1 INTRODUCTION
Let
be a finite-dimensional vector space over
and let
be a finite group. By the classical theorem of Chevalley–Serre–Shephard–Todd [
19, Théorèm 1'], the
linear quotient
of
by
is smooth if and only if
is generated by
reflections, that is, elements
with
. If
is singular, we also refer to this affine variety as a
quotient singularity. Regarding the class group of Weil divisors
of
, there is a nowadays well-known theorem by Benson [
3] which says that
, where
is the subgroup generated by the reflections contained in
. Restricting to subgroups
, we will show that the class group of a certain partial resolution of
can be described in a similar way.
The inclusion implies that cannot contain any reflections and is a singular variety. Further, by [5], there is a -factorial terminalization (or minimal model) , that is, a crepant, partial resolution of singularities of with at most terminal singularities; see below for the precise definition. -factorial terminalizations of quotient singularities are well-studied and, starting with the famous McKay correspondence [15], it is often observed that properties of are controlled only by the action of on , see, for example, the survey [17]. This leads to the expectation that questions regarding the birational geometry of should be answered by only considering , see Reid's “principle of the McKay correspondence” [17, Principle 1.1].
In this note, we give further evidence of this phenomenon by describing the class group of . By a version of the McKay correspondence due to Ito and Reid [11], the rank of the free part of this finitely generated abelian group coincides with the number of conjugacy classes of junior elements in ; we recall the definition of these distinguished elements of below. In the literature, one further finds a sufficient condition and a full characterization of the freeness of by Donten-Bury–Wiśniewski [7, Lemma 2.11] and Yamagishi [21, Proposition 4.14], respectively.
We study the torsion part of and obtain a theorem, which reads similar to Benson's result on .
Theorem 1.1. (= Theorem 5.1)Let be a finite group and let be the subgroup generated by the junior elements contained in . Let be a -factorial terminalization of . Then we have a canonical isomorphism of abelian groups
which is induced by the pushforward map
.
Combining our result with [11] gives a complete description of the class group.
Corollary 1.2. (= Corollary 5.2)With the assumptions in Theorem 1.1, we have
where
is the number of conjugacy classes of junior elements in
.
Remark 1.3.We emphasize that by Corollary 1.2, the class group of a -factorial terminalization is completely controlled by the group itself. This agrees well with the mentioned principle of Reid [17, Principle 1.1].
Remark 1.4.We make a further philosophical observation. Recall that by the theorem of Chevalley–Serre–Shephard–Todd, the variety is smooth if and only if is generated by reflections. This is mirrored by a theorem of Yamagishi [21, Theorem 1.1] that generalizes a result of Verbitsky [20, Theorem 1.1] and says that, if a -factorial terminalization is smooth, then must be generated by junior elements.
We feel that Theorem 1.1 mirrors Benson's theorem (Theorem 2.1) in the same way. In both cases, the geometry of the linear quotient is controlled by the reflections contained in and the junior elements control the geometry of the -factorial terminalization . Still, it appears that this picture is far from complete. The theorem of Verbitsky and Yamagishi on the smoothness of is not an equivalence and the freeness of the class group also depends in a somewhat convoluted way on the junior elements, see Remark 6.1.
Notice that the mentioned theorem of Verbitsky and Yamagishi together with our result implies that if is smooth, then the class group is free. This result was already shown in [7, Lemma 2.11] using different methods; a stronger statement appears in [21, Proposition 4.14], see also Corollary 5.4.
To prove Theorem 1.1, we study the Cox ring that is graded by . This directly extends the work of [21]. The Cox ring was introduced by Cox [6] in the context of toric varieties and transferred to a more general setting of birational geometry by Hu and Keel [10].
After fixing the notation and recalling fundamental results in Section 2, we present a correspondence of homogeneous elements in and effective divisors on in Section 3. We do not claim originality for the results in these sections, but only unify the different sources and transfer them to our setting. In Section 4, we analyse the grading of the Cox ring by in more detail and finally use this in Section 5 to prove our theorem. We close with some small examples in Section 6.
2 PRELIMINARIES
Let be a finite-dimensional vector space over and let be a finite group. Note that we will restrict to subgroups of after a short general discussion. Write for the linear quotient, where is the invariant ring of . The variety is normal and -factorial, see [3].
2.1 The class group of V/G
Recall that by a reflection (or pseudo-reflection), we mean an element with . The class group of was first described by Benson [3, Theorem 3.9.2] building on work of Nakajima [16, Theorem 2.11].
Theorem 2.1. (Benson)Let be a finite group and let be the subgroup generated by the reflections contained in . Then there is an isomorphism
of abelian groups.
With
as in the theorem, let
be the abelianization of
and write
for the group of irreducible (hence linear) characters of this group. By elementary character theory [
4, Theorem I.9.5], we have
and we hence often write
for the class group of
.
2.2 -factorial terminalizations
From now on, let be a finite subgroup.
Definition 2.2. (-factorial terminalization)Let be a normal -factorial variety. A -factorial terminalization of is a projective birational morphism such that is a normal -factorial variety with terminal singularities and is crepant.
In our context, a -factorial terminalization is often referred to as minimal model, see, for example, [11]. However, the usage of this terminology is not entirely uniform, which is why we decided to use the more unwieldy term “-factorial terminalization”.
We have the following special case of the deep result achieved in [5].
Theorem 2.3. (Birkar–Cascini–Hacon–McKernan)Let be a finite group. There exists a -factorial terminalization of .
This is [5, Corollary 1.4.3] together with the fact that has canonical singularities by the Reid–Tai criterion [13, Theorem 3.21], see [18, Theorem 2.1.15] for details.
2.3 McKay correspondence
Throughout, let be a -factorial terminalization. A deep connection between and the group itself is given by a version of the McKay correspondence due to Ito and Reid [11]. We recall some notation from [11].
For the following definition, let be of finite order and fix a primitive th root of unity . Then there are integers , such that the eigenvalues of are given by the powers , where .
Definition 2.4. (Age and junior elements)We call the number
the
age of
. Elements of age 1 are called
junior.
By construction, the number is an integer if and the junior elements in are hence the nontrivial elements of minimal age 1. The age is by definition invariant under conjugacy and we refer to the conjugacy classes of a group consisting (only) of junior elements as junior conjugacy classes.
Remark 2.5.The age as defined above depends on the choice of the root of unity , although this is hidden in the notation. See [11, p. 224, Remark 3] for an example demonstrating this. However, the subgroup generated by the junior elements, which is relevant for Theorem 1.1, is independent of any choices. We give a short argument for this in Appendix A, see Lemma A.4.
Definition 2.6. (Monomial valuation)For nonnegative integers with , we construct a discrete valuation defined on via
We call
a
monomial valuation.
This construction indeed gives a well-defined discrete valuation, see [12, Definition 2.1].
Notation 2.7.Let of finite order be a junior element with respect to the primitive th root of unity . That is, the eigenvalues of are with . For , we obtain as the age of with respect to the root of unity . As is an integer, we conclude and we can therefore define a monomial valuation
for
via
.
The construction of again depends on the choice of a root of unity, see Remark 2.5. The valuation is stable under conjugacy of and we can hence associate valuations to conjugacy classes in without needing to specify a particular representative.
Theorem 2.8. (Ito–Reid, “McKay correspondence”)Let be a finite group and let be a -factorial terminalization. Then there is a one-to-one correspondence between the junior conjugacy classes of and the irreducible exceptional divisors on .
More precisely, if is a divisor corresponding to a conjugacy class of a junior element of order in this way, then , where is the valuation of and we identify via the birational morphism .
See [11, section 2.8] for a proof.
Let
be the number of junior conjugacy classes in
. Using [
9, Proposition II.6.5(c)], we have an exact sequence of abelian groups
(1)where
are the irreducible exceptional divisors on
. As
is finitely generated, this implies that
is finitely generated as well.
Notation 2.9.We write for the torsion subgroup of and for the corresponding factor group, that is, .
The sequence in (1) can be extended to a short exact sequence by Grab [8].
Proposition 2.10. (Grab)There is a short exact sequence of abelian groups
where
are the irreducible components of the exceptional divisor of
and
is the induced pushforward map.
See [8, Proposition 4.1.3] for a proof.
Remark 2.11.As is a torsion group, we can deduce with Theorem 2.8 and Proposition 2.10 that for the free part of , where is the number of junior conjugacy classes in .
2.4 Cox rings
To understand the torsion part
of
we use the
Cox ring of
. The precise definition of this ring for a normal variety
is a bit involved and here we just remark that if the class group
is a free group, we have
see [
1, section 1.4] for the general case. The Cox ring is well-defined for normal varieties
with finitely generated class group
under the additional assumption
. These properties are in particular fulfilled for linear quotients
and we may hence speak about the Cox ring
.
We summarize a result on the Cox ring of
by Arzhantsev and Gaǐfullin [
2]. Recall that
as
cannot contain any reflections. There is an action of
on the ring
induced by the action of
. This action induces a grading by
by setting the graded component of a character
to be
where we write
for the action of
on
.
Theorem 2.12. (Arzhantsev–Gaǐfullin)Let be a finite group. Then there is an -graded isomorphism .
See [2, Theorem 3.1] for a proof.
3 CORRESPONDENCE OF EFFECTIVE DIVISORS AND HOMOGENEOUS ELEMENTS
To be able to deduce information about , we use the connection between effective divisors and canonical sections in the Cox ring of . We recall this correspondence and adapt it to our setting.
Notation 3.1.For a divisor , we write for the character corresponding to the class under the isomorphism in Theorem 2.1.
Remark 3.2.Working with the ring brings two subtle problems. First of all, homogeneous elements are only residue classes of elements of the function field as is a torsion group. We hence cannot immediately identify such elements with a function in . However, for a divisor we have an isomorphism
by [
1, Lemma 1.4.3.4]. That means, once we fixed a representative of the degree of a homogeneous element
we can uniquely lift
to an element of
.
The second problem comes from the fact that we make heavy use of the graded isomorphism as in Theorem 2.12 to the extent that one might forget that the isomorphism is not an identity. This is in particular important when we work with a valuation . We can only use on elements of and cannot apply to elements of in a well-defined way without choosing a system of representatives for the class group. For , we have an isomorphism of vector spaces
by setting
. Notice that for the trivial divisor, this gives an identity as we have
where 1 denotes the trivial character.
Notation 3.3.Let and let with . For a homogeneous element , let be the rational function mapping to via the isomorphism determined by as in Remark 3.2. We associate to an effective divisor
the
-
divisor of
. This construction is well-defined, see [
1, Proposition 1.5.2.2]. In particular, the
-divisor is independent of the choice of the representative
. We have
by definition.
The construction of a -divisor is not limited to our setting; see [1, Construction 1.5.2.1] for more details and the general case. We point out that is in general not an element of , that is, there is no meaning in writing .
The -divisor behaves well with respect to the multiplication of elements.
Lemma 3.4.For nonzero homogeneous elements and , we have
See [1, Proposition 1.5.2.2(iii)] for a proof.
We have a converse to the construction of the -divisor.
Proposition 3.5.Let be an effective divisor. There exist a class and an element with . The element is unique up to constants; it is called a canonical section of .
See [1, Proposition 1.5.2.2(i)] and [1, Proposition 1.5.3.5(ii)] for a proof.
Using the correspondence between effective divisors and homogeneous elements one can derive a precise description of the image of the strict transform of an effective divisor in the free group . The general idea of this argument appeared to our knowledge first in [7, Lemma 3.22]. We require a bit of notation.
Recall that by Theorem 2.8, we have a one-to-one correspondence between the junior conjugacy classes of and the irreducible components of the exceptional divisor of . Let be a minimal set of representatives of the junior conjugacy classes corresponding to exceptional prime divisors . For each , write for the monomial valuation on defined by and recall from Theorem 2.8 that we have , where is the divisorial valuation of and the order of .
The following also appears in [8, Proposition 4.1.9]. We present the argument from [21, Lemma 4.3] for completeness. Denote the canonical projection by .
Proposition 3.6.Let be an effective divisor on and let be a canonical section. Write for the strict transform of via . Then we have the equality
in
.
Proof.As is homogeneous with respect to the action of , there is such that and is principal. In particular, we have
where the first equality is by Lemma
3.4, the second by the independence of choice of representative and the third is by the fact that
, see Remark
3.2. Then we have
Hence, we have the equality of classes
in
. Now
by Theorem
2.8. Noting that
is a valuation on
(and not just
) this yields
We may finally cancel
in the free group
giving
Remark 3.7.Notice that the proof of Proposition 3.6 in fact computes the degree of a preimage of the canonical section under a graded surjective morphism induced by . See [1, Proposition 4.1.3.1] for the construction of this morphism between the Cox rings.
4 A DIGRESSION ON GRADINGS
To understand the group , we first have to get a better understanding of the grading of by . Unfortunately, there are a few subtle details involved, turning this into quite a technical discussion.
Again, let be representatives of the junior conjugacy classes corresponding to the exceptional divisors of and write for the monomial valuations corresponding to the .
At first, fix . Let the eigenvalues of be given by with a primitive th root of unity and integers , where is the order of in and . This induces a -grading on by putting . For a polynomial , the valuation is then the degree of the homogeneous component of of minimal degree with respect to . Notice that in this construction, we consider in an eigenbasis of giving rise to the isomorphism . However, the grading is well-defined on for any basis of , although the variables of the polynomial ring are in general not homogeneous. As we endow the same ring with gradings by different groups, we use the nonstandard notation for the ring graded by via .
The group acts on and hence induces a grading by , which we denote by . Write for the ring graded by via . We directly obtain:
Lemma 4.1.With the above notation, if is -homogeneous, then is -homogeneous as well and we have
In particular, there is a graded morphism
given by the identity on the rings and by the projection
on the grading groups.
Write
for the linear action of
on
. Observe that for every
we have an action of
on
. Indeed, for any
and
, we have
so
as required. Hence, the grading by
descends to
. As the actions of the elements
on
commute, we can consider all the induced gradings at the same time and hence obtain a grading by the group
on
.
The do not commute with each other in general, so we cannot decompose their actions on into a common eigenbasis. Hence, we cannot put the above gradings together to obtain a grading by or on as there are in general no polynomials that are homogeneous with respect to all gradings at the same time.
Let
be the subgroup of
generated by the junior elements contained in
. In general, the representatives
do not suffice to generate
. Let
and notice that this group is generated by the residue classes
modulo
. This gives a map
which is surjective onto
. This surjection corresponds to an embedding of character groups
. Further, the inclusion
induces a projection of characters
by restriction. We conclude:
Lemma 4.2.The gradings on coming from the actions of the groups , and are compatible in the sense that there is a graded morphism
which factors through
.
We state for later reference:
Lemma 4.3.We have and .
Proof.For the first statement, we note that the image of under the projection is . Hence,
and an application of the isomorphism theorems gives the claim. The second statement follows directly as
is a contravariant functor.
The following three lemmas are key ingredients for our theorem on .
Lemma 4.4.Let be -homogeneous. For every index , we have .
Proof.Let be -homogeneous. Fix an . Then is -homogeneous by Lemma 4.2. By Lemma 4.1, there exist - and -homogeneous elements such that and whenever . In particular, we have and . Hence, we conclude
by Lemma
4.1.
Lemma 4.5.Let be -homogeneous. We have for all if and only if , where is the subgroup generated by the junior elements contained in .
Proof.By Lemma 4.4, we have mod for every . Therefore, is equivalent to for every . Equivalently, every acts trivially on . As is furthermore -invariant, we conclude that this is the case if and only if every junior element in leaves invariant and hence .
Lemma 4.6.Let be a class of divisors. There exists a homogeneous element in of degree .
Proof.This is saying that the relative invariants with respect to the linear characters of on are nonempty which holds by [16, Lemma 2.1].
Notice that the lemma also implies that we can find an effective divisor in any class of divisors in .
5 THE CLASS GROUP
We are now prepared for our theorem.
Theorem 5.1.Let be a finite group and let be the subgroup generated by the junior elements contained in . Let be a -factorial terminalization of . Then we have a canonical isomorphism of abelian groups
which is induced by the pushforward map
.
Proof.For ease of notation, we identify with via Theorem 2.1 and use both groups synonymously. Notice that is the subgroup of consisting of those characters that take value 1 on every junior element. We claim that restricting to induces a bijection onto .
We first show that we indeed have . Let be a divisor on . By Lemma 4.6, there is of degree and we have the effective divisor on with . Write for the strict transform of via . Then , hence by Proposition 2.10 we have
(2) with
and where
are the irreducible components of the exceptional divisor of
. As before let
be the canonical projection. Applying
on both sides of (
2) and using Proposition
3.6 yields
(3) Assume now . Then and we conclude by (3) that for all and, in particular, . Hence, by Lemma 4.5 and therefore we can identify , or more precisely , with an element of . This means that we obtain a well-defined map
by restricting
to
.
We now prove that is bijective. For injectivity, notice that the morphism of groups
is injective. This follows from the exactness of the sequence in Proposition
2.10 noticing that the group
embeds into
, see also [
8, Lemma 4.1.4]. The injectivity of
implies the injectivity of
: if we have
for
, then
as by construction
.
Now let be a character, which we identify with a class of divisors . By Lemma 4.6, there exists and we may assume without loss of generality that is effective and is the canonical section of as in Proposition 3.5. By the assumption on , we have for all by Lemma 4.5. Let
and set
, where
is the strict transform of
via
. By Proposition
2.10, we have
and therefore
. Using Proposition
3.6, we have
, hence
and
. We conclude
and
is surjective.
Combining Theorem 2.8, see Remark 2.11, and Theorem 5.1 enables us to describe the class group of in general.
Corollary 5.2.Let be a finite group and let be the subgroup generated by the junior elements contained in . Let be a -factorial terminalization of . Then we have
where
is the number of junior conjugacy classes in
. Further, the canonical embedding
satisfies
with
as above.
Proof.The first part follows directly from the mentioned theorems. For the second part, we combine Proposition 2.10 and the first part to obtain and then the claim follows by Lemma 4.3.
Remark 5.3.As the isomorphism in Theorem 5.1 is induced by , we can see the sequence in Proposition 2.10 as the direct sum of the short exact sequences
and
We obtain [21, Proposition 4.14] as a further corollary.
Corollary 5.4. (Yamagishi)Let be a finite group and let be a -factorial terminalization of . Then the class group is free if and only if is generated by the junior elements contained in together with .
6 EXAMPLES AND CLOSING REMARKS
Remark 6.1.Note that in Corollary 5.4 we cannot drop the part “together with ” for the equivalence, that is, there are groups that are not generated by junior elements such that is free. For example, let be the binary icosahedral group [14, Theorem 5.14] and set . The abelianization is trivial, so the same is true for . However, every nontrivial element in is of age 1, hence all nontrivial elements of are of age 2 and does not contain any junior elements. Hence, the class group of a -factorial terminalization of is trivial and therefore free. For an example of a nontrivially free class group, one considers the direct product of with a group generated by junior elements.
Example 6.2.As a “reality check,” let be a group that does not contain any junior elements. Then for every nontrivial , so has terminal singularities by [13, Theorem 3.21]. Hence, is a -factorial terminalization of itself and Corollary 5.2 gives as in Theorem 2.1.
Example 6.3.For a nontrivial example, we consider the group
of order 6, where
is a primitive third root of unity. As
does not contain any reflections, we have
.
To determine the age of elements in , we need to fix a primitive sixth root of unity. However, the two possible choices and both result in the same junior elements of , namely
By the Reid–Tai criterion [
13, Theorem 3.21], the existence of junior elements in
implies that
is not terminal. As
is abelian, the conjugacy classes in
are trivial. So, the rank of the free part of the class group
of a
-factorial terminalization
is 2. For the torsion part, we determine that
is cyclic of order 2 and we conclude
We write the elements of as 3-tuples with the first two entries corresponding to the free part and the last entry corresponding to the torsion part. Then the pushforward morphism is induced by
where
denotes the identity matrix.
APPENDIX A: AGE REVISITED
Recall that for the integer depends on a choice of root of unity. In this appendix, we study this issue in more detail to show that the results in this paper are in fact independent of any choices.
Remark A.1.In [11], Ito and Reid avoid making any choices by defining the age not for the group , but for the set , where is the group of roots of unity of order and is a common multiple of the orders of the elements of . On , the notion of age is independent of any choices. Any primitive root of unity of order gives a bijection and one may endow with a group operation via this map. However, for the arguments in this paper we need a notion of age on ; this is quite common, see for example [17].
First of all, to be able to speak about junior elements in a uniform way, we introduce the following definition. Let be the exponent of and let be a primitive th root of unity. For any , we have , so there are such that the eigenvalues of are given by , , with . If is the order of , we must have and set .
Definition A.2.With the above notation, we set . We call the integers the weights of with respect to . We call a -junior element, if .
The integer coincides with as constructed above for an appropriate choice of .
Lemma A.3.Let be primitive th roots of unity. Then there is a bijection (of sets) such that for all . Further, the weights of with respect to and with respect to coincide.
Proof.By assumption, there is with and . Then there is with mod . Hence, we have a map
with inverse
.
Let be an element of order . There are such that the eigenvalues of are given by . Then the eigenvalues of are given by and
as claimed.
Lemma A.4.Let be primitive th roots of unity and write , respectively, , for the subgroup of generated by the -junior elements, respectively, the -junior elements. Then we have .
Proof.Let be the bijection in Lemma A.3, so taking powers by some . If is a -junior element, then is an -junior element. Further, we clearly have , hence and an analogous argument gives the reverse inclusion.