TERMINAL QUOTIENT SINGULARITIES IN

DIMENSION THREE VIA VARIATION OF GIT

arXiv:1502.03579v1 [math.AG] 12 Feb 2015

SEUNG-JO JUNG
Abstract. A 3-fold terminal quotient singularity X = C3 /G admits the economic resolution Y X, which is close to being
crepant. This paper proves that the economic resolution Y is isomorphic to a distinguished component of a moduli space of certain
G-equivariant objects using the King stability condition introduced by Kedzierski
[11].

Contents
1. Introduction
2. G-bricks and moduli spaces of G-constellations
2.1. Moduli spaces of G-constellations
2.2. G-prebricks and local charts of M
2.3. G-bricks and the biratioanl component Y
3. Weighted blowups and economic resolutions
3.1. Terminal quotient singularities in dimension 3
3.2. Weighted blowups and round down functions
3.3. Economic resolutions
4. Moduli interpretations of economic resolutions
4.1. G-bricks and stability parameters for 1r (1, r 1, 1)
4.2. G-bricks for 1r (1, a, r a)
4.3. Stability parameters for S(r, a)
4.4. Main Theorem
5. Kedzierskis
GIT chamber

1
6. Example: type 12
(1, 7, 5)
References

1
4
4
6
11
12
12
14
17
18
18
19
25
27
28
32
37

1. Introduction
The motivation of this work stems from the philosophy of the McKay
correspondence, which says that if a finite group G acts on a variety M,
then a crepant resolution of the quotient M/G can be realised as a
moduli space of G-equivariant objects on M.
2010 Mathematics Subject Classification. 14E15, 14E16, 14L24.
Key words and phrases. the McKay correspondence, terminal quotient singularities, economic resolutions.
1

S.-J. JUNG

Let G GLn (C) be a finite group. A G-equivariant coherent sheaf F

on Cn is called a G-constellation if H0 (F ) is isomorphic to the regular
representation C[G] of G as a C[G]-module. In particular, the structure
sheaf of a G-invariant subscheme Z Cn with H0 (OZ ) isomorphic to
C[G], which is called a G-cluster, is a G-constellation. Define the GIT
stability parameter space



= HomZ (R(G), Q)  (C[G]) = 0 ,

where R(G) is the representation ring of G. For , we say that

a G-constellation F is -(semi)stable if (G) > 0 ((G) 0) for every
nonzero proper subsheaf G of F . A parameter is called generic if
every -semistable G-constellation is -stable.
Let M be the moduli space of -semistable G-constellations. In
the celebrated paper [1], Bridgeland, King and Reid proved that for a
finite subgroup G of SL3 (C), M is a crepant resolution of C3 /G if is
generic. Craw and Ishii [2] showed that in the case of a finite abelian
group G SL3 (C), any projective crepant resolution can be realised
as M for a suitable GIT parameter .
While the moduli space M need not be irreducible [4] in general,
Craw, Maclagan and Thomas [3] showed that for generic , M has
a unique irreducible component Y containing the torus (C )n /G if G
is abelian. The component Y is birational to Cn /G and is called the
birational component1 of M .
On the other hand, in the case of G GL3 (C) giving a terminal
quotient singularity X = C3 /G in dimension 3, X has the economic
resolution : Y X satisfying
X i
KY = (KX ) +
Ei
r
1i<r
with Ei s prime exceptional divisors. Kedzierski
[11] proved that Y is

isomorphic to the normalization of Y for some . The main theorem

of this paper is that the economic resolution Y of X can be interpreted
as a component of a moduli space of G-constellations as follows.
Theorem 1.1 (Theorem 4.19). The economic resolution Y of a 3-fold
terminal quotient singularity X = C3 /G is isomorphic to the birational
component Y of the moduli space M of -stable G-constellations for
a suitable parameter .
To prove the theorem, first we generalize Nakamuras result [16]. Let
G GLn (C) be a finite diagonal group. Nakamura [16] introduced a
G-graph which is a C-basis of OZ for a torus invariant G-cluster Z.
Using G-graphs, he described a local chart of G-Hilb. In this paper, we
introduce a G-prebrick which is a C-basis of H0 (F )
= C[G] for a torus
invariant G-constellation F .
1This

component is also called the coherent component.

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

For a G-prebrick , by King [12], we have an affine scheme D()

parametrising G-constellations whose basis is . The affine scheme
D() is not necessarily irreducible, but D() has a distinguished component U() containing the torus T = (C )n /G. In addition, we can
show that U() = Spec C[S()] for a semigroup S(). If the toric affine
variety U() has a torus fixed point, then is called a G-brick. We
can prove that Y is covered by U()s for suitable G-bricks .
On the other hand, from [14, 17], we know that a 3-fold quotient
singularity X = C3 /G has terminal singularities if and only if the
group G is of type 1r (1, a, r a) with r coprime to a, i.e.

G = hdiag(, a , ra )  r = 1i.

In this case, the quotient variety X = C3 /G is not Gorenstein. While

X does not admit a crepant resolution, X has the economic resolution
: Y X obtained by a toric method called weighted blowups (or
Kawamata blowups). For each step of the weighted blowups, we define
three round down functions, which are maps between monomial lattices.
As Y is toric, Y is determined by its associated toric fan with the
lattice M of G-invariant monomials. From toric geometry, note that Y
is covered by torus invariant affine open subsets U = Spec C[ M]
for max where max denotes the set of maximal cones in .
Using the round down functions, we find a set S of G-bricks such
that there exists a bijective map max S sending to with
U( )
= U . We show that there exists a parameter such that
U( )s cover Y for S. This proves that the economic resolution
Y is isomorphic to the birational component Y of M .
Moreover, we further prove D()
= C3 for S. So the irreducible
component Y is actually a connected component. We conjecture that
the moduli space M is irreducible, which implies Y
= M .
Layout of this article. In Section 2, we define G-(pre)bricks and
describe the birational component Y using G-bricks. Section 3 explains
how to obtain the economic resolutions using toric methods and defines
round down functions. In Section 4, we explain how to find G-bricks
and a parameter such that the economic resolution is isomorphic
to the birational component Y . In Section 5 we describe Kedzierskis
GIT chamber using the Ar1 root system. In Section 6, we calculate
1
G-bricks and Kedzierskis GIT chamber for the group of type 12
(1, 7, 5).

Acknowledgement. This article is part of my PhD thesis in the University of Warwick [8]. I am very grateful to my advisor Miles Reid
for sharing his views on this subject and many ideas. I would like to
thank Diane Maclagan, Alastair Craw, Hiraku Nakajima, and Timothy
Logvinenko for valuable conversations. I thank Andrew Chan and Tom
Ducat for their comments on earlier drafts.

S.-J. JUNG

2. G-bricks and moduli spaces of G-constellations

In this section we define a G-prebrick which is a generalized version
of Nakamuras G-graph from [16]. By using G-prebricks, we describe
local charts of moduli spaces of G-constellations.
In this section, we restrict ourselves to the case where G is a finite
cyclic subgroup of GL3 (C). It is possible to generalize part of the
argument to include general finite small abelian groups in GLn (C) for
any dimension n. However we prefer to focus on this case where we
can avoid the difficulty of notation.
2.1. Moduli spaces of G-constellations. In this section, we review
the construction of moduli spaces M of -stable G-constellations as
described in [2, 12].

Define the group G = hdiag(1 , 2 , 3 )  r = 1i GL3 (C). We call
G the group of type 1r (1 , 2 , 3 ). We can identify the set of irreducible
representations of G with the character group G := Hom(G,LC ) of G.
Note that the regular representation C[G] is isomorphic to G C.

Definition 2.1. A G-constellation on C3 is a G-equivariant coherent sheaf F on C3 with H0 (F ) isomorphic to the regular representation C[G] of G as a C[G]-module.
L
The representation ring R(G) of G is G Z . Define the GIT
stability parameter space



= HomZ (R(G), Q)  (C[G]) = 0 .

Definition 2.2. For a stability parameter , we say that:

(i) a G-constellation F is -semistable if (G) 0 for every proper

submodule G F .
(ii) a G-constellation F is -stable if (G) > 0 for every nonzero
proper submodule G F .
(iii) is generic if every -semistable object is -stable.
It is known [4] that the language of G-constellations is the same as the
language of the McKay quiver representations. Thus by King [12], the
moduli spaces of G-constellations can be constructed using Geometric
Invariant Theory (GIT).
Let Rep G be the affine scheme whose coordinate ring is


C[Rep G] = C[xi , yi , zi  i G ] IG
where IG is the ideal generated by the following quadrics:

xi yi+1 yi xi+2 ,
(2.3)
xi zi+1 zi xi+3 ,

y z
i i+2 zi yi+3 .

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

Q
Let = (1, . . . , 1) Zr0 . The group GL() := iG C = (C )r acts
on Rep G via change of basis. For a parameter , define the GIT
quotient with respect to
Rep G // GL() := Repss
G/ GL()
ss
parametrising closed GL()-orbits in Repss
G where Rep G denotes
the -semistable locus in Rep G.

Theorem 2.4 (King [12]). Let us define M := Rep G // GL().

(i) The quasiprojective scheme M is a coarse moduli space of
-semistable G-constellations up to S-equivalence.
(ii) If is generic, the scheme M is a fine moduli space of -stable
G-constellations.
(iii) The scheme M is projective over M0 = Spec C[Rep G]GL() .
Birational component Y of the moduli space M . Let M denote the moduli space of -semistable G-constellations. Note that the
moduli space M need not be irreducible [4].
Note that for every parameter , there exists a natural embedding
of the torus T := (C )3 /G into M . Indeed, for a G-orbit Z in the
algebraic torus T := (C )3 C3 , since Z is a free G-orbit, OZ has
no nonzero proper submodules. Thus OZ is a -stable G-constellation.
Hence it follows that the torus T := (C )3 /G is the fine moduli space of
-stable G-constellations supported on T because any G-constellation
supporting on a free G-orbit Z is isomorphic to OZ .
Theorem 2.5 (Craw, Maclagan and Thomas [3]). Let be generic.
Then M has a unique irreducible component Y that contains the torus
T := (C )n /G. Moreover Y satisfies the following properties:
Y
M

y
y
C3 /G M0
(i) Y is a not-necessarily-normal toric variety which is birational
to the quotient variety C3 /G.
(ii) Y is projective over the quotient variety C3 /G.
Definition 2.6. The unique irreducible component Y in Theorem 2.5
is called the birational component of M .
Since Craw, Maclagan and Thomas [3] constructed Y as GIT quotient of a reduced affine scheme, it follows that Y is reduced.
Remark 2.7. Since T = (C )3 acts on C3 , the algebraic torus T acts
on the moduli space M naturally. Fixed points of the T-action play
a crucial role in the study of the moduli space M .

S.-J. JUNG

2.2. G-prebricks and local charts of M . Let G GL3 (C) be the

finite group of type 1r (1 , 2 , 3 ). Define the lattice
1
L = Z3 + Z (1 , 2 , 3 ),
r
3
which is an overlattice of L = Z of finite index. Let {e1 , e2 , e3 } be the
standard basis of Z3 . Set M = HomZ (L, Z) and M = HomZ (L, Z). The
two dual lattices M and M can be identified with Laurent monomials
and G-invariant Laurent monomials, respectively. The embedding of G
into the torus (C )3 GL3 (C) induces a surjective homomorphism
wt : M G
whose kernel is M. Note that there are two isomorphisms of abelian
groups L/Z3 G and M /M G .
Let M 0 denote genuine monomials in M , i.e.

M 0 = {xm1 y m2 z m3 M  m1 , m2 , m3 0}.

For a set A C[x , y , z ], let hAi denote the C[x, y, z]-submodule of

C[x , y , z ] generated by A.
Let + be the cone in LR := L Z R generated by e1 , e2 , e3 . Note

that the corresponding affine toric variety U+ = Spec C[+

M] is
3
G
isomorphic to the quotient variety C /G = Spec C[x, y, z] .
Definition 2.8. A G-prebrick is a subset of Laurent monomials in
C[x , y , z ] satisfying:
(i) the monomial 1 is in .
(ii) for each weight G , there exists a unique Laurent monomial m of weight , i.e. wt : G is bijective.
(iii) if n n m for m and n, n M 0 , then n m .
(iv) the set is connected in the sense that for any element m ,
there is a (fractional) path in from m to 1 whose steps
consist of multiplying or dividing by one of x, y, z.
For a Laurent monomial m M , let wt (m) denote the unique
element m in of the same weight as m.
Remark 2.9. Nakamuras G-graph in [16] is a G-prebrick because
if a monomial n n is in for two monomials n, n M 0 , then n is
in . The main difference between G-graphs and G-prebricks is that
elements of G-prebricks are allowed to be Laurent monomials, not just
genuine monomials.

Example 2.10. Let G be the group of type 17 (1, 3, 4). Then

n
o
2
z z2 z2
1 = 1, y, y , z, y , y , y2 ,
o
n
3
3
2
2 = 1, z, y, y 2, yz , yz , yz 2

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

are G-prebricks. For 1 , we have wt1 (x) =

z
y

and wt1 (y 3 ) =

z2
.
y2

For a G-prebrick = {m }, as an analogue of [16], define S() to

n m
be the subsemigroup of M generated by
for all n M 0 ,
wt (n m )
m . Define a cone () in LR = R3 as follows:
() = S()



= u LR  u,

n m
wt (n m )

0,


m , n M 0 .

Observe that:


(i) M 0 M S(),
(ii) () + ,


(iii) S() () M .

Lemma 2.11. Let be a G-prebrick. Define




B() := f m  m , f {x, y, z} \.

Then the semigroup S() is generated by wtb(b) for all b B() as a

semigroup. In particular, S() is finitely generated as a semigroup.
Proof. Let S be the subsemigroup of M generated by wtb(b) for all
b B(). Clearly, S S(). For the opposite inclusion, it is enough
to show that the generators of S() are in S.

for some
An arbitrary generator of S() is of the form wtnm
(nm )
n M 0 , m . We may assume that n m 6 . In particular,
n 6= 1. Since n has positive degree, there exists f {x, y, z} such
that f divides n, i.e. nf M 0 and deg( nf ) < deg(n). Let m denote
wt ( nf m ). Note that
n
wt (f m ) = wt (f m ) = wt (n m ).
f
Thus
n
m
f wt ( nf m )
n m
f
=

wt (n m )
wt ( nf m )
wt (n m )
n
m
f m
f
=

.
n
wt ( f m ) wt (f m )
By induction on the degree of monomial n, the assertion is proved. 
The set B() in the lemma above is called the Border bases of . As
B() is finite, the semigroup S() is finitely generated as a semigroup.
Thus the semigroup S() defines an affine toric variety. Define two
affine toric varieties:
U() := Spec C[S()],
U () := Spec C[() M].

S.-J. JUNG

Note that the torus Spec C[M] of U() is isomorphic to T = (C )3 /G

and that U () is the normalization of U().
Craw, Maclagan and Thomas [4] showed that there exists a torus
invariant G-cluster which does not lie in the birational component Y .
The following definition is implicit in [4].
Definition 2.12. A G-prebrick is called a G-brick if the affine toric
variety U() contains a torus fixed point.
From toric geometry, U() has a torus fixed point if and only if
S() (S())1 = {1}, i.e. the cone () is a 3-dimensional cone.
Example 2.13. Consider the G-prebricks 1 , 2 in Example 2.10. By
5
3
Lemma 2.11, S(1 ) is generated by yz 2 , yz 4 , xy
. We have
z

n
 y5 z 3 xy o
3
,
(1 ) = u R hu, mi 0, for all m z 2 , y4 , z


= Cone (1, 0, 0), 71 (3, 2, 5), 17 (1, 3, 4) .

Similarly, we can see that


n
 y4 z 4 xz 2 o
3
(2 ) = u R hu, mi 0, for all m z 3 , y3 , y3
,


= Cone (1, 0, 0), 71 (1, 3, 4), 17 (6, 4, 3) .

Since S(1 ) = (1 ) M and S(2 ) = (2 ) M, the two G-prebricks

1 , 2 are G-bricks. Moreover the two toric varieties U(1 ) and U(2 )
are smooth.

Let be a G-prebrick. Define

C() := hi/hB()i.
The module C() is a torus invariant G-constellation. A submodule G
of C() is determined by a subset A , which forms a C-basis of G.

Lemma 2.14. Let A be a subset of . The following are equivalent.

(i) The set A forms a C-basis of a submodule of C().
(ii) If m A and f {x, y, z}, then f m implies f m A.
Let p be a point in U(). Then the evaluation map
evp : S() (C, ),
is a semigroup homomorphism.
To assign a G-constellation C()p to the point p of U(), first consider the C-vector space with basis whose G-action is induced by the
G-action on C[x, y, z]. Endow it with the following C[x, y, z]-action:


n m
n m := evp
(2.15)
wt (n m ),
wt (n m )
for a monomial n M 0 and m .

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

Lemma 2.16. Let be a G-prebrick.

(i) For every p U(), C()p is a G-constellation.
(ii) For every p U(), is a C-basis of C()p .
(iii) If p and q are different points in U(), then C()p 6
= C()q .
3
(iv) Let Z T = (C ) be a free G-orbit and p the corresponding
point in the torus Spec C[M] of U(). Then C()p
= OZ as
G-constellations.
(v) If U() has a torus fixed point p, then C()p
= C().
Proof. From the definition of C()p , the assertions (i), (ii) and (v) follow immediately. The assertion (iii) follows from the fact that points
on the affine toric variety U() are in 1-to-1 correspondence with semigroup homomorphisms from S() to C.
It remains to show (iv). Let Z T = (C )3 be a free G-orbit and
p the corresponding point in Spec C[M] U(). There is a surjective
G-equivariant C[x, y, z]-module homomorphism
C[x, y, z] C()p

given by f 7 f 1.

whose kernel is equal to the ideal of Z. This proves (iv).

Definition 2.17. A G-prebrick is said to be -stable if C() is -stable.

Deformation space D(). We introduce deformation theory of C()
for a -stable G-prebrick . We deform C(), keeping the same vector space structure, but perturbing the structure of C[x, y, z]-module.
Since we fix a C-basis of C(), deforming C() involves 3r parameters {x , y , z  G } with

x m = x wt (x m ),
y m = y wt (y m ),

z m = z wt (z m ),

with the following commutation relations:

x ywt(xm ) y xwt(ym ) ,
(2.18)
x zwt(xm ) z xwt(zm ) ,

y z
wt(ym ) z ywt(ym ) .

Note that wt (m) is the base of the same weight as m. Fixing a

basis means that we set f = 1 if wt (f m ) = f m for f {x, y, z}.
Define a subset of the 3r parameters
 

() := f  wt (f m ) = f m , f {x , y , z } ,

i.e. () is the set of parameters fixed to be 1. Define the affine scheme




(2.19)
D() := Spec C[x , y, z  G ] I



where I = the quadrics in (2.18), f 1  f () . By King [12],
the affine scheme D() is an open set of M containing the point

10

S.-J. JUNG

f
corresponding to C(). More precisely, consider an affine open set U
in Rep G, which is defined by f to be nonzero for all f (). Note
f is GL()-invariant and that U
f is in the -stable locus. Since
that U
ss
the quotient map Rep G M is a geometric quotient for generic ,
f ]GL() is an affine open subset of
from GIT [15], it follows that Spec C[U
M . On the other hand, setting f () to be 1 for all f () gives
f ]GL() .
a slice of the GL()-action. Thus D() is isomorphic to Spec C[U
Remark 2.20. The affine open subset D() of the moduli space M
parametrises G-constellations whose basis is .

Proposition 2.21. For generic , let be a -stable G-brick and Y

the birational component of M . Then C()p is -stable for every
p U(). Furthermore, there exists an open immersion
U() = Spec C[S()] Y M ,
which fits into the following commutative diagram:
U()

D() M
where the vertical morphisms are closed embeddings.
Proof. Let us assume that the G-constellation C() is -stable. Let p
be an arbitrary point in U() and G a submodule of C()p . By the
definition of C()p , there is a submodule G of C() whose support is
the same as G. Since C() is -stable, (G) = (G ) > 0. Thus C()p
is -stable.
Since there is a C-algebra epimorphism from C[D()] to C[S()]
given by
f m
f 7
wt (f m )
for f {x , y, z }, it follows that U() is a closed subscheme of D().
As Craw, Maclagan, and Thomas [3] proved that the birational component Y is a unique irreducible component of M containing the
torus T = (C )3 /G, it follows that Y D() is a unique irreducible
component of D() containing the torus T .
The morphism U() D() M induces an isomorphism between the torus Spec C[M] and the torus T of Y by Lemma 2.16 (iv).
Note that U() is in the component Y D() because U() is a closed
subset of D() containing T . Since both U() and Y D() are reduced and of the same dimension, U() is equal to Y D(). Thus
there exists an open immersion from U() to Y .


TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

11

2.3. G-bricks and the biratioanl component Y . In this section,

we present a 1-to-1 correspondence between the set of torus fixed points
in Y and the set of -stable G-bricks.
Proposition 2.22. Let G GL3 (C) be the group of type 1r (1 , 2 , 3 ).
For a generic parameter , there is a 1-to-1 correspondence between the
set of torus fixed points in the birational component Y of the moduli
space M and the set of -stable G-bricks.
Proof. 2 In Section 2.2, we have seen that if is a -stable G-brick,
then C() is a torus invariant G-constellation corresponding to a torus
fixed point in Y .
Let p Y be a torus fixed point and F the corresponding torus
invariant G-constellation. For a one parameter subgroup
: C T Y
with limt0 (t) = p, induces a flat family V of G-constellations over
A1C . Note that V has generic support; for every nonzero t A1C , the
G-constellation Vt over t is isomorphic to OZ for a free G-orbit Z in
T = (C )3 . There is a set = {m M  G } satisfying:

(1) is a C-basis of Vt for every t A1C .

(2) 1 .
(3) Vt is isomorphic to hi/Nt for a submodule Nt of hi, where
hi denotes the C[x, y, z]-module generated by .
We prove that is a G-prebrick and that F
= C(). Note that F can
be written as hi/N for a submodule N. For any m , since m is a
base, m is not in N. Moreover if nm 6 for n M 0 , m , then
nm N because the dimension of H0 (F ) is r = ||. This proves that
N = hB()i, where B() is the Border bases in Lemma 2.11. From this,
it follows that satisfies the conditions (i),(ii),(iii) in Definition 2.8.
As F is -stable for generic , the connectedness condition (iv) follows.
To see that the G-prebrick is a G-brick, note that the point p Y
corresponds to the isomorphism class of C() so p D(). Thus p is
in U() = Y D().

Corollary 2.23. Let be a G-prebrick. Then C() lies in the birational component Y if and only if is a G-brick.

Theorem 2.24. Let G GL3 (C) be a finite diagonal group and a

generic GIT parameter for G-constellations. Assume that S is the set
of all -stable G-bricks.
(i) The birational component Y of M
S is isomorphic to the notnecessarily-normal toric variety S U().
2In

[8], there is another proof using the language of the McKay quiver
representations.

12

S.-J. JUNG

(ii) The normalization of Y is isomorphic to the normal toric variety whose toric fan consists of the 3-dimensional cones ()
for S and their faces.
Proof. Let G be the group of type 1r (1 , 2 , 3 ). Consider the lattice
L = Z3 + Z 1r (1 , 2 , 3 ).
Let Y be the birational component of the moduli space of -stable
G-constellations and Y the normalization
of Y . Let Y denote the
S
not-necessarily-normal toric variety S U(). Define the fan in LR
whose maximal cones are () for S. Note that the corresponding
toric variety Y := X is the normalization of Y .
By Proposition 2.21, there is an open immersion : Y Y . From
Proposition 2.22, it follows that the image (Y ) contains all torus
fixed points of Y . The induced morphism : Y Y is an open
embedding of normal toric varieties with the same number of torus
fixed points. Thus the morphism should be an isomorphism. This
proves (ii).
To show (i), suppose that Y \(Y ) is nonempty so it contains a torus
orbit O of dimension d 1. Since the normalization morphism is torus
equivariant and surjective, there exists a torus orbit O in Y
= Y
of dimension d which is mapped to the torus orbit O. At the same
time, from the fact that Y is the normalization of Y and that the
normalization morphism is finite, it follows that the image of O is a
torus orbit of dimension d, so the image is O. Thus O is in (Y ), which
is a contradiction.

Corollary 2.25. With the notation as in Theorem 2.24, Y is a normal
toric variety if and only if S() = () M for all S.
3. Weighted blowups and economic resolutions
Let G GL3 (C) be the finite subgroup of type 1r (1, a, r a) with r
coprime to a, i.e.

G = hdiag(, a , ra )  r = 1i.

The quotient X = C3 /G has terminal singularities and has no crepant

resolution. However there exist a special kind of toric resolutions, which
can be obtained by a sequence of weighted blowups. In this section,
we review the notion of toric weighted blowups and define round down
functions which are used for finding admissible G-bricks.

3.1. Terminal quotient singularities in dimension 3. In this section, we collect various facts from birational geometry. Most of these
are taken from [17].
Definition 3.1. Let X be a normal quasiprojective variety and KX
the canonical divisor on X. We say that X has terminal singularities
(resp. canonical singularities) if it satisfies the following conditions:

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

13

(i) there is a positive integer r such that rKX is a Cartier divisor.

(ii) if : Y X is a resolution with Ei prime exceptional divisors
such that
X
rKY = (rKX ) + r
ai Ei ,
then ai > 0 (resp. 0) for all i.

In the definition above, ai is called the discrepancy of Ei . A crepant

resolution of X is a resolution with all discrepancies zero.
Birational geometry of toric varieties. Let L be a lattice of rank
n and M the dual lattice of L. Let be a cone in L Z R. Fix a
primitive element v L . The barycentric subdivision of at v is
the minimal fan containing all cones Cone(, v) where varies over all
subcones of with v 6 .
Proposition 3.2 (see e.g. [17]). Let be the barycentric subdivision of
an n-dimensional cone at v. Let X := U be the affine toric variety
corresponding to and Y the toric variety corresponding to the fan .
(i) The barycentric subdivision induces a projective toric morphism
: Y X.
(ii) The set of 1-dimensional cones of consists of 1-dimensional
cones of and Cone(v).
(iii) The torus invariant prime divisor Dv corresponding to the 1dimensional cone Cone(v) is a Q-Cartier divisor on Y .
Furthermore if v is an interior lattice point in , then
KY = (KX ) + (hx1 x2 xn , vi 1)Dv ,
i.e. the discrepancy of the exceptional divisor Dv is hx1 x2 . . . xn , vi 1.
Example 3.3. Define the lattice L = Z3 + Z 1r (1, a, r a) with r
coprime to a and M = HomZ (L, Z) the dual lattice. Let {e1 , e2 , e3 } be
the standard basis of Z3 and + the cone generated by e1 , e2 , e3 . As in
Section 2.2, the toric variety X := U+ is the quotient variety C3 /G
where G is the group of type 1r (1, a, r a).
Set vi := 1r (i, ai, r ai) L for each 1 i < r 1 where denotes
the residue modulo r. Let Ei be the torus invariant prime divisor
corresponding to vi . From Proposition 3.2, the discrepancy of Ei is
i ai r ai
i
+ +
1 = > 0.
r
r
r
r
This shows that X has only terminal singularities.

We have seen that the quotient singularity X = C3 /G has terminal

singularities if G is the group of type 1r (1, a, r a) with r coprime to a.
Conversely, these groups are essentially all the cases, by the following.

14

S.-J. JUNG

Theorem 3.4 (Morrison and Stevens [14]). A 3-fold cyclic quotient

singularity X = C3 /G has terminal singularities if and only if the
group G GL3 (C) is of type 1r (1, a, r a) with r coprime to a.
3.2. Weighted blowups and round down functions. Define the
lattice L = Z3 + Z 1r (1, a, r a). Set L = Z3 L. Consider the
two dual lattices M = HomZ (L, Z), M = HomZ (L, Z). Note that a
(Laurent) monomial m M is G-invariant if and only if m is in M.
Let {e1 , e2 , e3 } be the standard basis of Z3 and + the cone generated

by e1 , e2 , e3 . Then Spec C[+

M] is the quotient variety X = C3 /G.
Set v = 1r (1, a, r a) L, which corresponds to the exceptional divisor
of the smallest discrepancy (see Example 3.3). Define three cones
1 = Cone(v, e2 , e3 ),

2 = Cone(e1 , v, e3 ),

3 = Cone(e1 , e2 , v).

Define to be the fan consisting of the three cones 1 , 2 , 3 and their

faces. The fan is the barycentric subdivision of + at v. Let Y1 be
the toric variety corresponding to the fan together with the lattice L.
The induced toric morphism : Y1 X is called the weighted blowup
of X with weight (1, a, r a).
e1
2

3
v = 1r (1, a, r a)

e3

e2

Figure 3.1. Weighted blowup of weight (1, a, r a)

Let us consider the sublattice L2 of L generated by e1 , v, e3 . Define
M2 := HomZ (L2 , Z) with the dual basis
1

2 := xy a ,

2 := y a ,

2 := y

ar
a

z.

The lattice inclusion L2 L induces a toric morphism

: Spec C[2 M2 ] U2 := Spec C[2 M].
Since C[2 M2 ]
= C[2 , 2 , 2 ] and the group G2 := L/L2 is of type
1
(1, r, r) with eigencoordinates 2 , 2 , 2 , the open subset U2 has a
a
quotient singularity of type a1 (1, r, r).
Similarly, consider the sublattice L3 of L generated by e1 , e2 , v. Let
us define the lattice M3 := HomZ (L3 , Z) with basis
1

3 := xz ra ,

3 := yz ra ,

3 := z ra .

The open set U3 = Spec C[3 , 3 , 3] has a quotient singularity of type

1
(1, r, r) with eigencoordinates 3 , 3 , 3 . Set G3 := L/L3 .
ra

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

15

Lastly, consider the sublattice L1 of L generated by v, e2 , e3 . Let us

define M1 := HomZ (L1 , Z) with the dual basis
1

1 := x r ,

1 := x r y,

1 := x

ra
r

z.

Since {v, e2 , e3 } forms a Z-basis of L, i.e. G1 = L/L1 is the trivial

group, the open set U1 = Spec C[1 , 1 , 1 ] is smooth.
Example 3.5. Let G be the group of type 17 (1, 3, 4) as in Example 2.10.
The toric fan of the weighted blowup with weight (1, 3, 4) is shown in
Figure 3.2.
e1

v6 = 71 (6, 4, 3)
v5 = 71 (5, 1, 6)
v3 = 71 (3, 2, 5)

v4 = 17 (4, 5, 2)
v2 = 71 (2, 6, 1)

3
v1 =

e3

1
(1, 3, 4)
7

e2

Figure 3.2. Weighted blowup of weight (1, 3, 4)

The affine toric variety corresponding to the cone 2 on the left
side of v = 17 (1, 3, 4) has a quotient singularity of type 31 (1, 2, 1) with
7
4
1
eigencoordinates xy 3 , y 3 , y 3 z. The affine toric variety corresponding
to the cone 3 on the right side of v has a singularity of type 14 (1, 3, 1)
7
3
1
with eigencoordinates xz 4 , yz 4 , z 4 . On the other hand, the affine
toric variety corresponding to the cone 1 = Cone(e2 , e3 , v) is smooth
as e2 , e3 , v form a Z-basis of L.

Definition 3.6 (Round down functions). With the notation above,

the left round down function 2 : M M2 of the weighted blowup
with weight (1, a, r a) is defined by
m3 m3
1 m1 + ar m2 + ra
r
2 .

2 (xm1 y m2 z m3 ) = 2m1 2 r

where is the floor function. In a similar manner, the right round

down function 3 : M M3 of the weighted blowup with weight

16

S.-J. JUNG

(1, a, r a) is defined by
m3
1 m1 + ar m2 + ra
r

3 (xm1 y m2 z m3 ) = 3m1 3m2 3 r

and the central round down function 1 : M M1 of the weighted

blowup with weight (1, a, r a) by
1 m1 + ar m2 + ra
m3 m2 m3
r
1 1 .

1 (xm1 y m2 z m3 ) = 1 r

Lemma 3.7. For each k = 1, 2, 3, let k be the round down function of

the weighted blowup with weight (1, a, r a). For a monomial m M
and a G-invariant monomial n M,
k (m n) = k (m) n.
Thus the weight of k (m n) and the weight of k (m) are the same in
terms of the Gk -action. Therefore k induces a surjective map
k : G Gk ,

7 k (),

where k () is the weight of k (m) for a monomial m M of weight .

Proof. Since Mk contains M as the lattice of Gk -invariant monomials,
n is in Mk . By definition, the assertions follow.

Remark 3.8. Davis, Logvinenko, and Reid [6] introduced a related
construction in a more general setting.

Lemma 3.9. For each k = 1, 2, 3, let k be the round down function of

the weighted blowup with weight (1, a, r a). Let m M be a Laurent
monomial of weight j.
(i) If 0 j < r a, then 2 (y m) = 2 (m).
(ii) If 0 j < a, then 3 (z m) = 3 (m).
(iii) If 0 j < r 1, then 1 (x m) = 1 (m).
(iv) If k (m) = k (m ), then m = n m or m = n m for some
n M 0 .
Proof. Let m = xm1 y m2 z m3 be a Laurent monomial of weight j. To
prove (i), assume that 0 j < r a. This means that
a
ra
1
a
ra
ra
1
m3 m1 + m2 +
m3 <
.
0 m1 + m2 +
r
r
r
r
r
r
r
Thus 2 (y m) = 2 (xm1 y m2 +1 z m3 ) = 2 (xm1 y m2 z m3 ) = 2 (m). The
assertions (ii) and (iii) can be proved similarly. The definition of k
implies (iv).

Lemma 3.10. For each k = 1, 2, 3, let k be the round down function
of the weighted blowup with weight (1, a, r a). Let k be a lattice point
in the monomial lattice Mk and g a monomial of degree 1 in Mk .There
exist a monomial f {x, y, z} of degree 1 and m M such that
k (f m) = g k
satisfying k (m) = k.

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

17

Proof. Here we prove the assertion for the left round down function.
Let , , denote the eigencoordinates for the G2 -action. Let k be a
monomial in M2 and g {, , }.
Consider the case where g = . Since 2 is surjective, there exists
m = xm1 y m2 z m3 M such that 2 (m) = k. If k = 2 (z m), then
we are done.
Suppose k 6= 2 (z m). This means that


1
a
ra
ra
1
a
ra
m1 + m2 +
m3 +

m1 + m2 +
m3 + 1.
r
r
r
r
r
r
r
There is a positive integer l0 3 such that 2 ( m
) = k for all 0 l l0
yl


6= k. Since 2 (z yml0 ) = k, the assertion follows.
with 2 ylm
0 +1
For the other cases, we can prove the assertion similarly.

3.3. Economic resolutions. By the fact that the quotient variety
X = C3 /G has terminal singularities, X does not admit crepant resolutions. However X has a certain toric resolution introduced by
Danilov [5] (see also [17]).
Definition 3.11. Let G GL3 (C) be the group of type 1r (1, a, r a).
For each 1 i < r, let vi := 1r (i, ai, r ai) L where denotes the
residue modulo r. The economic resolution of C3 /G is the toric variety
obtained by the consecutive weighted blowups at v1 , v2 , . . . , vr1 from
C3 /G.
Proposition 3.12 (see [17]). Let : Y X = C3 /G be the economic
resolution of C3 /G. For each 1 i < r, let Ei denote the exceptional
divisor of corresponding to the lattice point vi .
(i) The toric variety Y is smooth and projective over X.
(ii) The morphism satisfies
X i
Ei .
KY = (KX ) +
r
1i<r
In particular, each discrepancy is 0 <

i
r

< 1.

From the fan of Y , we can see that Y can be covered by three open
sets U2 , U3 and U1 , which are the unions of the affine toric varieties
corresponding to the cones on the left side of, the right side of, and
below the vector v = 1r (1, a, r a), respectively. Note that U2 and U3
are isomorphic to the economic resolutions for the singularity of type
1
1
(1, r, r) and of type ra
(1, r, r), respectively.
a
3This

integer l0 is the largest integer satisfying



1
a
ra
a
a
ra
1
m1 + m2 +
m3 l
m1 + m2 +
m3 .
r
r
r
r
r
r
r

18

S.-J. JUNG

Remark 3.13. Let be the toric fan of the economic resolution Y .

Note that the number of 3-dimensional cones in is 2r 1 and that
the number of 3-dimensional cones containing e1 is r.

Example 3.14. Let G be the group of type 17 (1, 3, 4) as in Example 2.10. The fan of the economic resolution of the quotient variety
C3 /G is shown in Figure 3.3.
e1
1

2
v6

v5
v4
v3
v2

e3

v = 17 (1, 3, 4)

e2

Figure 3.3. Fan of the economic resolution for 71 (1, 3, 4)

The toric variety corresponding to the fan consisting of the cones on
the left side of v = 71 (1, 3, 4) is the economic resolution of the quotient
1
(1, 2, 1). On the other hand, the toric variety corresponding to the fan
3
consisting of the cones on the right side of v is the economic resolution
of the quotient 41 (1, 3, 1).

4. Moduli interpretations of economic resolutions

This section contains our main theorem. First, we explain how to
find a set S(r, a) of G-bricks using the round down functions and a
recursion process. In Section 4.3, we show that there exists a stability
parameter such that every G-brick in S(r, a) is -stable.
4.1. G-bricks and stability parameters for 1r (1, r 1, 1). Let G
be the group of 1r (1, r 1, 1) type, i.e. a = 1 or r 1. In this case,
Kedzierski
[10] proved that G-Hilb C3 is isomorphic to the economic

resolution of C3 /G.

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

19

Theorem 4.1 (Kedzierski

[10]). Let G GL3 (C) be the finite group

1
of type r (1, a, r a) with a = 1 or r 1. Then G-Hilb C3 is isomorphic
to the economic resolution of the quotient variety C3 /G.
For each 1 i < r, set vi = 1r (i, r i, i). Set v0 = e2 and vr = e3 .
The toric fan corresponding to G-Hilb C3 consists of the following 2r1
maximal cones and their faces:
i = Cone(e1 , vi1 , vi )

for 1 i r,

r+i = Cone(e3 , vi1 , vi )

for 1 i r 1.

Each 3-dimensional cone has a corresponding (Nakamuras) G-graph:

(4.2)

i = {1, y, y 2, . . . , y i1, z, z 2 , . . . , z ri }
r+i = {1, y, y 2, . . . , y i1, x, x2 , . . . , xri }

for 1 i r,
for 1 i r 1,

with S(j ) = j M. As the cone j is 3-dimensional, the G-prebrick

j is a G-brick. Furthermore, U(j ) = D(j )
= C3 .
By Ito-Nakajima [7], all G-bricks in (4.2) are -stable for any +
where



(4.3)
+ :=  () > 0 for 6= 0 .

Example 4.4. Let G be the finite group of type 31 (1, 2, 1) with eigencoordinates , , . Set v1 = 31 (1, 2, 1) and v2 = 13 (2, 1, 2). Recall that the
economic resolution Y of X = C3 /G can be obtained by the sequence
of the weighted blowups:
2

Y Y1 X,
where 1 is the weighted blowup with weight (1, 2, 1) and 2 is the
toric morphism induced by the weighted blowup with weight (2, 1, 2).
The fan corresponding to Y consists of the following five 3-dimensional
cones and their faces:
1 = Cone(e1 , e2 , v1 ), 2 = Cone(e1 , v1 , v2 ), 3 = Cone(e1 , v2 , e3 ),
4 = Cone(e3 , e2 , v1 ), 5 = Cone(e3 , v1 , v2 ).
The following
1 = {1, , 2}, 2 = {1, , },
4 = {1, , 2}, 5 = {1, , }.
are their corresponding G-bricks.

3 = {1, , 2},

4.2. G-bricks for 1r (1, a, r a). In this section, we assign a G-brick

with S( ) = M to each maximal cone in the fan of the
economic resolution Y .
Let G be the group of type 1r (1, a, r a) with r coprime to a, X the
quotient C3 /G, and : Y X the economic resolution of X. Then
Y can be covered by U2 , U3 and U1 , which are the unions of the affine

20

S.-J. JUNG

toric varieties corresponding to the cones on the left side of, the right
side of, and below the lattice point v = 1r (1, a, r a), respectively.
Proposition 4.5. For k = 1, 2, 3, let be a Gk -brick. Define



:= m M  k (m) .
The set is a G-brick with S() = S( ).

Proof. Since k (1) = 1 , we have 1 . To show that satisfies

(ii) in Definition 2.8, we need to show that there exists a unique monomial of weight in for each G . Fix a positive integer i such
that the weight of xi is . Consider the monomial k (xi ) in Mk and its
weight Gk . Since is a Gk -brick, there exists a unique element
k of weight . Since the Gk -invariant monomial kk(x i ) is in the lattice
M, it follows from Lemma 3.7 that
 k 

7 k ,
k : xi
2 (xi )


i.e. xi 2k(x i ) is in . To show uniqueness, assume that two monomials
m, m of the same weight are mapped into . As k (m) and k (m ) are
of the same weight, we have k (m) = k (m ) . From Lemma 3.7,

m
m
k (m) = k m = k (m ) ,
m
m

and hence m = m . From Lemma 3.10, it follows that is connected

as is connected.
For (iii) in Definition 2.8, assume that n n m for m and
n, n M 0 . We need to show k (n m ) . From
k (n n m ) =

k (n n m ) k (n m )

k (m ) ,
k (n m )
k (m )

)
it follows that k (n m ) = kk(nm
k (m ) is in because is a
(m )
Gk -prebrick. This proves that is a G-prebrick.
It remains to prove that S() = S( ). Note that S() is generated

by wtnm
for n M 0 and m . Let n be a genuine monomial in
(nm )
M 0 and m an element in . Let k denote k (m ) . Define k to
)
. From the definition of the round down functions, we know
be kk(nm
(m )

is G-invariant,
that k is a genuine monomial in , , . Since wtnm
(nm )
it follows that

k (nm )
k (m )
k (n m )
n m
k k

= k (m )

=
=
wt (n m )
wt (k k )
k wt (n m )
k wt (n m )

from Lemma 3.7. This proves S() S( ).

For the opposite inclusion, by Lemma 2.11, it suffices to show that
gk
is in S() for all g {, , } and k . By Lemma 3.10
wt (gk )

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

21

there are n M 0 , m such that k (n
m ) = g k . Lemma 4.6

implies that wt (g k ) = k wt (n m ) . Thus
k (n m )
k (n m )
n m
g k

=

=
=
,
wt (g k )
wt (n m )
wt k (n m )
k wt (n m )

and we proved the proposition.

Lemma 4.6. With the notation as in Proposition 4.5, if m M , then

wt k (m) = k wt (m) .


Proof. Since k (m) is of the same weight as k wt (m)
Lemma 3.7,

by

the assertion follows from the fact that k wt (m) .


Lemma 4.7. With the notation as in Proposition
4.5, let m be the


Laurent monomial of weight j in = m M k (m) .
(i) If k = 2 and 0 j < r a, then 2 (y m) .
(ii) If k = 3 and 0 j < a, then 3 (z m) .
(iii) If k = 1 and 0 j < r 1, then 1 (x m) .

Proof. Lemma 3.9 implies the assertion.

Proposition 4.8. Let G be the group of type 1r (1, a, r a) with r coprime to a. Let max be the set of maximal cones in the fan of the
economic resolution Y of X = C3 /G. Then there exists a set S(r, a)
of G-bricks such that there is a bijective map max S(r, a) sending
to satisfying S( ) = M. In particular, U( ) is isomorphic
to the smooth toric variety U = Spec C[ M] corresponding to .
Proof. From Section 4.1, the assertion holds when a = 1 or r 1. We
use induction on r and a.
Let be a 3-dimensional cone in the fan of the economic resolution
Y of X = C3 /G. For v = 1r (1, a, r a), we have three cases:
(1) the cone is below the vector v.
(2) the cone is on the left side of the vector v.
(3) the cone is on the right side of the vector v.
Case (1) the cone is below the vector v. Since there is a unique
3-dimensional cone below v, = Cone(v, e2 , e3 ). Consider the central
round down function 1 of the weighted blowup with weight (1, a, ra).
For m = xm1 y m2 z m3 M, note that
1 (m) = 1 if and only if m2 = m3 = 0 and 0

m1
r

< 1.

2
r1
The set := 1
} is a G-prebrick with the
1 (1) = {1, x, x , . . . , x

property S() = M. Since the corresponding cone () is equal

to , the G-prebrick is a G-brick.

22

S.-J. JUNG

Case (2) the cone is on the left side of v. From the fan of the
economic resolution, it follows that U2 is isomorphic to the economic
resolution Y2 of a1 (1, r, r) with eigencoordinates , , . There exists
a unique 3-dimensional cone in the toric fan of Y2 corresponding to
. Let G2 be the group of type a1 (1, r, r). Note that a is strictly less
than r so that we can use induction.
Assume that there exists a G2 -brick with S( ) = ( ) M. By
Proposition 4.5, there is a G-brick with S() = S( ) = M.
Case (3) the cone is on the right side of v. The case where the
cone is on the right side of v can be proved similarly.

Definition 4.9. A G-brick in S(r, a) described above is called a
Danilov G-brick.
Proposition 4.10. With the notation as is in Proposition 4.5, we have
D( )
= D(). Moreover we have a commutative diagram

U( ) U()

y
y

D( ) D()
with the vertical morphisms closed embeddings. Therefore for a G-brick
S(r, a), we have U() = D()
= C3 .
Proof. Let be a G-brick and the corresponding Gk -brick. Let
, , denote the eigencoordinate for the Gk -action. From (2.19), the
coordinate rings of the affine schemes D(), D( ) are


C[D()] = C[x , y , z  G ] I ,


C[D( )] = C[ , ,  Gk ] I



 f () and the
where the ideal
I
is
the
quadrics
in
(2.18),
f


ideal I is the commutative relations, g 1  g ( ) .
By Lemma 3.10, we have an algebra epimorphism

: C[x , y , z  G ] C[D( )] f 7 k()

defined as follows on the 3r generators f {x , y , z }. Let m be

the unique element of weight in and the weight of k (f m ).
m )
Then k := kk(f(m
is a monomial, so k induces a linear map k() on
)
the vector space C k (m ). Then is the morphism sending f to
k() . Since the generators of I are in ker , induces an epimorphism
: C[D()] C[D( )].
To construct the inverse of , first we show that if (f ) = (f ),
then f f mod I . If (f ) = (f ), then k (f m ) = k (f m )
and k (m ) = k (m ). Since both f and f are degree of 1, f = f .
By (iv) in Lemma 3.9, we may assume that m = n m for some
n M 0 . Since k (m ) = k (m ), n induces a linear map equal to

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

23

1 on m , i.e. m() 1 mod I because m = n m is a base. From

the following commutative diagram
C m

yf

C m

yf

C wt (f m ) C wt (f m ),
it suffices to show that n wt (f m ) is a base, which implies that n
induces a linear map equal to 1 on wt (f m ). Since

wt (f m ) 

k (n wt (f m )) = k f m
f m
wt (f m )
= k (f m )
f m


wt (f m )
= k (f m )
= k wt (f m ) ,
f m
the monomial n wt (f m ) is in and is a base.
Now define the algebra morphism : C[ , ,  Gk ] C[D()]
by (g ) = f for g { , , } so that (f ) = g . Since the generators of I are in ker , induces : C[D( )] C[D()].
To show is surjective, we prove that generators f are in the image
m )
is of degree 1, f is in the image of
of . For f such that kk(f(m
)
by definition. By Lemma 4.11 below, it follows that is surjective.
Since and are the inverses of each other, D() is isomorphic to
D( ). Note that U() = D()
= C3 for S(r, 1) from Section 4.1.


Using induction, we get D() = C3 for S(r, a).
Lemma 4.11. In the situation as in Proposition 4.10, define
n
o

m )
S := f {x , y, z }  kk(f(m
is
of
degree

1
.
)

m )
is of degree 2 for some f {x , y , z }, then f can be
If kk(f(m
)
written as a multiple of some elements in S modulo I .

Proof. We prove this for the left round down function 2 . Note that
k (ym )
is of degree 1 for all G . Thus y s are in S.
k (m )
m )
is of degree 2 with m = xm1 y m2 z m3 . Then
Suppose that kk(f(m
)
the monomial f is either x or z. In the case where f = z, this means
that


a
ra
1
a
ra
a
1
m1 + m2 +
m3 m1 + m2 +
m3 .
r
r
r
r
r
r
r

As in the proof of Lemma 3.10, there is a positive
integer l such that

2 ( m
) = 2 (m ) for all 0 l l with 2 yml+1 6= 2 (m ). Note
y l

24

S.-J. JUNG
2 (f m )

that 2 (m ) is of degree 1 where m =

commutation relations
C m

yf

m
.
yl

y l

Thus f S. From the

C m

yf

y l

C wt (f m ) C wt (f m ),
since y l induces a linear map on C m set to be 1, we have
l
l
f f y(
) y() f

mod I .

As all y s are in S, the assertion follows.

Example 4.12. Let G be the group of type 71 (1, 3, 4) as in Example 2.10. The fan of the economic resolution of the quotient variety is
shown in Figure 3.3.
e1

e1

1 2

e1

1
(1, 2, 1)
3

e3

1
(1, 2, 1)
3

1
(1, 3, 1)
7

e2

e3

1
(1, 3, 1)
4

e2

1
(1, 3, 1)
7

e3

1
(1, 3, 1)
4

e2

Figure 4.1. Recursion process for 17 (1, 3, 4)

We now calculate G-bricks associated to the following cones:


1 := Cone (1, 0, 0), 71 (1, 3, 4), 17 (3, 2, 5) ,


2 := Cone (1, 0, 0), 17 (6, 4, 3), 17 (1, 3, 4) .

Note that the left side of the fan corresponds to the economic resolution
for the quotient singularity of type 13 (1, 2, 1), which is G2 -Hilb C3 , where
G2 is of type 31 (1, 2, 1). Let , , denote the eigencoordinates. Let 1
be the cone in the fan of G2 -Hilb C3 which corresponds to 1 . Observe
that the corresponding G2 -brick is


1 = 1, , 2 .
Since the left round down function 2 is
1

2 (xm1 y m2 z m3 ) = m1 7 m1 + 7 m2 + 7 m3 m3 ,

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

25

the G-brick corresponding to 1 is



def 
1 = xm1 y m2 z m3 M  2 (xm1 y m2 z m3 ) 1

2
2
= 1, y, y 2, z, yz , zy , yz 2 .

On the other hand, the right side of the fan corresponds to the economic resolution of the quotient variety 14 (1, 3, 1) which is G3 -Hilb C3 ,
where G3 is of type 41 (1, 3, 1) with eigencoordinates , , . Let 2 be
the cone in the fan of G2 -Hilb C3 which corresponds to 2 . Observe
that the corresponding G3 -brick is


2 = 1, , 2, 3 .
Since the right round down function 3 is

3 (xm1 y m2 z m3 ) = m1 m2 7 m1 + 7 m2 + 7 m3 ,
the G-brick corresponding to 2 is


def 
2 = xm1 y m2 z m3 M  2 (xm1 y m2 z m3 ) 2

2
3
3
= 1, z, y, y 2, yz , yz , yz 2 .

From Example 2.13, (1 ) = 1 and (2 ) = 2 .

4.3. Stability parameters for S(r, a). Let G GL3 (C) be the finite
subgroup of type 1r (1, a, r a) with r coprime to a. We may assume
2a < r. Let G2 and G3 be the groups of type a1 (1, r, r) and of type
1
(1, r, r), respectively.
ra
Given stability conditions (2) for Danilov G2 -bricks and (3) for
Danilov G3 -bricks, take a GIT parameter P satisfying the following system of linear equations:

P
(2)

P ()
for all G2 ,
() =
2 ()=
P
(4.13)
(3)

P () for all G3 .

(
)
=

3 ()=

Define the GIT parameter by

1 if 0 wt() < a ,
(4.14)
() = 1
if r a wt() < r,

0
otherwise.
P
Observe that
() = 0 for all Gk 4. For a sufficiently large
k ()=

natural number m, set

(4.15)

:= P + m.

P
addition, if any satisfies that k ()= () = 0 for all G
k and
k = 2, 3, then must be a constant multiple of . This also explains the existence
of a solution P for (4.13).
4In

26

S.-J. JUNG

We claim that every S(r, a) is -stable.

Example 4.16. As in Example 4.12, let G be the group of type
1
(1, 3, 4). For each 0 i 6, let i denote the irreducible repre7
sentation of G whose weight is i. We saw that the left side of the fan
is G2 -Hilb C3 , where G2 is of type 31 (1, 2, 1) and that the right side of
the fan is G3 -Hilb C3 , where G3 is of type 14 (1, 3, 1). Let {0 , 1 , 2 }
and {0 , 1 , 2 , 3 } be the characters of G2 and G3 , respectively. Take
GIT parameters (2) , (3) corresponding to G-Hilb such as (see (4.3)):
(2) = (2, 1, 1),
We have the following

1
3

1
1

(3) = (3, 1, 1, 1).

system of linear equations:

=
=
=
=
=
=
=

P (0 ) + P (3 ) + P (6 ),
P (1 ) + P (4 ),
P (2 ) + P (5 ),
P (0 ) + P (4 ),
P (1 ) + P (5 ),
P (2 ) + P (6 ),
P (3 ).

Take P = (1, 3, 3, 1, 2, 2, 2) as a partial solution. For the parameter = (1, 1, 1, 0, 1, 1, 1), define = P + m for large m.
Consider the following G-brick
n
o
2
2
= 1, y, y 2, z, yz , zy , yz 2 .

2
Let F be the submodule of C() with basis A = z, yz , zy . Note
that (F ) > 0 and

n z z2 z2 o
1

(A)
= z, y , y , y2 ) A.
2
2
Thus (F ) is positive for large enough m. More precisely,

(F ) = 3 m + (2 + m) + (2 + m) = m 1
is positive if m > 1.
Onthe other
hand, consider the submodule G
of C() with basis

z z2
B = y , y . Note that (G) = 0 and 1
2 (B) = B. In this case,
2

the set 2 (B) gives a submodule G of C( ) with

(2) (G ) = (G).
Since C( ) is (2) -stable, (2) (G ) is positive. Hence (G) is positive.
Lemma 4.17. Let be the parameter in (4.15). For the set S(r, a) in
Proposition 4.8, if is in S(r, a), then is -stable.
Proof. Let be a G-brick in S and the cone corresponding to . We
have the following three cases as in Section 4.2:
(1) the cone is below the vector v.

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

27

(2) the cone is on the left side of the vector v.

(3) the cone is on the right side of the vector v.
In Case (1), = {1, x, x2 , . . . , xr2 , xr1 }. By Lemma 2.14, any
nonzero proper submodule G of C() is given by
A = {xj , xj+1 , . . . , xr2 , xr1 }
for some 1 j r 1. Since (G) > 0, is -stable for sufficiently
large m.
We now consider Case (2). Let be the G2 -brick corresponding to
. Let G be a submodule of C() with C-basis A . Lemma 3.9
and Lemma
2.14 imply that if m A for 0 wt(m ) < a, then
1
2 (m ) A. Thus (G) 0 from the definition of .
2
If (G) > 0, then it follows that (G) > 0 for sufficiently
large m.
1
Let us assume that (G) = 0.

Note that A = 2 2 (A) ; otherwise
1
there exists m in 2 2 (A) \ A with 0 wt(m ) < a. To show
that (G) is positive, we prove that 2 (A) gives a submodule G of
C( ) and that (G) = (2) (G ). Since satisfies the equations (4.13),
it suffices to show that 2 (A) gives a submodule of C( ). Let , , be
the coordinates of C3 with respect to the action of G2 . By Lemma 2.14,
it is enough to show that if g 2 (m ) for some g {, , } and
m A, then g 2 (m ) 2 (A). Suppose that g 2 (m ) for
some m A. By Lemma 3.10, there exists m such that
2 (f m ) = g 2 (m )
with 2 (m ) = 2 (m ) for

some f {x, y, z}. In particular, f m is in
. Since A = 1

(A)
, we have m A, which implies f m A
2
2
as A is a C-basis of G. Thus g 2 (m ) is in 2 (A).

Remark 4.18. At this moment, our stability parameter in (4.15) has
nothing to do with Kedzierskis
GIT chamber C(r, a) described in [11].

In Section 5, it is shown that the parameter is in C(r, a).

4.4. Main Theorem.

Theorem 4.19. The economic resolution Y of a 3-fold terminal quotient singularity X = C3 /G is isomorphic to the birational component
Y of the moduli space M of -stable G-constellations for a suitable
parameter .
Proof. From Proposition 4.8 and Lemma 4.17, Proposition 2.21 implies
that there exists an open immersion from Y to Y fitting in the following
commutative diagram:
Y Y

X.

28

S.-J. JUNG

Since both Y and Y are projective over X, the open immersion Y

Y is a closed embedding. As both Y and Y are 3-dimensional and
irreducible, this embedding is an isomorphism.

Conjecture 4.20. The moduli space M is irreducible.
Proposition 4.10 implies that the irreducible component Y is actually
a connected component. In addition, if every torus invariant -stable
G-constellation lies over the birational component Y , then M is irreducible. For a = 2, we can prove Conjecture 4.20 so the economic
resolution is isomorphic to M for S(r, a) (See [8]). We hope to
establish this more generally in future work.
Remark 4.21. By construction, M0 = Spec C[Rep G]GL() is the moduli space of 0-semistable G-constellations up to S-equivalence. Since
there exists an algebra isomorphism C[Rep G]GL() C[x, y, z]G , M0
is isomorphic to C3 /G. In particular, M0 is irreducible.

5. Kedzierskis
GIT chamber

Kedzierski
[11] described his GIT cone in using a set of inequalities.

Using his lemma, we can prove further that the cone is actually a GIT
chamber C. In this section, we provide a description of C using the
Ar1 root system. Define

S(r, a)0 = { S(r, a)  x 6 }.
Kedzierskis
lemma. By the same argument as in Lemma 6.7 of [11],

we can prove that it suffices to check the -stability for G-bricks not
containing x.
Lemma 5.1 (Kedzierskis
lemma [11]). For a parameter , the

following are equivalent.

(i) Every S(r, a) is -stable.
(ii) Every S(r, a)0 is -stable.
Let A be the finite group of type 1r (a, r a). Since A
= G as groups,
the GIT parameter space of G-constellations can be canonically identified with that of A-constellations.
Since G-constellations which x acts trivially on are supported on the
hyperplane (x = 0) C3 , they can be considered as A-constellations.
As S(r, a)0 is the set of G-bricks corresponding to G-constellations
supported on (x = 0) C3 , Lemma 5.1 implies that the GIT chamber
for S(r, a) is equal to a GIT chamber of A-constellations.
Kedzierskis
GIT chamber. We describe a set of simple roots so

that Y is isomorphic to the economic resolution for C(). After

considering the case of a = 1, we describe simple roots for the case of
1
(1, a, r a) using a recursion process.
r

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

29

 
Root system Ar1 . Identify I := Irr(G) with Z/rZ. Let i  i I be
an orthonormal basis of Qr , i.e. hi , j i = ij . Define

:= {i j  i, j I, i 6= j}.

Let h be the subspace of Qr generated by . Elements in are called

roots.
For each nonzero i I, set i = i ia . Let i denote the irreducible representation of G of weight i. Note that each root can be
considered as the support of a submodule of a G-constellation. In other
words, i corresponds
P to the dimension vector of i . In general we consider a root = i ni i as the dimension
vector of the representation
P
ni i . Abusing notation, let = i ni i denote the corresponding
representation ni i .
Let be a set of simple roots. Define C() associated to as

C() := {  () > 0 }.
Note that for the cone + for G-Hilb in (4.3), the corresponding set of
simple roots is


+ = {i ia  i I, i 6= 0} = {i  i I, i 6= 0}.

The case of 1r (1, r 1, 1). From Theorem 4.1, we know that the economic resolution of X = C3 /G is isomorphic to G-Hilb C3 if G is of
type 1r (1, r 1, 1). Thus in this case, the G-bricks are just Nakamuras
G-graphs, which are -stable for + , where



+ :=  () > 0 for 6= 0 .
The corresponding set of simple roots is



= i i+1  i I, i 6= 0

= {1 2 , 2 3 , . . . , r1 0 } .

 
Example 5.2. For the group of type 31 (1, 2, 1), let Lj  j = 0, 1, 2 be
the standard basis of Q3 . The corresponding set of simple roots is


L = L1 L2 , L2 L0 .
 

 k = 0, 1, 2, 3 the
Similarly, for the group of type 14 (1, 3, 1) with R
k
standard basis of Q4 ,
R R
R R
R
R = {R
1 2 , 2 3 , 3 0 }

is the corresponding set of simple roots for type 14 (1, 3, 1).

30

S.-J. JUNG

The case of 1r (1, a, r a). Let G be the group of type 1r (1, a, r a). Let
L and R denote the sets of simple roots for the types of a1 (1, r, r)
1
and of ra
(1, r, r), respectively. As in Section 5, let

 L


l  l = 0, 1, . . . , a 1}, {R
k k = 0, 1, . . . , r a 1

be the standard basis of Qa and Qra , respectively.

From the two sets L and R , we construct a set of simple
in

 roots


Ar1 as follows. First, as in Section 5, let the standard basis i i I
of Qr be identified with the union of the two sets


 L

l  l = 0, 1, . . . , a 1} and {R
k k = 0, 1, . . . , r a 1
using the following identification:
 L
l = i with i l mod a
if r a i < r,
(5.3)
R
k = i with i k mod (ra) if 0 i < r a.
With the identification above, define to be
(5.4)

= L { r1 a r1 (ra)a } R .
a

ra

The root r1 a r1 (ra)a is called the added root in . Note that

a
ra
is actually a set of simple roots in Ar1 .
Definition 5.5. With as above, the corresponding Weyl chamber

C(r, a) := C() = {  () > 0 }

is called Kedzierskis
GIT chamber for G = 1r (1, a, r a).

Proposition 5.6. Let C(r, a) be Kedzierskis

GIT chamber.

(i) The parameter in (4.14) is a ray of C(r, a).

(ii) Any G-brick in S(r, a) is -stable for C(r, a).
(iii) The cone C(r, a) is a full GIT chamber.
Proof. We may assume a < r a. First, by construction, is zero
on the sets L and R with the identification (5.4). To prove (i), it
remains to show that () is positive where is the added root in .
Since
X
= r1 a r1 (ra)a =
i + ra ,
a

ra

2 (i )=0

where 0 is the trivial representation of G2 , (i) follows.

For defined by (4.15), every S(r, a)0 is -stable. For the
group A of type 1r (a, a), Kronheimer [13] showed that the chamber
structure of the GIT parameter space of A-constellations is the same
as the Weyl chamber structure of Ar1 5. Thus for S(r, a)0 considered
as A-constellations, we have a Weyl chamber of the Ar1 root system
containing the parameter .
By Kedzierskis
lemma, to prove (ii), it suffices to show that C(r, a)

contains the parameter . Observe that every parameter in C(r, a)

5For

an explicit description, see Section 5.1 in [8]

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

31

satisfies the system of equations (4.13) for some (2) C(a, r) and
(3) C(r a, r) by construction. Since in (4.14) is a ray of the
chamber C(r, a), it follows that C(r, a).
It remains to prove (iii). By considering G-constellations supported
on the hyperplane (x = 0) C3 , it follows that any facet of C(r, a) is
an actual GIT wall in . Therefore Kedzierskis
GIT chamber C(r, a)

is a full GIT chamber in the stability parameter space (see [8,9]). 

Proposition 5.7. Assume that a < r a. Let be an element in
C(r, a). Then (i ) is negative if and only if 0 i < a. Thus any
-stable G-constellation is generated by 0 , 1 , . . . , a1 .
Proof. Let be the set of simple roots corresponding to C(r, a). Recall
that any positive sum of simple roots is positive on .
Suppose that 0 i < a. From the identification (5.3), note that i
is identified with R
k for some k and that ia = i+(ra) is identified
with Ll for some l. Note that r1 a is identified with a vector L and
a
that r1 (ra)a is identified with a vector R . Since we added the
ra
L
root r1 a r1 (ra)a to , the root i = i ia = R
k l is a
a
ra
negative sum of simple roots in .
Suppose that a i < r a. The root i = i ia is a sum of
simple roots in R . A recursive argument yields that i is a positive
sum of simple roots in R . Thus i is a positive sum of simple roots
in .
Consider the case where r a i < r and the root i = i ia .
From the identification (5.3), i is identified with Lk for some k and ia
L
R
is identified with R
l for some l. Thus i = k l is a positive sum of
simple roots in with the same reason as the case where 0 i < a. 
Example 5.8. Let G be the group of type 71 (1, 3, 4). From the fan of
the economic resolution of this case (see Example 3.14), the left and
right sides are the economic resolutions of singularities of 31 (1, 2, 1) and
1
(1, 3, 1), respectively. By Example 5.2, we have two sets
4
R R
R R
R
L = {L1 L2 , L2 L0 } and R = {R
1 2 , 2 3 , 3 0 }.

As in the construction (5.4), the corresponding set of simple roots is

= {4 5 , 5 6 , 6 1 , 1 2 , 2 3 , 3 0 }
= {4 + 1 , 5 + 2 , 1 5 2 , 1 + 5 , 2 + 6 , 3 },
where the underlined root is the added root as in (5.4). Thus the set
of parameters satisfying
(4 1 ) > 0, (5 2 ) > 0, (1 5 2 ) < 0,
(1 5 ) > 0, (2 6 ) > 0,
(3 ) > 0
is Kedzierskis
GIT chamber C(r, a) where i is the irreducible repre
sentation of G of weight i.

32

S.-J. JUNG

e1
v5
12

11

10

v10
3

v3
v8
v1
6
2

v6

v11

7
v4

v9

8
v2

4
10
0

v7

e3

e2
Figure 6.1. Toric fan of the economic resolution for

The rays of the chamber C(r, a) are the

1
0
0 1 0
1
0
0 0 0

0 1 0 0
1

1
1
1 0 1

1 1
0 0 1
1
0
0 0 1

1
(1, 7, 5)
12

row vectors of the matrix

0 0
0 1

1 1

1 1
1 0
0 0

with the dual basis {i } with respect to {i }. Observe that for any
C(r, a), (i ) is negative if and only if 0 i < 3.

6. Example: type

1
(1, 7, 5)
12

In this section, as a concrete example, we calculate Danilov G-bricks

and the corresponding set of simple roots for the group G of type
1
(1, 7, 5).
12

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

33

1
(1, 7, 5) with eigencoordinates
Let G be the finite group of type 12
1
3
(1, 7, 5). Let X denote the
x, y, z and L the lattice L = Z + Z 12
3
quotient variety C /G and Y the economic resolution of X. The toric
fan of Y is shown in Figure 6.1.
To use the recursion process in Section 4, first we need to investigate
the cases of type 71 (1, 2, 5) and of type 51 (1, 2, 3). Let G2 be the group
of type 17 (1, 2, 5) with eigencoordinates 2 , 2 , 2 and G3 be the group of
type 15 (1, 2, 3) with eigencoordinates 3 , 3 , 3 . Consider the toric fans
2 and 3 of the economic resolutions for the type 17 (1, 2, 5) and the
type 51 (1, 2, 3), respectively.

G-bricks. We now calculate G-bricks corresponding to the following

two maximal cones in :


1
1
1
(12, 0, 0), 12
(3, 9, 3), 12
(8, 8, 4) ,
4 = Cone 12


1
1
1
(1, 7, 5), 12
(3, 9, 3), 12
(8, 8, 4) .
3 = Cone 12
The cones 4 , 3 are on the right side of the lowest vector v =
Their corresponding cones 4 , 3 in 3 , respectively, are


4 = Cone 51 (5, 0, 0), 15 (1, 2, 3), 51 (1, 1, 4) ,


(6.1)
3 = Cone 15 (0, 0, 5), 15 (1, 2, 3), 51 (1, 1, 4) .

1
(1, 7, 5).
12

Observe that the cones 4 , 3 are on the left side of 3 . To use the
recursion, let G32 be the group of type 21 (1, 1, 1) with eigencoordinates
32 , 32 , 32 . Let 32 denote the fan of the economic resolution of the
quotient C3 /G32 . In 32 , there exist two cones 4 , 3 corresponding to
4 , 3 , respectively:


4 = Cone 21 (2, 0, 0), 21 (0, 2, 0), 12 (1, 1, 1) ,


3 = Cone 12 (0, 0, 2), 21 (0, 2, 0), 12 (1, 1, 1) .
e1

e1

4
3

3
e3

e2
1
(1, 1, 1)
2

e1

e3

e2

e3

1
(1, 2, 3)
5

Figure 6.2. Recursion process for

e2
1
(1, 7, 5)
12

1
(1, 7, 5)
12

As in Section 4.1, the G32 -bricks 4 , 3 corresponding to 4 , 3 are



4 = 1, 23 ,


3 = 1, 23 .

34

S.-J. JUNG

Using the left round down function 32 for 51 (1, 2, 3)

a+2b+3c
c
5
32 ,

a
32 : 3a 3b 3c 7 32
32

we can see that the G3 -bricks 4 , 3 corresponding to 4 , 3 are


def

4 = 1
= 1, 3 , 32 , 3 , 33 ,
32 4




def

3 = 1
= 1, 3 , 32 , 3 , 3 3 .
32 2

To get the G-bricks 4 and 3 corresponding to 4 and 3 , respectively,

1
we use the right round down function 3 for 12
(1, 7, 5):

a+7b+5c
12

3 : xa y b z c 7 3a 3b 3

We get

o
n
def
y y2 y2
2 3 4 z4 z5 z6

,
,
,
,
z,
z
,
z
,
z
,
,
,
4 = 1
(
)
=
1,
y,
2
3
4
z z z
y y y
o
n
2
2
def
y y
y
xy
2

2
.
3 = 1
3 (2 ) = 1, x, xz, xz , xy, z , y, z , z , z 2 , z, z

Let us consider the following two cones in :

,


1
1
1
7 = Cone 12
(2, 2, 10), 12
(9, 3, 9), 12
(4, 4, 8) .

9 = Cone

1
1
1
(12, 0, 0), 12
(9, 3, 9), 12
(4, 4, 8)
12

Observe that the cones 9 , 7 are on the left side of v. The cones in 2
corresponding to 9 , 7 are


9 = Cone 17 (12, 0, 0), 71 (5, 3, 4), 17 (2, 4, 3) ,


7 = Cone 17 (1, 2, 5), 17 (5, 3, 4), 71 (2, 4, 3) .

Note that the cones 9 , 7 are on the right side of the fan 2 and that
the right side is equal to the fan 3 of the economic resolution for
1
(1, 2, 3). Moreover, the cones in 3 corresponding to 9 , 7 are 4 , 3 ,
5
respectively, in (6.1). Thus the corresponding G23 -bricks 9 , 7 are:


2
9 = 1, 23 , 23
, 23 , 23
,
23


2
7 = 1, 23 , 23 23 , 23 , 23
,
where G23 denotes the group of type 51 (1, 2, 3) with eigencoordinates
23 , 23 , 23 . Using the right round down function 23 for 71 (1, 2, 5)

a+2b+5c
7

a b
23 : 2a 2b 2c 7 23
23 23

we can calculate the G2 -bricks corresponding to 9 , 7 :

o
n
3
2
def

2
2 2 2
,
9 = 1
(
)
=
1,

,
2
2
23
9
2
2 2 2 ,
n
o
def

2
2
7 = 1
(
)
=
1,

2
2 2
2 2
2
2
23
7
2
2 .

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

Lastly, from the left round down function 2 for

35

1
(1, 7, 5)
12

c
a+7b+5c
12
2 ,

2 : xa y b z c 7 2a 2

it follows that the G-bricks 9 , 7 corresponding to 9 , 7 are:

n
o
2
2
2
3
9 = 1, y, y 2, y 3, y 4, y 5, z, z 2 , zy , yz 2 , yz 3 , yz 3 ,
n
o
2
3
2 3
4 5
7 = 1, x, xy, xy , xy , xz, y, y , y , y , y , z .
1
(7i, i, 12 i) in L.
For 0 i 12, let vi denote the lattice point 12
For each 3-dimensional cone in Figure 6.1 on page 32, Table 6.1 on
page 36 shows the corresponding G-brick .

Kedzierskis
GIT chamber. We calculate Kedzierskis
GIT chamber

1
for 12 (1, 7, 5). Since the economic resolution is G-Hilb for the group of
type 1r (1, r 1, 1), the sets of simple roots for 21 (1, 1, 1) and 13 (1, 2, 1)
are {1 0 }, {1 2 , 2 0 }, respectively. By the identification (5.3),
the set of simple roots for 51 (1, 2, 3) is
{3 4 , 4 1 , 1 2 , 2 0 },
where the underlined root is the added root as in (5.4). Similarly, the
admissible set of simple roots for 17 (1, 2, 5) is
{5 6 , 6 3 , 3 4 , 4 1 , 1 2 , 2 0 }.
1
Thus the corresponding set of simple roots for 12
(1, 7, 5) is


5 6 , 6 10 , 10 11 , 11 8 , 8 9 , 9 7 ,
.
7 3 , 3 4 , 4 1 , 1 2 , 2 0

With the dual

following matrix

1
1

1
1

1
1

basis {i } with respect to {i }, the row vectors of the

are the rays of the admissible Weyl chamber C(r, a):

0
0
0
0 0 0 1 0 0 0 0
0 1
0
0 0 0 1 0 1 0 0

1 1
0
0 0 0 1 1 1 0 0

1 1
0 1 0 0 1 1 1 0 1

1 1 1 1 0 0 1 1 1 1 1

1
0 1 1 0 0 0 1 1 1 1 .

1
0 1
0 0 0 0 1 0 1 1
1
0
0
0 0 0 0 0 0 1 1

1
0
0
0 0 1 0 0 0 1 0

1
0
0
0 1 1 0 0 0 0 0
0
0
0
0 1 0 0 0 0 0 0

36

S.-J. JUNG

Cone Generators

Coordinates
on U

G-brick

e1 , e2 , v11

1, z, z 2 , z 3 , z 4 , z 5 , z 6 , z 7 , z 8 , z 9 , z 10 , z 11

x
, y , z 12
z 5 z 11

e1 , v10 , v11

1, y, yz , z, z 2 , z 3 , z 4 , z 5 , z 6 , z 7 , z 8 , z 9

x y 2 z 11
, ,
z 5 z 10 y

e1 , v9 , v10

1, y, yz , yz , yz 2 , yz 3 , yz 4 , yz 5 , z, z 2 , z 3 , z 4

e1 , v8 , v9

1, y, yz , yz , yz 2 , z, z 2 , z 3 , z 4 , zy , zy , zy

e1 , v7 , v8

1, y, yz , yz , yz 2 , yz 2 , yz 3 , yz 4 , yz 4 , yz 5 , z, z 2

e1 , v6 , v7

1, y, z, z 2 , zy , zy , zy2 , yz 2 , zy2 , yz 3 , yz 3 , yz 4

e1 , v5 , v6

1, y, y 2, y 3 , z, z 2 , zy , zy , yz 2 , yz 2 , yz 2 , yz 3

e1 , v4 , v5

1, y, y 2, y 3, y 4, y 5 , yz , yz 2 , yz 2 , z, z 2 , zy

e1 , v3 , v4

1, y, y 2, y 3, y 4, y 5 , z, z 2 , zy , yz 2 , yz 3 , yz 3

10

xz 5 y 3 z 10
, ,
y2 z 9 y2

xy y 4 z 9
, ,
z 4 z 8 y3

xz 4 y 5 z 8
, ,
y3 z 7 y4

xy 2 y 6 z 7
, ,
z 3 z 6 y5

xy 2 y 7 z 6
, ,
z 3 z 5 y6

xz 2 y 8 z 5
, ,
y5 z 4 y7

xy 3 y 9 z 4
, ,
z 2 z 3 y8

e1 , v2 , v3

1, y, y 2, y 3 , y 4, y 5, y 6, yz , yz , yz , yz , z

xz y 10 z 3
, ,
y6 z 2 y9

11

e1 , v1 , v2

1, y, y 2, y 3, y 4, y 5, y 6 , z, yz , yz2 , yz3 , yz4

xy 4 y 11 z 2
, z 1 , y10
z

12

e1 , e3 , v1

1, y, y 2, y 3, y 4 , y 5, y 6, y 7, y 8 , y 9, y 10, y 11

x
, y 12, yz11
y7

e2 , v9 , v11

1, x, xz, xz 2 , xz 3 , xz 4 , x2 , x2 z, z, z 2 , z 3 , z 4

5
x3
, y ,z
z 3 x2 z x

v9 , v10 , v11

1, x, z, xz, z 2 , xz 2 , z 3 , xz 3 , z 4 , xz 4 , y, yz

v7 , v8 , v9

1, x, xy, xy
, xz, xz 2 , y, yz , yz , yz 2 , z, z 2
z

e2 , v7 , v9

1, x, x2 , x3 , x4 , xz, xz 2 , x2 z, x3 z, x4 z, z, z 2

v4 , v6 , v7

1, x, xy, xz, xz 2 , xzy , x2 , x2 y, y, z, z 2, zy

v4 , v5 , v6

1, x, xy, xz, xz 2 , xzy , y, y 2, y 3 , z, z 2 , zy

3
x2 y 5
, , z
y 2 xz 2 xy 2

v2 , v3 , v4

1, x, xy, xy 2, xy 3 , xz, y, y 2, y 3, y 4 , y 5, z

x2 y 6 z 2
, ,
y 2 xz xy 3

v2 , v4 , v7

1, x, xy, xz, x2 , x2 y, x3, x3 y, x4 , x4 y, y, z

x5 y 2 z 2
, ,
z x2 x3 y

e3 , v1 , v2

1, x, xy, xy 2, xy 3, xy 4 , y, y 2, y 3, y 4 , y 5, y 6

x2 y 7
, , z
y 2 x xy 4

10

e3 , v2 , v7

1, x, xy, x2 , x2 y, x3 , x3 y, x4, x4 y, x5 , x6 , y

x7 y 2 z
, ,
y x2 x5

e2 , e3 , v7

1, x, x2 , x3 , x4 , x5 , x6 , x7 , x8 , x9 , x10 , x11

x12 , xy7 , xz5

Table 6.1. G-bricks for G =

x2 z y 3 z 4
, xz 4 , xy
y

1
(1, 7, 5)
12

x2 z y 2 z 5
, xz 5 , x
y

3
x5
, y ,z
z x2 z x3

x3 y y 2 z 3
, ,
z 2 x2 xy

TERMINAL QUOTIENT SINGULARITIES IN DIMENSION THREE

37

References
[1] T. Bridgeland, A. King, M. Reid, The McKay correspondence as an equivalence of derived categories, J. Amer. Math. Soc. 14 (2001), no. 3, 535554.
[2] A. Craw, A. Ishii, Flops of G-Hilb and equivalences of derived categories by
variation of GIT quotient, Duke Math. J. 124 (2004), no. 2, 259307.
[3] A. Craw, D. Maclagan, R. R. Thomas, Moduli of McKay quiver representations I: The coherent component, Proc. Lond. Math. Soc. (3) 95 (2007), no.
1, 179198.
[4] A. Craw, D. Maclagan, R. R. Thomas, Moduli of McKay quiver representations II: Gr
obner basis techniques, J. Algebra 316 (2007), no. 2, 514535
[5] V. Danilov, Birational geometry of three-dimensional toric varieties, Izv.
Akad. Nauk SSSR Ser. Mat. 46 (1982), no. 5, 971982, 1135; English translation: Math. USSR-Izv. 21 (1983), no. 2, 269280.
[6] S. Davis, T. Logvinenko, M. Reid, How to calculate A-Hilb Cn for
1
r (a, b, 1, . . . , 1), preprint.
[7] Y. Ito, H. Nakajima, McKay correspondence and Hilbert schemes in dimension three, Topology 39 (2000), no. 6, 11551191.
[8] S.-J. Jung, McKay Quivers and Terminal Quotient Singularities in Dimension 3, PhD thesis, University of Warwick, 2014.
[9] S.-J. Jung, Note on GIT chambers for G-constellations, in preparation.
[10] O. Kedzierski,
Cohomology of the G-Hilbert scheme for 1r (1, 1, r 1), Serdica

Math. J. 30 (2004), no. 23, 293302.

[11] O. Kedzierski,
Danilovs resolution and representations of the McKay quiver,

Tohoku Math. J. (2) 66 (2014), no. 3, 355375.

[12] A. King, Moduli of representations of finite dimensional algebras, Quart. J.
Math. Oxford Ser.(2) 45 (1994), no. 180, 515530.
[13] P. Kronheimer, The construction of ALE spaces as a hyper-K
ahler quotients,
J. Differential Geom. 29 (1989), no. 3, 665683.
[14] D. Morrison, G. Stevens, Terminal quotient singularities in dimension 3 and
4, Proc. Amer. Math Soc. 90 (1984), 1520.
[15] D. Mumford, J. Fogarty, F. Kirwan, Geometric invariant theory, Third edition, Ergebnisse der Mathematik und ihrer Grenzgebiete (2), 34. SpringerVerlag, Berlin, 1994.
[16] I. Nakamura, Hilbert schemes of abelian group orbits, J. Algebraic. Geom. 10
(2001), no.4, 757779.
[17] M. Reid, Young persons guide to canonical singularities, Algebraic geometry, Bowdoin, 1985 (Brunswick, Maine, 1985), 345414, Proc. Sympos. Pure
Math., 46, Part 1, Amer. Math. Soc., Providence, RI, 1987.
Korea Institute for Advanced Study, 85 Hoegiro, Dongdaemun-gu,
Seoul, 130-722, Republic of Korea
E-mail address: seungjo@kias.re.kr