Journal of Algebra 303 (2006) 358372

www.elsevier.com/locate/jalgebra
Generalized Laurent polynomial rings
as quantum projective 3-spaces
Thomas Cassidy
a,
, Peter Goetz
a
, Brad Shelton
b
a
Department of Mathematics, Bucknell University, Lewisburg, PA 17837, USA
b
Department of Mathematics, University of Oregon, Eugene, OR 97403-1222, USA
Received 20 June 2005
Available online 28 November 2005
Communicated by Michel Van den Bergh
Abstract
Given a ring R, we introduce the notion of a generalized Laurent polynomial ring over R. This
class includes the generalized Weyl algebras. We show that these rings inherit many properties from
the ground ring R. This construction is then used to create two new families of quadratic global
dimension four ArtinSchelter regular algebras. We show that in most cases the second family has
a nite point scheme and a dening automorphism of nite order. Nonetheless, a generic algebra in
this family is not nite over its center.
2005 Elsevier Inc. All rights reserved.
Keywords: ArtinSchelter regular algebras; Point modules
1. Introduction
Let K be an algebraically closed eld of characteristic different than 2. The primary
goal of this paper is to construct two new families of quadratic ArtinSchelter regular K-
algebras of global dimension 4. We recall that an N-graded K-algebra A is ArtinSchelter
*
Corresponding author.
E-mail address: [email protected] (T. Cassidy).
0021-8693/$ see front matter 2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.jalgebra.2005.10.027
T. Cassidy et al. / Journal of Algebra 303 (2006) 358372 359
regular if it has nite GK-dimension, nite global dimension d and satises the Gorenstein
condition
Ext
k
(T, A) =
k,d
T,
where T is the trivial module A/A
+
.
The classication of such algebras in various global dimensions has been a driving
problem in noncommutative algebraic geometry for years. Classication in dimension 2
is quite easy and classication in dimension 3 is contained in the celebrated papers of
Artin, Schelter, Tate and Van den Bergh [13]. Lu et al. [9] have recently classied some
Noetherian, 2-generated, global dimension 4 AS-regular algebras (having one relation of
degree 3 and one of degree 4).
It is well known that an ArtinSchelter regular algebra of global dimension 4 on four
generators must be a quadratic and Koszul algebra of Hilbert series H
A
(t ) =(1 t )
4
(cf.
[11]), very much like a polynomial ring in four variables. For this reason we refer to such
an algebra as a quantum P
3
. The classication of such algebras is very far from complete.
The purpose of this paper is to provide two new families of such algebras, examples which
do not appear to be similar to any of the examples already known (for example, graded Ore
extensions, normal extensions of 3-dimensional AS-regular algebras [6], regular Clifford
algebras and their noncommutative analogs [11], Sklyanin algebras [14] and others).
We can state our main results at once. Let x, y, d and u be indeterminates. Let F =
Kx, y and F

=Kx, y, d, u be free algebras, graded with generators in degree 1.

Theorem 1.1. Let q = ax
2
+ b(xy + yx) + cy
2
be a symmetric 2-tensor in F and let
be any graded automorphism of F. Let B

= F

/I be the graded algebra where I is the

homogeneous ideal of F

generated by the elements:

xd d(x), yd d(y), ux (x)u,
uy (y)u, du q, ud (q).
Then the following are equivalent:
(1) B

is a quantum P
3
;
(2) B

is Koszul;
(3) q is nonsingular (i.e., q is not factorable in F).
Theorem 1.2. Fix r, K

. Let C(r, ) be the algebra F

/J where J is the homogeneous

ideal generated by
xd dy, yd dx, ux ryu,
uy r
1
xu, du (xy ryx), ud +r

xy r
1
yx

.
Then C(r, ) is a quantum P
3
and is Koszul.
360 T. Cassidy et al. / Journal of Algebra 303 (2006) 358372
These theorems are proved, in part, by realizing the algebras in question via a different
construction, which we call a generalized Laurent polynomial ring. The construction is
quite simple and is closely related to the notion of down-up algebras [5] and generalized
Weyl algebras [4]. This construction and its basic properties are discussed in Section 2.
Theorems 1.1 and 1.2 are then proved in Sections 3 and 4, respectively.
We also include some information about the basic noncommutative algebraic geometry
of the algebras in Theorems 1.1 and 1.2.
Associated to any quantum P
3
, say A = F

/I , there is a standard family of graded

modules known as point modules. These are the graded, cyclic modules of Hilbert se-
ries 1/(1 t ). It is well known that this family of modules is parameterized by a scheme
called the point scheme. The point scheme is easily described as the scheme of zeroes
in P
3
P
3
of the quadratic component I
2
of the dening ideal of A. We denote this
scheme as = (A). The point schemes of the regular algebras B

are always at least

one-dimensional. The point schemes of the algebras C(r, ), on the other hand, are gener-
ically zero-dimensional.
It is also well known [12] that is the graph of an automorphism. It was long thought
that a quantum P
3
was nite over its center if and only if the associated automorphism
dening the point scheme had nite order, as is the case for ArtinSchelter regular algebras
of global dimension 3 [2]. Recently Vancliff and Stephenson [15] have found quantum
P
3
s with nite automorphisms which are not nite over their centers. For certain values
of the parameters r and the algebras C(r, ) also exhibit this phenomenon, that is, the
automorphism dening (C(r, )) has nite order while C(r, ) is not a nite module
over its center. This information is detailed in Section 4.
2. Generalized Laurent polynomial rings
Let R be a ring with identity. Fix Aut R and x a normal regular element q R.
Let q

= (q). The element q denes an automorphism Aut(R) via the formula

rq =q(r) for all r R. Let d be an indeterminate and consider the (right) skew Laurent
polynomial ring R[d, d
1
; ] (cf. [10]). To be precise, this is the ring of Laurent polyno-
mials in d with right R-coefcients and the multiplication rule rd =d(r) for r R.
Denition 2.1. The generalized Laurent polynomial ring associated to R, and q is the
subring R[d, d
1
q] of R[d, d
1
; ], generated by R, d and d
1
q = q

d
1
. We will de-
note d
1
q as u and write this ring as R[d, u; , q].
The basic properties of R[d, u; , q] that we will need are all straightforward.
Proposition 2.2. Let R, , q be as above and let S =R[d, u; , q].
(1) S is a free left and right R-module on the basis {d
i
, i 0} {u
j
, j >0}.
(2) If R is a domain then S is a domain.
(3) If R is Noetherian then S is Noetherian.
T. Cassidy et al. / Journal of Algebra 303 (2006) 358372 361
(4) Suppose f : R H is a ring homomorphism and d

, u

are in H such that: f (r)d

=
d

f ((r)), u

f (r) = f ((
1
(r)))u

for all r R, d

= f (q) and u

= f (q

).
Then f extends uniquely to a ring homomorphism

f : S H with

f (d) = d

and

f (u) =u

.
(5) If the automorphisms and commute and the left global dimension of R is n then
the left global dimension of S is either n +1 or innite.
Proof. The rst statement is obvious from the assumption that q is normal and regular.
The second follows immediately. Let T be the subring of S generated by R and d. Then
T is the right Ore extension R[d; ]. Moreover, S is generated as a ring by T and u,
and T + T u = T + uT . The third statement follows from this in a standard way, see, for
example, [10, Theorem 1.2.10].
Statement (4) follows from the denition of S and (1), once we have checked some
formulas. We need: rd = d(r) and ur = (
1
(r))u for all r R. The rst of those is
just the denition and the second is ur = d
1
qr = d
1

1
(r)q = (
1
(r))u. We also
need du =q and ud =q

, which are true by denition.

Finally, assume that the automorphisms and r commute and let =
1
=
1
. By
direct calculation, extends to an automorphism of T with (d) =d. Also by calculation,
there is a left -derivation of T dened by (r) = 0 for r R and (d) = q

q. Let
U be another indeterminate. Then the element dU q is normal and regular in the left
Ore extension T [U; , ] and we have S

= T [U; , ]/(dU q). Statement (5) follows
immediately from this observation. 2
When the element q is central in R (and is thus the identity), our denition of gener-
alized Laurent polynomial ring coincides with the denition of a generalized Weyl algebra
as given by Bavula in [4]. Notationally one just takes d = X

and u = X
+
. Generalized
Weyl algebras can always be realized as quotients of iterated Ore extensions as in part (5)
of the above proof, and from this it follows that the generalized Weyl algebras inherit some
properties of the ground ring. For example, if R is ArtinSchelter regular then graded gen-
eralized Weyl algebras over R are again ASGorenstein [7, Proposition 3.2].
However the method in part (5) of Proposition 2.2 does not apply to S in general. If
the automorphisms and do not commute then we cannot realize S as a quotient of an
iterated Ore extension.
Generalized Weyl algebras include the Noetherian subclass of the family of K-algebras
known as down-up algebras. Since we will utilize several down-up algebras later, we might
as well formalize this notationally as follows. Let r, s and be scalars. In the algebra
Kx, y, for any scalar , let H

=xy yx. Then the down-up algebra associated to r, s

and is
A(r, s, ) :=Kx, y/xH
r
sH
r
x x, H
r
y syH
r
y.
This is not the usual presentation of a down-up algebra, as given in [5], but it coincides
with that denition by taking =(r +s) and =rs.
We note that the roles of r and s in the denition of the algebra are interchangeable, due
to the formula: xH
r
sH
r
x =xH
s
rH
s
x and similarly for y.
362 T. Cassidy et al. / Journal of Algebra 303 (2006) 358372
Remark 2.3. In our notation, the algebra A(r, s, ) is the generalized Laurent polynomial
ring R[d, u; , q] where R = K[t
1
, t
2
], q = t
2
and is dened by (t
1
) = st
1
+ and
(t
2
) =rt
2
+t
1
. It is well known that for rs =0 the algebras A(r, s, 0) are ArtinSchelter
regular of global dimension 3. This observation is essentially the starting point of our
endeavor, as we will use generalized Laurent polynomial rings to form ArtinSchelter
regular algebras of global dimension 4.
Finally, we note that S, as in Proposition 2.2, is a Z-graded ring where deg(d) = 1,
deg(u) =1 and deg(r) =0 for r R. This will generally be referred to as the Z-grading,
to differentiate it from natural N-gradings that our later examples will have.
3. A family of quantum 3-spaces
Let K be an algebraically closed eld not of characteristic 2. Let A be the N-graded
K-algebra
A=Kx, y/

x(xy +yx) (xy +yx)x, y(xy +yx) (xy +yx)y

,
where the indeterminates x and y have degree 1. This is the down-up algebra A(1, 1, 0).
It has a wealth of quadratic central elements. The degree two component of the center is
the span of the elements x
2
, y
2
and xy +yx and we will therefore identify it with the space
of symmetric 2-tensors.
The graded automorphism group of A is the same as that of the graded free algebra
Kx, y, which we identify with GL
2
(K). We x an automorphism of A and let be
the corresponding element of GL
2
(K), with the formal correspondence given by (ax +
by) = (ab)

x
y

. We x a quadratic central element q and let Q be the corresponding

symmetric element of M
2
(K), i.e., q = (x y)Q

x
y

. We write q

for (q) = (x y)Q

x
y

where Q

=
t
Q.
Denition 3.1. In the notation of the previous section we let B =B(, q) =A[d, u; , q].
We note that the algebra B has both an N-grading, where the generators x, y, d and u
all have degree 1, as well as the Z-grading of the previous section, where x and y have
degree 0, d has degree 1 and u degree +1. We will refer to the rst of these as simply
the grading and the second as the Z-grading. The rst thing to note is that the Hilbert
series of B, with respect to the grading, is easily computed from the fact that B is a free
module over A with basis d
i
, i >0, and u
j
, j 0. Since the Hilbert series of A is H
A
(t ) =
((1 t )
2
(1 t
2
))
1
, we get H
B
(t ) =H
A
(t ) ((1 +t )/(1 t )) =1/(1 t )
4
.
We wish to determine the conditions under which B is ArtinSchelter regular. From the
previous paragraph we know that B has GK-dimension four. The challenge now is to show
that B has nite global dimension.
Throughout our discussion of B, there is a marked difference between the cases Q
singular and Q nonsingular. Moreover, the singular case has two distinctive subcases as
distinguished in the following lemma.
T. Cassidy et al. / Journal of Algebra 303 (2006) 358372 363
Lemma 3.2. Let A and B be as above. Assume that either Q is nonsingular or that Q
and Q

are linearly independent. Then the algebra B is isomorphic to the free algebra
Kx, y, d, u factored by the ideal generated by the six 2-tensors:
xd d(x), yd d(y), ux (x)u, uy (y)u,
du q, ud q

.
Conversely, if we assume that Q and Q

are singular and linearly dependent, then the

graded algebra B is not quadratic.
Proof. Let B

be Kx, y, d, u factored by the six quadratic 2-forms above and let A

be
the subalgebra of B

generated by x and y. The relations imply that q and q

centralize x
and y and are thus in the center of A

.
Now assume that Q is nonsingular or that Q and Q

are linearly independent. Either

of these hypotheses implies that every symmetric 2-tensor in A

is in the center of A

.
Thus there exists an algebra epimorphism : A A

with (x) = x and (y) = y. By

universality (Proposition 2.2, part (4)), extends to an algebra epimorphism : B B

with (d) = d and (u) = u. But B is also generated by x, y, d and u and satises (at
least) the six quadratic relations of B

, so is invertible.
Conversely, assume Qis singular and linearly dependent with Q

. Then we may choose

coordinates x and y so that q = x
2
and q

= x
2
for some K

. It is clear that the

six quadratic relations given above span the quadratic relations of B. Since none of the
monomials y
2
, xy or yx appear in those six 2-tensors, it is not possible for the relation
xy
2
y
2
x =0 to be generated by them. Thus B cannot be quadratic. 2
Under the hypotheses of Lemma 3.2, it is very convenient to express the relations of the
algebra B in the following shorthand:
du =q, ud =q

x
y

d =d

x
y

, u

x
y

x
y

u.
We can also write these in transposed form, for example (x y)d =d(x y)
t
.
Remark 3.3. We need one nal remark before getting to our main results. Let be the
matrix of 2-tensors =

x
y

(x y) =

x
2
xy
yx y
2

. We record the following interesting formula

for later use
QQ=qQ+det(Q)

y
2
xy
yx x
2

.
A similar formula holds for q

and Q

.
Denition 3.4. Let T be the trivial graded left B-module B/B
+
. If Q is nonsingular, let
=det(Q

)/det(Q) and if Q is singular, let be an arbitrary nonzero scalar. Let P

be
364 T. Cassidy et al. / Journal of Algebra 303 (2006) 358372
the following graded augmented sequence of free left B modules with indicated grading
shifts:
P
4
P
3
P
2
0 B(4)

4
B(3)
4

3
B(2)
6
P
1
P
0

2
B(1)
4

1
B

T 0.
Here is the usual augmentation map given by the action of B on T , and the other
maps are right multiplication by matrices with entries from B
1
as described below in block
forms:

4
=(x y d u) and
1
=(x y d u)
t
,

3
=

u
t
Q d(
1
)
t
Q

x
y

x
y

0 0 (x y)Q

u 0
(x y)Q 0 0 0 d

and

2
=

d
x
y
0
0
d
1
0
0
x
y
(x y)Q 0 d
(x y)Q

u 0

.
Proposition 3.5. The augmented sequence P

T 0 is a complex.
Proof. We simply need to observe that
i+1

i
= 0 in B for 1 i < 3. The block form of
the matrices simplies this calculation. We have

1
=

x
y

x
y

d
u
1

x
y

x
y

u
(x y)Q

x
y

du
(x y)Q

x
y

ud

which is 0 in B. Similarly
4

3
is 0 in B.
Let H =

y
2
xy
yx x
2

and recall our formulas of Remark 3.3 for QQ and Q

. By
direct calculation, the upper left 2 by 2 submatrix of
3

2
is:
T. Cassidy et al. / Journal of Algebra 303 (2006) 358372 365
ud
t
Q du

t
Q

1
+ Q

x
y

(x y)Q+Q

x
y

(x y)Q

=q

qQ+ QQ+Q

=det

H + det(Q)H =0.
Calculating the rest of
3

2
now yields

0
0
0
0
Q

u
1

x
y

x
y

x
y

x
y

((x y)
t
u u(x y)) Q 0 (q

ud)
((x y)d d(x y)
t
)Q q +du 0

which is 0 in B. 2
Theorem 3.6. The following are equivalent:
(1) the algebra B is ArtinSchelter regular of global dimension 4,
(2) the algebra B is Koszul,
(3) the symmetric matrix Q is nonsingular.
Proof. Based on the fact that the Hilbert series of B is 1/(1t )
4
, the proof that (1) implies
(2) is standard and can be found, for example, in [11].
To prove (2) implies (3), let us assume that B is Koszul, but that Q is singular. Koszul
algebras must be quadratic, so by Lemma 3.2, we must have Q and Q

linearly indepen-
dent. Based on this, we may change variables in the algebra A so that q =x
2
, Q =

1 0
0 0

,
q

=y
2
, Q

0 0
0 1

, and =

0 1

for some , K with =0.

We consider the complex P

T 0. Since the entries of

2

1
span the relations of B,
and B is a domain, the complex is automatically exact at P
0
, P
1
and P
4
. Moreover, since
the Hilbert series of B is 1/(1 t )
4
, the kernel of
2
is generated in degree 3 by Koszul,
where it has dimension 4. We examine the map
3
(with =1) given by the matrix

3
=

0 0 0 d x 0
u 0 0 0 0 y
0 0 0 y u 0
x 0 0 0 0 d

.
The rows of this matrix are clearly linearly independent, from which we conclude that
the complex is also exact at P
2
. By the Hilbert series, the complex must now also be
exact at P
3
. But this is impossible, since the linearly independent vectors (x, y, d, u),
(0, d, 0, x) and (u, 0, y, 0) in P
3
are all in the kernel of
3
. This contradiction proves that
(2) implies (3).
Finally, assume that Qis nonsingular. Then B is a quadratic domain with relations given
by Lemma 3.2. As above, we conclude that P

is exact at P
4
, P
1
and P
0
. It remains to
prove exactness at P
3
and P
2
. Since B has Hilbert series 1/(1 t )
4
, exactness at either of
366 T. Cassidy et al. / Journal of Algebra 303 (2006) 358372
these will imply exactness at the other. It sufces to prove that the kernel of
3
is contained
in the image of
4
.
Let h =(h
1
, h
2
, h
3
, h
4
) P
3
be an element of the kernel of
3
. This implies in particu-
lar that
(h
1
, h
2
)
t
uQh
4
(x, y)Q=0,
(h
1
, h
2
)

t
dQ

h
3
(x, y)Q

=0.
Since Q and Q

are nonsingular and =0, we may cancel them from these equations and
get the simpler equations
(h
1
, h
2
)
t
u =h
4
(x, y),
(h
1
, h
2
)

t
d =h
3
(x, y).
We conclude, in particular, that h
4
x Bu.
We claim that h
4
x Bu implies h
4
Bu. Recall that B has a Z-grading and let B
(n)
denote the degree n component of B in that grading. Since x and u are homogeneous
in the Z-grading, we may assume, in the proof of the claim, that h
4
is also homoge-
neous, i.e., h
4
B
(n)
for some xed n Z. Suppose rst that n > 0. Then h
4
x B
(n)
x
B
(n1)
u = Au
n
x Au
n1
u = A
n
(x)u
n
Au
n
= A
n
(x)u
n
= Au
n
x Bux. Next sup-
pose that n 0. Then h
4
x B
(n)
x B
(n1)
u =d
n
Ax d
n+1
Au =d
n
Ax d
n
Adu =
d
n
(Ax Aq). By assumption, Q is nonsingular, hence q is not a scalar multiple of x
2
and so Ax Aq =Axq =Aqx =dAux. Thus h
4
x d
n+1
Aux Bux. In either case we
get h
4
x Bux from which we can conclude h
4
Bu.
Now dene b B by h
4
= bu. Let h

= (h

1
, h

2
, h

3
, 0) = h b(x, y, d, u). Then h

is
in the kernel of
3
. But the equations above, applied to h

, immediately imply that h

= 0.
Thus h =b(x, y, d, u) and h is in the image of
4
, as required. This completes the proof of
(3) implies (1). 2
Theorem 1.1 follows immediately from Theorem 3.6 and Lemma 3.2.
Remark 3.7. Let (q, ) P
3
P
3
be the scheme representing the point modules of
B(q, ) (when q is nonsingular). (q, ) is always at least one-dimensional. Generically,
(q, ) has two one-dimensional components, each isomorphic to P
1
, and at most two
additional points.
4. A second family of quantum 3-spaces
Throughout this section we will assume r is a nonzero scalar and we take A(r) to be
the graded down-up algebra A(r, r
1
, 0). For ease of notation we will continue to use the
notation s under the assumption that rs = 1. We write h
r
and h
s
for the images of H
r
and
H
s
in A(r). These are both normal and regular elements of A(r).
T. Cassidy et al. / Journal of Algebra 303 (2006) 358372 367
Let be another nonzero scalar and let be the graded automorphism of A(r) de-
ned by (x) = y and (y) = x. We note that (h
r
) = rh
s
. We also note that the
automorphism dened by h
r
is given by (x) =sx and (y) =ry.
Denition 4.1. Given the notation above, we let C =C(r, ) =A(r)[d, u; h
r
, ].
We remark that as in the previous section, C is an N-graded algebra with all four gener-
ators having degree one. It is also a Z-graded algebra where x and y have degrees 0, d has
degree 1 and u degree 1. The Hilbert series with respect to the N-grading is 1/(1 t )
4
.
Lemma 4.2. The algebra C is isomorphic to the algebra Kx, y, d, u factored by the ideal
generated by the six quadratic elements
xd dy, yd dx, ux ryu, uy sxu,
du (xy ryx), ud +r(xy syx).
Proof. Let C

be the algebra Kx, y, d, u factored by the six given elements and let
A

be the subalgebra generated by x and y. Let h

r
= xy ryx = du in C

. Then
xh

r
= xdu = dyu = sdux = sh

r
x and similarly h

r
y = syh

r
. Thus we have an epimor-
phism : A(r) A

which by universality extends to an algebra epimorphism : C C

.
Since C is generated by x, y, d and u and satises at least the six given relations, this is
invertible. 2
Theorem 4.3. The algebra C(r, ) is ArtinSchelter regular and Koszul.
Proof. Let
C
T = C/C
+
and let P

T 0 be the augmented sequence of graded pro-

jective left C modules
0 C(4)

4
C(3)
4

3
C(2)
6

2
C(1)
4

1
C

T 0,
where the maps
i
are given by right multiplication by the following matrices of degree
one elements

4
=(x y d u),
1
=(x y d u)
t
,

3
=

1
u 0 0 rd y
1
ry
0
2
ru d 0 rx
1
x
0 0 x ry u 0

2
ry
2
x 0 0 0
2
d

and
368 T. Cassidy et al. / Journal of Algebra 303 (2006) 358372

2
=

0 d x 0
d 0 y 0
u 0 0 ry
0 u 0 sx
ry x 0 d
y rx u 0

.
The fact that P

T 0 is a complex is a straightforward check, similar to Propo-

sition 3.5, which we omit. Moreover, the entries of
2

1
span the relations of C. By
Lemma 4.2 and Proposition 2.2, C is a quadratic domain. Hence this complex is exact
at P
4
, P
1
and P
0
. Since the Hilbert series of C is 1/(1 t )
4
, to see that the complex is
exact it sufces to check it is exact at P
3
, i.e., that the kernel of
3
, is contained in the
image of
4
.
Let h = (h
1
, h
2
, h
3
, h
4
) P
3
be in the kernel of
3
. From the second column of
3
we
obtain rh
2
u h
4
x = 0 in C, i.e., h
4
x Cu. It follows exactly as in the proof of The-
orem 3.6 that h
4
Cu. The remainder of the proof of exactness is also the same as in
Theorem 3.6.
The complex P

T 0 is now seen to be a graded projective resolution of the trivial

left C-module T from which ArtinSchelter regularity and the Koszul property follow
immediately. 2
Let [X, Y, D, U] be homogeneous coordinates on P
3
. We let e
1
=[1, 0, 0, 0] P
3
, e
2
=
[0, 1, 0, 0], etc. Recall that the point scheme of any quantumP
3
is the graph of a scheme
automorphism : E E, where E =
1
( ) P
3
.
Proposition 4.4. Assume r
4
=1. Then the point scheme of C =C(r, ) is zero-dimensional
with four closed points: e
1
e
1
, e
2
e
2
, e
3
e
3
and e
4
e
4
of multiplicities 1, 1, 9 and 9,
respectively. The automorphism is the identity on closed points and has nite order if
and only if is a root of unity.
Proof. Let f = (r + s)/2. We recall that rs = 1 and we note that since r
4
= 1, f = 0.
A point p q =(X

, Y

, D

, U

) (X, Y, D, U) in P
3
P
3
is in if and only if
X

D =D

Y, Y

D =D

X, D

U =X

Y rY

X,
U

X =rY

U, U

Y =sX

U, U

D =r(X

Y sY

X).
The fact that the four points e
i
e
i
are the only solutions to this system of equations is
straightforward. We calculate the local ring over e
3
e
3
by setting D =D

= 1 and using
afne coordinates (x

, y

, u

) (x, y, u). Then x

= y, y

= x, u = x

y rxy

= y
2

rx
2
and u

= r(x

y sy

x) = r(y
2
sx
2
). Substituting these into the remain-
ing two equations yields r(y
2
sx
2
)x = rx(y
2
rx
2
) and r(y
2
sx
2
)y =
sy(y
2
rx
2
). These simplify to the equations f x
3
xy
2
= 0 and fy
3

2
x
2
y = 0.
Thus we may eliminate the variables u, u

, x

and y

and are left with the local ring

K[x, y]/(f x
3
xy
2
, fy
3

2
x
2
y) and the automorphism
#
of that ring is given by
T. Cassidy et al. / Journal of Algebra 303 (2006) 358372 369

#
(x) = y/ and
#
(y) = x. This ring has dimension 9 and the automorphism is nite if
and only if is a root of unity.
The local ring over the point e
4
e
4
is calculated similarly and turns out to be
K[x, y]/((r +s)x
3
2r
2
xy
2
, r
2
(r +s)y
3
2x
2
y) with
#
(x) =ry and
#
(y) =sx.
This ring also has dimension 9 and the automorphism has nite order if and only if is a
root of unity.
Since the total multiplicity of the scheme must be 20 (see, for example, [13]), the
remaining two points are seen to have multiplicity 1. 2
The fact that the automorphism will have nite order when is a root of unity leads
us to ask the question of whether or not the algebra C(r, ) if nite over its center in the
same circumstances. Our goal in the rest of this section is to prove that C(r, ) is not nite
over its center as long as r is not a root of unity. We must rst calculate the center of A(r)
in the following lemma. This result is well known, see, for example, [16], but we include a
short proof for the convenience of the reader.
The Z-grading that A(r) inherits from C puts all of A(r) in degree zero. However A(r)
has its own Z-grading if we give x degree 1 and y degree 1. This grading will be used in
the proof of Lemma 4.5 and Proposition 4.7.
Lemma 4.5. If r is not a root of unity then the center of A(r) is the polynomial ring
K[h
r
h
s
].
Proof. Let A(r)
(n)
denote the degree n component of the Z-grading on A(r). Since r =s,
it follows from [8] that A(r)
(0)
is the polynomial ring K[h
r
, h
s
] and A(r) is a free left or
right 4(r)
(0)
-module on the basis x
i
, i 0, and y
j
, j >0.
The center of A(r), Z(r), inherits both the N-grading and the Z-grading. Suppose
that an element of the form p(h
r
, h
s
)x
n
is in Z(r)
(n)
for some n 0 and some ho-
mogeneous polynomial p. Then p(h
r
, h
s
)x
n+1
= xp(h
r
, h
s
)x
n
= p(sh
r
, rh
s
)x
n+1
. We
conclude that p(sh
r
, rh
s
) = p(h
r
, h
s
). If the degree of p is k, this can only happen if
p(h
r
, h
s
) = (h
r
h
s
)
k/2
for some K (since r is not a root of unity). So we have
p K[h
r
h
s
], which is clearly contained in Z(r). Since p is in the center and A(r) is a
domain, we must now have x
n
in the center.
But then yx
n
=(yx)x
n1
=(
h
r
h
s
sr
)x
n1
and
yx
n
=x
n
y =x
n1

sh
r
rh
s
s r

s
n
h
r
r
n
h
s
s r

x
n1
.
We conclude that s
n
h
r
r
n
h
s
= h
r
h
s
, which can only happen if r
n
= 1. Thus n = 0.
A similar argument works for p(h
r
, h
s
)y
n
and we conclude that the center of A(r) is
contained in K[h
r
h
s
], as claimed. 2
The following statement is now clear.
Lemma 4.6. Suppose that r is not a root of unity. If is a primitive kth root of unity then
the -invariant elements of Z(A(r)) are given by
370 T. Cassidy et al. / Journal of Algebra 303 (2006) 358372
Z

A(r)

K[(h
r
h
s
)
k/2
] if k is even,
K[(h
r
h
s
)
k
] if k is odd.
If is not a root of unity then Z(A(r))

=K.
We will now calculate the center of C(r, ) when r is not a root of unity.
Proposition 4.7. Assume r is not a root of unity. If is a primitive kth root of unity then
the center of C =C(r, ) is given by
Z(C) =

K[(h
r
h
s
)
k/2
, d
2k
, u
2k
] if k is even,
K[(h
r
h
s
)
k
, d
2k
, u
2k
] if k is odd.
If is not a root of unity then the center of C is the eld K.
Proof. First assume is a primitive kth root of unity. Let

Z denote the algebra on the right-
hand side of the equation above and let Z denote the center of C. We begin by observing
that since
k
= 1,
2k
is the identity automorphism of A(r). For any a A(r) we have
ad
2k
= d
2k

2k
(a) = d
2k
a and ud
2k
= (h
r
)d
2k1
= d
2k1

2k
(h
r
) = d
2k1
h
r
= d
2k
u.
Thus d
2k
and similarly u
2k
are in the center. Hence

Z Z.
We also need some easy formulas, proved inductively:

n
(h
r
) =

n
rh
s
if n is odd,

n
(h
r
) if n is even,
and similarly with r and s interchanged.
Since Z inherits a Z-grading from C, we may consider only Z-homogeneous elements.
Let a be in A(r) and suppose that the element d
n
a is in Z for some n > 0. We will prove
that d
n
a

Z. The proof for central elements of the form u
n
a

is similar and we omit it.

For any b A(r) we have d
n
ab = bd
n
a = d
n

n
(b)a and similarly d
n+1

1
(a)b =
d
n
a(db) = (db)d
n
a = d
n+1

n
(b)a. Since C is a domain we obtain ab =
n
(b)a and

1
(a)b =
n
(b)a = ab for all b A(r). In particular a is -invariant as well as normal
and regular in A(r). We claim that n must be even. To see this, consider the Z-grading on
A(r) (as in Lemma 4.5). The automorphism
n
either preserves this grading or switches
signs in the grading, depending as n is even or odd. But
n
is dened by the formula

n
(b)a =ab and it therefore preserves the grading. So n is even.
Now we get
n
h
r
a =
n
(h
r
)a = ah
r
= h
r
(a). Canceling h
r
leaves
n
a = (a). But
the eigenvalues of on A(r) are all of the form r
j
for j Z. Since r is not a root of unity,
but is a root of unity, we conclude that
n
=1 and (a) =a. In particular k divides n.
Suppose rst that 2k divides n and thus d
n
is in the center Z. Since C is a domain, it
follows that a is in Z as well. But then a is in Z(A(r))

and so d
n
a

Z as required.
Suppose alternatively that 2k does not divide n. Then k must be even as we may write
n =kl, with l odd. Since n is even, xd
n
=
n/2
d
n
x =(1)
l
d
n
x =d
n
x and so d
n
is not in
the center. However, d
2n
a
2
=(d
n
a)
2
is in the center and hence by the arguments above a
2
is in the center. Moreover, a
2
K[(h
r
h
s
)
k/2
]. But this clearly implies a K[h
r
h
s
], and
since a is -invariant we get a in Z. This contradicts d
n
not being in the center and so the
T. Cassidy et al. / Journal of Algebra 303 (2006) 358372 371
case where 2k does not divide n does not occur. This completes the proof when is a root
of unity.
Now assume that is not a root of unity. Since we need only consider Z-homogeneous
elements of the center, let d
m
b be in Z(C) with b A(r). Then h
r
h
s
d
m
b =d
m
bh
r
h
s
and
h
r
h
s
d
m
b = d
m

m
(h
r
h
s
)b =
2m
d
m
h
r
h
s
b =
2m
d
m
bh
r
h
s
. Since is not a root of unity
m = 0. The case for elements of the form u
m
b

is identical and consequently Z(C)

Z(A(r)). However elements in Z(C) must be -invariant in order to commute with d, so
by Lemma 4.6 Z(C) =K. 2
When is a kth root of unity one can compute the formula d
2k
u
2k
= (r)
k
(h
r
h
s
)
k
.
This shows that the center of C(r, ), as given in the proposition, is itself a generalized
Laurent polynomial ring over the base ring Z(A(r))

. In particular the Hilbert series of

the center is H
Z(C)
(t ) = H
Z(A(r))
(t ) (
1+t
2k
1t
2k
). Since Z(A(r))

has GK-dimension 1, we
see that Z(C) has GK-dimension 2. In particular:
Corollary 4.8. If r is not a root of unity and is a root of unity then the algebra C(r, ) is
not a nite module over its center.
It is of course easy to prove that C(r, ) is not nite over its center any time that r is
not a root of unity. But it is only the case when is a root of unity that the automorphism
of the point scheme has nite order.
References
[1] M. Artin, J. Tate, M. Van den Bergh, Some algebras associated to automorphisms of elliptic curves, in: The
Grothendieck Festschrift, vol. I, in: Progr. Math., vol. 86, Birkhuser Boston, Boston, MA, 1990, pp. 3385,
MR MR1086882 (92e:14002).
[2] M. Artin, J. Tate, M. Van den Bergh, Modules over regular algebras of dimension 3, Invent. Math. 106 (2)
(1991) 335388, MR MR1128218 (93e:16055).
[3] M. Artin, W.F. Schelter, Graded algebras of global dimension 3, Adv. Math. 66 (2) (1987) 171216, MR
88k:16003.
[4] V. Bavula, Generalized Weyl algebras, kernel and tensor-simple algebras, their simple modules, in: Pro-
ceedings of the Sixth International Conference on Representations of Algebras, Ottawa, ON, 1992, in:
Carleton-Ottawa Math. Lecture Note Ser., vol. 14, Carleton Univ. Press, Ottawa, ON, 1992, p. 23, MR
94i:16016.
[5] G. Benkart, T. Roby, Down-up algebras, J. Algebra 209 (1) (1998) 305344, MR 2000e:06001a.
[6] T. Cassidy, Global dimension 4 extensions of ArtinSchelter regular algebras, J. Algebra 220 (1) (1999)
225254, MR MR1713429 (2000g:16047).
[7] E. Kirkman, J. Kuzmanovich, Fixed subrings of Noetherian graded regular rings, J. Algebra, 2005, in press.
[8] E. Kirkman, I.M. Musson, D.S. Passman, Noetherian down-up algebras, Proc. Amer. Math. Soc. 127 (11)
(1999) 31613167, MR 2000b:16042.
[9] D.M. Lu, J.H. Palmieri, Q.S. Wu, J.J. Zhang, Regular algebras of dimension 4 and their a

-ext-algebras,
preprint, 2004.
[10] J.C. McConnell, J.C. Robson, Noncommutative Noetherian Rings, revised ed., Grad. Stud. Math., vol. 30,
Amer. Math. Soc., Providence, RI, 2001, with the cooperation of L.W. Small, MR 2001i:16039.
[11] B. Shelton, C. Tingey, On Koszul algebras and a new construction of ArtinSchelter regular algebras, J. Al-
gebra 241 (2) (2001) 789798, MR MR1843325 (2002e:16045).
372 T. Cassidy et al. / Journal of Algebra 303 (2006) 358372
[12] B. Shelton, M. Vancliff, Embedding a quantum rank three quadric in a quantum P
3
, Comm. Algebra 27 (6)
(1999) 28772904, MR MR1687269 (2000m:16044).
[13] B. Shelton, M. Vancliff, Schemes of line modules. I, J. London Math. Soc. (2) 65 (3) (2002) 575590,
MR MR1895734 (2003i:16044).
[14] S.P. Smith, J.T. Stafford, Regularity of the four-dimensional Sklyanin algebra, Compos. Math. 83 (3) (1992)
259289, MR MR1175941 (93h:16037).
[15] D. Stephenson, M. Vancliff, Some nite quantum P
3
s that are innite modules over their centers, J. Alge-
bra, in press.
[16] K. Zhao, Centers of down-up algebras, J. Algebra 214 (1) (1999) 103121, MR MR1684900 (2000c:16041).