, , April 1, 2003

April 1, 2003

1991 Mathematics Subject Classification. Primary 16D25, 16S38, 18E30.The second author is a director of research at the FWO .

<ph f="cmbx">Ideal classes of three dimensional Sklyanin algebras</ph>

Koen De Naeghel and Michel Van den Bergh

Departement WNI, Limburgs Universitair Centrum, Universitaire Campus, Building D, 3590 Diepenbeek, Belgium E-mail address, K. De Naeghel: koen.denaeghel@luc.ac.be E-mail address, M. Van den Bergh: michel.vandenbergh@luc.ac.be

Abstract. In this paper we classify graded reflexive ideals, up to isomorphism and shift, in certain three dimensional Artin-Schelter regular algebras. This classification is similar to the classification of right ideals in the first Weyl algebra, a problem that was completely settled recently. The situation we consider is substantially more complicated however.

Contents

1 Introduction

This paper is motivated by the recent developments on the classification of right ideals in the first Weyl algebra. We start by recalling the (for now) definite result in this subject, as it was formulated by Berest and Wilson.

Theorem 1.1. [8] Let

A_{1} = C 〈 x, y 〉 / (y x - x y - 1)

be the first Weyl algebra.

Put

G = Aut (A_{1})

and let

ℛ

be the set of isomorphism classes of right

A_{1}

-ideals. Then the orbits of the natural

G

-action on

ℛ

are indexed by

N

, and the orbit corresponding to

n \in N

is in natural bijection with the

n

'th Calogero-Moser space

\begin{matrix} C_{n} = {X, Y \in M_{n} (C) | rk (Y X - X Y - id) = 1} / {Gl}_{n} (C) \end{matrix}

(1.1)

where

{Gl}_{n} (C)

acts by simultaneous conjugation on

(X, Y)

The fact

ℛ / G \sim = N

has also been proved by Kouakou in his (unpublished) PhD-thesis [18] .

The first proof of Theorem 1.1 used the fact that there is a description of

ℛ

in terms of the adelic Grassmanian (due to Cannings and Holland [12] ). Using methods from integrable systems, Wilson established a relation between the adelic Grassmanian and the Calogero-Moser spaces [36] .

In [7] Berest and Wilson gave a new proof of Theorem 1.1 using noncommutative algebraic geometry [1, 34] (some of the original proofs in [7] were slightly simplified by the second author). See also [6, 16] . That an approach based on noncommutative geometry should be possible was in fact anticipated very early by Le Bruyn who in [20] already came very close to proving Theorem 1.1 .

Let us briefly indicate how the methods of noncommutative algebraic geometry may be used to prove Theorem 1.1 . We introduce the homogenized Weyl algebra

H = k 〈 x, y, z 〉 / (z x - x z, z y - y z, y x - x y - z^{2})

and then we consider

A_{1}

as the coordinate ring of an open affine part of a noncommutative space

P_{q}^{2}

, with “homogeneous coordinate ring”

H

(see below for more precise definitions). The problem of describing

ℛ

then becomes equivalent to describing certain objects on the “noncommutative projective plane”

P_{q}^{2}

. Objects on

P_{q}^{2}

have finite dimensional cohomology groups and these may be used to define moduli spaces, just as in the ordinary commutative case.

The current paper starts from the observation that there are many more noncommutative projective planes than just the one associated to the Weyl algebra (this is in fact a fairly degenerate one) [2, 3, 11] . So below we let

P_{q}^{2}

be a so-called “elliptic” quantum projective plane. By definition

P_{q}^{2}

is noncommutative projective scheme which has as homogeneous coordinate ring a graded ring

A

with generators

x, y, z

(in degree one) satisfying the relations

\begin{matrix} {\begin{matrix} a y z + b z y + c x^{2} = 0 \\ a z x + b x z + c y^{2} = 0 \\ a x y + b y x + c z^{2} = 0 \end{matrix} \end{matrix}

(1.2)

where

a, b, c

are generic scalars (see below).

Such a

P_{q}^{2}

is called an elliptic quantum plane because there is an inclusion (in a noncommutative geometry sense)

E ↪ P_{q}^{2}

where

E

is a smooth (commutative) elliptic curve.

We let

ℛ

be the set of reflexive graded

A

-ideals, considered up to isomorphism and shift of grading. We may think of the elements of

ℛ

as line bundles on

P_{q}^{2}

. In this paper we prove the following result (see Theorem 5.5.5 below).

Theorem 1.2. There exist smooth affine varieties

D_{n}

of dimension

2 n

such that

ℛ

is naturally in bijection with

\prod_{n} D_{n}

We would like to think of the

D_{n}

as elliptic Calogero-Moser spaces. We show below that

D_{0}

is a point and

D_{1}

is the complement of

E

under a natural embedding in

P^{2}

Remark 1.3. In fact

D_{n}

is connected, which we will prove in a subsequent paper [13] .

Remark 1.4. A theorem similar to Theorem 1.2 has been announced by [24] . They work in a more general setting where the associated automorphism

σ

E

may have finite order.

The reader will notice that Theorem 1.2 is weaker than Theorem 1.1 but this is probably unavoidable. Although we have a fairly succinct description of the varieties

D_{n}

(see 5.9 below) it is not as explicit as 1.1 . And very likely

D_{n}

can also not be viewed in a natural way as the orbit of a group.

Our proof of Theorem 1.2 is similar in spirit to the proof of Theorem 1.1 . However it is substantially more involved. The reason for this is that the proofs for the Weyl algebra rely heavily on the fact that

H

contains a central element in degree one (namely

z

) and the lowest central element in

A

has degree three.

We also have a result which explicitly describes the elements of

ℛ

. Recall that a line module over

A

is a graded

A

-module of the form

A / u A

with

u \in A_{1} - {0}

The following theorem can be deduced easily from Theorem 5.6.6 below.

Theorem 1.5. Let

I \in ℛ

. Then there exists an

m \in N

together with a monomorphism

I (- m) ↪ A

such that there exists a filtration of reflexive graded

A

-ideals

A = M_{0} \supset M_{1} \supset \dots \supset M_{u} = I (- m)

with the property that the

M_{i} / M_{i + 1}

are shifted line modules, up to finite length modules.

It seems plausible that this result may be used to obtain an analogue of the Cannings-Holland classification of ideals in the Weyl algebra (see [12] ) but we have not sorted out the details. We hope to come back on this in a subsequent paper.

2 Preliminaries

Throughout we work over an algebraically closed field

k

of characteristic zero. In this section we recall some basic notions of noncommutative projective geometry.

These are collected from [1, 22, 26, 27, 28, 33] . We use the following convention:

Convention 2.1. If

X y U v w (\dots)

denotes an abelian category then

x y u v w (\dots)

denotes the full subcategory of

X y U v w (\dots)

consisting of noetherian objects.

To simplify the notations we often use implicitly the following result

Lemma 2.2. Assume that

C

is a locally noetherian category and

C_{f}

is the full subcategory of

C

consisting of noetherian objects. Then the natural map

D^{b} (C_{f}) \to D_{C_{f}}^{b} (C)

is an equivalence of categories.

Proof. This follows for example from the dual of [17,1.7.11] . □

2.1 Graded algebras and modules

Let

A = \oplus_{i \in Z} A_{i}

be a

Z

-graded algebra. If

A_{i} = 0

for all

i < 0

we say that

A

is positively graded, and if in addition

A_{0} = k

we say that

A

is connected. Any graded connected Noetherian

k

-algebra

A

is locally finite, i.e.

{dim}_{k} A_{i} < \infty

, for all

i \in Z

We write

GrMod (A)

for the category of graded right

A

-modules with morphisms the

A

-module homomorphisms of degree zero. Let

M

be a graded right

A

-module.

We use the notation (for all

n \in Z

)

M_{\geq n} = \oplus_{d \geq n} M_{d}

and

M_{\leq n} = \oplus_{d \leq n} M_{d}

We say that

M

is left (resp. right) bounded if

M_{\leq n} = 0

(resp.

M_{\geq n} = 0

) for some

n \in Z

. For any integer

n

, define

M (n)

as the graded

A

-module that is equal to

M

with its original

A

action, but which is graded by

M (n)_{i} = M_{n + i}

. We refer to the functor

M \mapsto M (n)

as the

n

-th shift functor.

Since

GrMod (A)

is an abelian category with enough injective objects we may define the functors

{Ext}_{A}^{n} (M, -)

GrMod (A)

as the right derived functors of

{Hom}_{A} (M, -)

. It is convenient to write (for

n \geq 0

)

{\underset{̲}{Ext}}_{A}^{n} (M, N) : = \oplus_{d \in Z} {Ext}_{A}^{n} (M, N (d));

whence

{\underset{̲}{Ext}}_{A}^{n} (M, -)

are the right derived functors of

{\underset{̲}{Ext}}_{A}^{0} (M, -) : = {\underset{̲}{Hom}}_{A} (M, -)

, for

n \geq 1

Finally, recall that a module

M \in GrMod (A)

is reflexive if

M^{* *} = M

where

M^{*} = {\underset{̲}{Hom}}_{A} (M, A)

is the graded dual of

M

2.2 Tails

Let

A

be a Noetherian connected graded

k

-algebra. We denote by

τ

the functor that sends a graded right

A

-module to the the sum of all its finite dimensional submodules.

Denote by

Tors (A)

the full subcategory of

GrMod (A)

consisting of all modules such that

τ M = M

and write

Tails (A)

for the quotient category

GrMod (A) / Tors (A)

. We write

π : GrMod (A) \to Tails (A)

for the (exact) quotient functor. By localization theory [29]

π

has a right adjoint which we denote by

ω

. It is well-known that

π \circ ω = id

. The object

π A

Tails (A)

will be denoted by

O

and it is again easy to see that

ω = {\underset{̲}{Hom}}_{Tails} (O, -)

. The objects in

Tails (A)

will be denoted by script letters, like

ℳ

The shift functor induces an automorphism

sh : ℳ \mapsto ℳ (1)

Tails (A)

which we also call the shift functor (in analogy with algebraic geometry it should perhaps be called the “twist” functor).

When there is no possible confusion we write

Hom

instead of

{Hom}_{A}

and

{Hom}_{Tails (A)}

. The context will make clear in which category we work.

ℳ \in Tails (A)

then

Hom (ℳ, -)

is left exact, so we may define its right derived functors

{Ext}^{n} (ℳ, -)

. We also use the notation

{\underset{̲}{Ext}}^{n} (ℳ, N) : = \oplus_{d \in Z} {Ext}^{n} (ℳ, N (d))

and we set

\underset{̲}{Hom} (ℳ, N) = {\underset{̲}{Ext}}^{0} (ℳ, N)

Our Convention 2.1 fixes the meaning of

grmod (A)

tors (A)

and

tails (A)

. It is easy to see that

tors (A)

consists of the finite dimensional graded

A

-modules.

Furthermore

tails (A) = grmod (A) / tors (A)

M

is finitely generated and

N

is arbitrary we have

\begin{matrix} {\underset{̲}{Ext}}^{n} (π M, π N) \sim = {lim}_{⟶} {\underset{̲}{Ext}}_{A}^{n} (M_{\geq m}, N) . \end{matrix}

(2.1)

M

and

N

are both finitely generated, then ( 2.1 ) implies

π M \sim = π N in tails (A) \Leftrightarrow M_{\geq n} \sim = N_{\geq n} in grmod (A) for some n

explaining the word “tails”.

For

M \in GrMod (A)

there is an exact sequence (see [1] , Proposition 7.2)

\begin{matrix} 0 \to τ M \to M \to ω π M \to {lim}_{⟶} {\underset{̲}{Ext}}_{A}^{1} (A / A_{\geq n}, M) \to 0 . \end{matrix}

(2.2)

An object

ℳ \in Tails (A)

is said to be reflexive if

ℳ = π M

for some reflexive

M \in GrMod (A)

We say that

A

satisfies condition

χ

{dim}_{k} {Ext}^{j} (k, M) < \infty

for all

j

and all

M \in grmod (A)

In case

A

satisfies condition

χ

then for every

M \in grmod (A)

the cokernel of the map

M \to ω π M

in the exact sequence ( 2.2 ) is right bounded. In particular, for

M \in grmod (A)

we have

M_{\geq d} \sim = (ω π M)_{\geq d}

for some

d

Every graded quotient of a polynomial ring satisfies condition

χ

and so do most noncommutative algebras of importance. The condition is essential to get a theory for noncommutative schemes which resembles the commutative theory.

Proposition 2.2.1. [1] Let

A

be a right Noetherian connected

k

-algebra satisfying condition

χ

. Then

{Ext}^{j} (ℳ, N)

is finite dimensional for all

j

and all

ℳ, N \in tails (A)

2.3 Serre duality

It was shown in [38] that under reasonable hypotheses the category

tails (A)

satisfies a classical form of Serre duality. However we will need a stronger form of Serre duality introduced by Bondal and Kapranov in [10] . Let

A

be a

k

-linear

Ext

-finite triangulated category. By this we mean that for all

ℳ, N \in A

we have

\sum_{n} {dim}_{k} Hom (ℳ, N [n]) < \infty

. The category

A

is said to satisfy Bondal-Kapranov-Serre (BKS) duality if there is an auto-equivalence

F : A \to A

together with for all

A, B \in A

natural isomorphisms

Hom (A, B) \to Hom (B, F A)^{'}

(where

(-)^{'}

denotes the

k

-dual).

Let

C

be an abelian category. We say that

C

has finite global dimension if there exists an

n

such that

{Ext}_{C}^{i} (A, B) = 0

for all

A, B \in C

and for all

i > n

. The minimal such

n

is called the global dimension of

C

In this section we assume that

A

is a connected graded noetherian ring over a

k

(-)^{'}

we denote the functor on graded vector spaces which sends

M

\oplus_{n} M_{- n}^{*}

If we use notations which refer to the left structure of

A

then we adorn them with a superscript “

\circ

”.

We make the following additional assumptions on

A

(1) $A$ satisfies $χ$ and the functor $τ$ has finite cohomological dimension.
(2) $A$ satisfies $χ^{\circ}$ and the functor $τ^{\circ}$ has finite cohomological dimension.
(3) $tails (A)$ has finite global dimension.

Note that if

A

has finite global dimension then so does

tails (A)

by 2.1 .

Put

R = R τ (A)^{'}

. According to [32]

R

is a complex of bimodules with finitely generated cohomology on the left and on the right, which in addition has finite injective dimension, also on the left and on the right. We now have the following result

Theorem 2.3.1. (Serre duality) The category

D^{b} (tails (A))

satisfies BKS-duality with Serre functor defined by

F (π M) = π (M \otimes L R) [- 1]

This result is certainly not unexpected but as far as we know a written proof does not exist in the literature. We prove a more general version of Theorem 2.3.1 in Appendix A .

2.4 Projective schemes

We use the definition of

Proj A

(for a noncommutative algebra

A

) suggested by Artin and Zhang (see [1] ). Let

A

be a Noetherian graded

k

-algebra. We define the (polarized) projective scheme

Proj A

A

as the triple

(tails (A), O, sh)

. In what follows we shall refer to the objects of

tails (A)

(resp.

Tails (A)

) as the coherent (resp. quasicoherent) sheaves on

X = Proj A

, even when

A

is not commutative, and we shall use the notation

coh (X) : = tails (A)

Qcoh (X) : = Tails (A)

. By analogy we sometimes write

O_{X} = O = π A

The following definitions agree with the classical ones for projective schemes.

ℳ

is be a quasicoherent sheaf on

X = Proj A

, we define the cohomology groups of

ℳ

H^{n} (X, ℳ) : = {Ext}^{n} (O_{X}, ℳ) .

We refer to the graded right

A

-modules

{\underset{̲}{H}}^{n} (X, ℳ) : = \oplus_{d \in Z} H^{n} (X, ℳ (d))

as the full cohomology modules of

ℳ

Finally, we mention the cohomological dimension of

X

cd X : = max {n \in N | H^{n} (X, -) \neq 0} .

It is easy to prove that

cd X = max (0, cd τ - 1) .

2.5 The Grothendieck group, the Euler form and Hilbert series

In this subsection

A

will be a Noetherian connected graded

k

-algebra with finite global (homological) dimension. We recall some basic tools.

2.5.1 The Euler form

Let

C

be an

Ext

-finite

k

-linear abelian category of finite global dimension. We define

χ (A, B) = \sum_{i} (- 1)^{i} dim {Ext}_{C}^{i} (A, B)

for

A, B \in C

. It is clear that

χ

defines a bilinear form

χ : K_{0} (C) \times K_{0} (C) \to Z

which we call the Euler form for

C

2.5.2 Hilbert series

The Hilbert series of

M \in grmod (A)

is the Laurent power series

h_{M} (t) = \sum_{i = - \infty}^{+ \infty} ({dim}_{k} M_{i}) t^{i} \in Z ((t)) .

This definition makes sense since

A

is right Noetherian. Let

M \in grmod (A)

. Given a resolution

0 \to P^{r} \to . . . \to P^{1} \to P^{0} \to M \to 0

we have

h_{M} (t) = \sum_{i = 0}^{r} (- 1)^{i} h_{P^{i}} (t) .

Since

A

is connected, left bounded graded right

A

-modules are projective if and only if they are free hence isomorphic to a sum of shifts of

A

. So if we write

P^{i} =^{r_{i}} \oplus_{j = 0} A (- l_{i j})

we obtain the formula

\begin{matrix} q_{M} (t) = h_{M} (t) h_{A} (t)^{- 1} \end{matrix}

(2.3)

where

q_{M} (t)

is the characteristic polynomial of

M

, it is defined by

q_{M} (t) = \sum_{i = 0}^{r} (- 1)^{i} \sum_{j = 0}^{r_{i}} t^{l_{i j}} \in Z [t, t^{- 1}] .

2.5.3 The Grothendieck group $K_{0} (X)$ and the rank function $K_{0} (X) \to Z$

Set

X = Proj A

. If

ℳ

is a coherent sheaf on

X

, we denote by

[ℳ]

its image in

K_{0} (X)

The shift functor on

coh (X)

induces an automorphism of

K_{0} (X)

. Following [22] , we view

K_{0} (X)

as a

Z [t, t^{- 1}]

-module with

t

acting as the shift functor

{sh}^{- 1} : ℳ \mapsto ℳ (- 1)

. Now

K_{0} (X)

may be described in terms of the Hilbert series of

A

Theorem 2.5.1 ([22] , Theorem 2.3). Let

A

be a Noetherian connected graded

k

-algebra of finite global dimension. Set

X = Proj A

and let

q = q_{k} (t)

. Then

K_{0} (X) \sim = Z [t, t^{- 1}] / (q)

and for each

M \in grmod (A)

, the isomorphism sends

[π M]

to the characteristic polynomial

q_{M} (t)

M

θ : [π M] \mapsto \bar{q_{M} (t)}

In particular,

[O (n)]

is sent to

t^{- n}

Let us assume that

A

is a domain generated in degree one, so it has a graded division ring of fractions (graded version of Goldie's Theorem; see [23,Ch.C,Cor.I.1.7] )

Fract (A) : = {a b^{- 1} | a, b \in A homogeneous, b \neq 0} .

The degree zero component

k (X)

Fract (A)

is a division algebra which is called the function field of

X

Fract (A)

is isomorphic to a skew Laurent extention

k (X) [z, z^{- 1}; σ]

where

z

has degree one (see [23] , Chapter A, Corollary I.4.3). The rank of a finitely graded right

A

-module

M

rk M = {dim}_{k (X)} {(M \otimes_{A} Fract (A))}_{0}

This also defines an additive rank fuction on

coh (X)

and hence a homomorphism

K_{0} (X) \to Z

also denoted by “

rk

”. Obviously

rk O = 1

and

rk ℳ = rk ℳ (1)

2.6 Artin-Schelter regular algebras

Definition 2.6.1. [1] A connected graded

k

-algebra

A

is called an Artin-Schelter regular algebra of dimension

d

if it has the following properties:

(i) A has finite global dimension $d$ ;
(ii) A has polynomial growth, that is, there exists positive real numbers $c, δ$ such that ${dim}_{k} A_{n} \leq c n^{δ}$ for all positive integers $n$ ;
(iii) A is Gorenstein, meaning there is an integer $l$ such that ${\underset{̲}{Ext}}_{A}^{i} (k_{A}, A) \sim = {\begin{matrix} _{A} k (l) & if i = d, \\ 0 & otherwise. \end{matrix}$ where $l$ is called the Gorenstein parameter of $A$ .

A

is commutative, then the condition (i) already implies that

A

is isomorphic to a polynomial ring

k [x_{1}, . . . x_{n}]

with some positive grading. If in this case the grading is standard then

n = l

The Gorenstein property determines the full cohomology groups of

O

Theorem 2.6.2. [1] Let

A

be a Noetherian Artin-Schelter regular algebra of dimension

d = n + 1

, and let

X = Proj A

. Then

cd X = n

, and the full cohomology modules of

O = π A

are given by

{\underset{̲}{H}}^{i} (X, O) \sim = {\begin{matrix} A & if i = 0 \\ 0 & if i \neq 0, n \\ A^{'} (l) & if i = n \end{matrix}

Let

A

be an Artin-Schelter regular algebra as in the previous Theorem and put

X = Proj A

. It is easy to see that

A

satisfies the hypotheses for Theoreom 2.3.1 . In this case the Serre functor has a particularly simple form: indeed in [1] it is shown that

R = (R^{n + 1} τ A)^{'} \sim = A [n + 1] (- l)

as left

A

-modules and in [32] it is proved that

R τ A \sim = R τ^{\circ} A

as complexes of bimodules. Thus we also have that

R = A [n + 1] (- l)

as right

A

-modules. In other words

R = A_{φ} [n + 1] (- l)

where

φ

is some graded automorphism of

A

. The automorphism

M \mapsto M_{φ}

GrMod (A)

passes to an automorphism

Tails (A)

for which we also use the notation

(-)_{φ}

We find the the following formula for the Serre functor on

tails (A)

F ℳ = ℳ_{φ} (- l) [n]

From this we easily obtain:

Proposition 2.6.3. One has

gldim tails (A) = gldim A - 1

Proof. As above put $gldim A = n + 1$ . The inequality $gldim tails (A) \leq n$ follows directly from BKS-duality and the above discussion. The other inequality follows from Theorem 2.6.2 . □

2.7 Dimension and multiplicity

Let

A

be a noetherian Artin-Schelter regular algebra. If

0 \neq M \in grmod (A)

then the Gelfand-Kirilov dimension

GKdim M

M

[19] can be computed as the order of the pole of

h_{M} (t)

1

[3] (in particular it is an integer). If

GKdim M \leq n

then we define

e_{n} (M)

{lim}_{t \to 1} (1 - t)^{n} h_{M} (t)

. Clearly

e_{n}

is additive on short exact sequences of objects with

GKdim \leq n

. We have

e_{n} (M) \geq 0

and furthermore

e_{n} (M) = 0

if and only if

GKdim M < n

. If

e_{n} (M) > 0

then we put

e (M) = e_{n} (M)

and we call this the multiplicity of

M

. If

u = GKdim A

and if

A

is a domain generated in degree one then it is easy to see that

rk M = e_{u} (M) / e_{u} (A)

ℳ = π M \neq 0

then we put

dim ℳ = GKdim M - 1

and

e_{n} (ℳ) = e_{n + 1} (M)

e (ℳ) = e (M)

An object in

grmod (A)

tails (A)

is said to be pure if it contains no subobjects of strictly smaller dimension. It is critical if all non-trivial subobjects have the same multiplicity. It is easy to see that if

M \in grmod (A)

is pure or critical then so is

π M

, and conversely if

ℳ \in tails (A)

is pure or critical then there exists a module

M \in grmod (A)

which has the corresponding property such that

ℳ = π M

2.8 Three dimensional Artin-Schelter regular algebras

There exists a complete classification for Artin-Schelter regular algebras of dimension three [2, 3, 30, 31] :

Theorem 2.8.1. The Artin-Schelter regular algbras

A

of dimension three can be classified. They are all Noetherian domains with Hilbert series of a weighted polynomial ring

k [x, y, z]

It is known that three dimensional Artin-Schelter regular algebras have all expected nice homological properties. For example they are both left and right noetherian domains.

In this paper we restrict ourselves to Koszul three dimensional Artin-Schelter regular algebra. These have three generators and three defining relations in degree two. The minimal resolution of

k

has the form

0 \to A (- 3) \to A (- 2)^{3} \to A (- 1)^{3} \to A \to k_{A} \to 0

hence

q_{k} (t) = (1 - t)^{3}

and the Hilbert series of

A

is the same as that of the commutative polynomial algebra

k [x, y, z]

with standard grading.

Such algebras are also referred to as quantum polynomial ring in three variables.

The corresponding

Proj A

will be called a quantum projective plane and will be denoted by

P_{q}^{2}

So let

A

be a quantum polynomial ring in three variables. A linear module of dimension

d

over

A

is a cyclic

A

-module generated in degree zero with Hilbert series

(1 - t)^{- d}

. Linear modules of dimension one and two are respectively called point and line modules. The images of these objects in

coh (P_{q}^{2})

will be called point and line objects respectively. Line and point modules are classified in [2, 3] .

Line modules are of the form

A / u A = L

with

u \in A_{1}

. Hence line modules correspond naturally to lines in

P^{2}

. To classify point modules we write the relations of

A

f_{i} = \sum_{j = 1}^{3} m_{i j} x_{j}

. Set

M = (m_{i j})_{i, j}

. We introduce auxiliary (commuting) variables

x_{i}^{(p)}

(for

p \in Z

) and for a monomial

u = x_{i_{0}} \dots x_{i_{n}}

we define the multilinearization of

m

\tilde{m}

x_{i_{0}}^{(0)} \dots x_{i_{n}}^{(n)}

. We extend this operation linearly to homogeneous polynomials in the variables

(x_{i})_{i}

Let

Γ \subset P^{2} \times P^{2}

denote the locus of common zeros of the

{\tilde{f}}_{i}

. It turns out that

Γ

is the graph of an automorphism

σ

E = {pr}_{1} (Γ)

, the locus of zeros of the multihomogenized polynomial

det (\tilde{M})

. If

det (\tilde{M})

is not identically zero then

E

is a divisor of degree

3

P^{2}

. We then say that

A

is elliptic. Otherwise,

E

is all of

P^{2}

and we call

A

linear in this case.

The connection between

E

and point modules is as follows: let

P = \sum k e_{u}

be a point module where

e_{u} \in P_{u}

. Put

e_{u} x_{i} = e_{u + 1} λ_{i}^{(u)}

with

λ_{i}^{(u)} \in k

. From the fact that

e_{0} f_{i} = 0

we deduce that

((λ_{i}^{(0)})_{i}, (λ_{i}^{(1)})_{i}) \in Γ

and hence

(λ_{i}^{(0)})_{i} \in E

. This construction is reversible and defines a bijection between the closed points of

E

and the point modules over

A

. If

P_{q}

is the pointmodule corresponding to

q \in E

then we have

P_{q} (1)_{\geq 0} = P_{σ p}

Let

j : E \to P^{2}

be the inclusion and put

ℒ = j^{*} O_{P^{2}} (1)

. Associated to the geometric data

(E, σ, ℒ)

is a so-called “twisted” homogeneous coordinate ring

B = B (E, σ, ℒ)

. This is a special case of a general construction in [1] . See also [4] . Denote the auto-equivalence

σ_{*} (- \otimes_{E} ℒ)

- \otimes ℒ_{σ}

. For

ℳ \in Qcoh (X)

put

Γ_{*} (ℳ) = \oplus_{u} Γ (E, ℳ \otimes (ℒ_{σ})^{\otimes u})

and

B = Γ_{*} (O_{E})

. It is easy to see that

B

has a natural ring structure and

Γ_{*} (ℳ)

is a right

B

-module. A straightforward verification shows

Γ_{*} (O_{q}) = P_{q} .

In [3] it is shown that there is a surjective morphism

p : A \to B

of graded

k

-algebras.

Its kernel is trivial in the linear case and it is generated by a regular normalizing element

g

of degree three in the elliptic case. All point modules are

B

-modules. In other words:

g

annihilates all point modules.

By analogy with the commutative case we may say that

Proj A

contains

Proj B

as a “closed” subscheme. Though the structure of

Proj A

is somewhat obscure, that of

Proj B

is well understood.

Indeed it follows from [1, 4] that the functor

Γ_{*} : Qcoh (E) \to GrMod (B)

defines an equivalence

Qcoh (E) \sim = Tails (B)

. The inverse of this equivalence and its composition with

π : GrMod (B) \to Tails (B)

are both denoted by

\tilde{(-)}

For further properties of point modules and line modules over three dimensional quantum polynomial algebras we refer to [2, 3] .

We will frequently use the following result

Lemma 2.8.2. Assume that we are in the elliptic case. Let

M \in grmod (A)

be such that

M / M g \in tors A

. Then

GKdim M = 1

. If

σ

has infinite order then

M \in tors (A)

Proof. Multiplication induces an isomorphism $M_{n} \sim = M_{n + 3}$ for large $n$ . Hence $GKdim M = 1$ . Furthermore $(M_{g})_{0}$ is a finite dimensional representation of $(A_{g})_{0}$ .
It is is shown in [3] that if $σ$ has infinite order then $(A_{g})_{0}$ is a simple ring. In particular it has no finite dimensional representations. Thus $(M_{g})_{0} = 0$ . This implies $M \in tors (A)$ . □

In the sequel it will be useful to cast the relationship between the noncommutative graded ring

A

and the commutative scheme

E

into the language of noncommutative algebraic geometry exhibited in [26, 33] although we will use this language only in an intuitive way. Let

X = Proj A

Y = Proj B

We define a map of noncommutative schemes

i : E \to X

\begin{matrix} i^{*} π M & = (M \otimes_{A} B) \tilde{} \end{matrix}

\begin{matrix} i_{*} ℳ & = π (Γ_{*} (ℳ)_{A}) \end{matrix}

\begin{matrix}  \end{matrix}

We will call

i^{*} (π M)

the restriction of

π M

E

i_{*}

is clearly an exact functor.

For the left derived functor of

i^{*}

we have:

Lemma 2.8.3. If

M \in D^{-} (GrMod (A))

then

L i^{*} (π M) = (M \otimes L_{A} B) \tilde{}

Proof. One shows first that the objects $π F$ where $F$ is a finitely generated graded free $A$ -module are acyclic for $i^{*}$ in the sense of [14] . Then the lemma follows by replacing $M$ by a resolution of finitely generated free $A$ -modules. □

We easily obtain the following consequence:

Lemma 2.8.4. Assume that we are in the elliptic case and let

ℳ \in D^{-} (Qcoh (P_{q}^{2}))

. Then there are short exact sequences:

0 \to i^{*} H^{j} (ℳ) \to H^{j} (L i^{*} ℳ) \to L_{1} i^{*} H^{j + 1} (ℳ) \to 0

Proof. Take $M \in D^{-} (GrMod (A))$ such that $ℳ = π M$ . We may assume that $M$ is given by a right bounded complex of graded projective $A$ -modules. The lemma now follows by applying $π$ to the long exact homology sequence associate to the short exact sequence of complexes $0 \to M g \to M \to M / M g \to 0 □$

2.9 Three dimensional Sklyanin algebras

Below we are interested in Sklyanin algebras of dimension three which are elliptic Artin-Schelter regular algebras such that the corresponding elliptic curve

E

is smooth and the automorphism is a translation. More specificly, we are interested in the algebras

{Skl}_{3} (a, b, c) = k {x, y, z} / (f_{1}, f_{2}, f_{3})

where

f_{1}, f_{2}, f_{3}

are the quadratic equations

\begin{matrix} {\begin{matrix} f_{1} = a y z + b z y + c x^{2} \\ f_{2} = a z x + b x z + c y^{2} \\ f_{3} = a x y + b y x + c z^{2} \end{matrix} \end{matrix}

(2.4)

and

(a, b, c) \in P^{2} \ F

where

F = {(a, b, c) \in P^{2} | a b c = 0 or a^{3} = b^{3} = c^{3} or (3 a b c)^{3} = (a^{3} + b^{3} + c^{3})^{3}} .

The algebras

{Skl}_{3} (a, b, c)

are elliptic quantum polynomial rings. They correspond to Artin-Schelter algebras of dimension three where, in the associated geometric data,

E

is a smooth elliptic curve and

σ

is given by translation under the group law. We refer to [2] for the description of

E

and

σ

. The regular normalizing element

g

of degree three turns out to be central in this case.

Put

A = {Skl}_{3} (a, b, c)

. Combining the results in [32] with Theorem 2.3.1 we see that Serre duality for

A

takes a particularly simple form:

Theorem 2.9.1. Let

ℳ, N \in D^{b} (tails (A))

. Then there are natural isomorphisms

{Ext}^{i} (ℳ, N) \sim = {Ext}^{2 - i} (N, ℳ (- 3))^{*}

Corollary 2.9.2. Let

ℳ \in D^{b} (tails (A))

and let

P \in tails (A)

be a point object corresponding to

p \in E

. Then

{Ext}^{i} (P, ℳ) \sim = {Ext}^{2 - i} (ℳ, P^{'})^{*}

where

P^{'}

is the point object corresponding to

σ^{- 3} p

3 Cohomology of rank one sheaves on a quantum projective plane

In this section,

A

will be a quantum polynomial ring in three variables, and

P_{q}^{2} = Proj A

the associated quantum projective plane. As usual

O = π A

We say that a graded right

A

-module

M \neq 0

is torsion if

rk M = 0

M

is called torsion-free if

M

contains no torsion submodule. This is the same as saying that

M

is pure three dimensional. We use the same terminology for objects in

coh (P_{q}^{2})

The graded right ideals of

A

are, up to isomorphism, precisely the shifts of torsion-free rank one right

A

-modules.

A torsion-free rank one graded

A

-module

I

gives rise to a torsion-free coherent sheaf

ℐ = π I

P_{q}^{2}

of rank one. Conversely, every torsion-free

ℐ \in coh (P_{q}^{2})

determines a torsion-free rank one graded

A

-module

ω ℐ

Any shift

l

of a torsion-free rank one graded

A

-module

I

gives rise to a torsion-free rank one coherent sheaf

ℐ (l) = π I (l)

P_{q}^{2}

. Our first aim is to normalize this shift.

We will use the following natural basis for

K_{0} (P_{q}^{2})

Proposition 3.1. Let

P

be a point module and

S

a line module over

A

. Denote the corresponding objects in

coh (P_{q}^{2})

P

and

S

Then

{[O], [S], [P]}

is a

Z

-module basis of

K_{0} (P_{q}^{2})

, which does not depend on the particular choice of

S

and

P

, and the action of the shift functor on that basis is

\begin{matrix} \begin{matrix} [O (1)] & = [O] + [S] + [P] \\ [S (1)] & = [S] + [P] \\ [P (1)] & = [P] \end{matrix} \end{matrix}

(3.1)

Proof. It follows from Theorem 2.5.1 that the class in $K_{0} (P_{q}^{2})$ of an object $π M$ depends only on the Hilbert series of $M$ . Thus $[S]$ and $[P]$ are indeed independent of the particular choice of $S$ and $P$ .
Using a computation with Hilbert series we see that the images of $[O], [S]$ and $[P]$ under the isomorphism $θ$ of Theorem 2.5.1 $θ : K_{0} (P_{q}^{2}) \to Z [t, t^{- 1}] / (1 - t)^{3}$ are respectively $\bar{1}$ , $\bar{1 - t}$ , $\bar{(1 - t)^{2}}$ . Furthermore the shift functor corresponds to multiplication by $t^{- 1}$ . This easily yields what we want. □

From now on, we fix such a

Z

-module basis

{[O], [S], [P]}

K_{0} (P_{q}^{2})

. For any coherent sheaf

J

P_{q}^{2}

we may write

[J] = r [O] + a [S] + b [P]

where

r

is the rank of

J

It follows from ( 3.1 ) that we have

\begin{matrix} [J (l)] = r [O] + (a + l r) [S] + (\frac{1}{2} l (l + 1) r + l a + b) [P] \end{matrix}

(3.2)

for all integers

l

Proposition 3.2.

(1) Let $ℐ$ be a coherent sheaf on $P_{q}^{2}$ of rank one, and write $[ℐ] = [O] + a [S] + b [P]$ . Then there is an unique shift $c$ (namely $- a$ ) and an integer $n$ such that $[ℐ (c)] = [O] - n [P] .$ Moreover, $n = \frac{1}{2} a (a + 1) - b$ .
(2) Let $ℱ$ be a coherent sheaf on $P_{q}^{2}$ of rank zero, and write $[ℱ] = u [S] + v [P]$ . Then $u = e_{1} (ℱ)$ . If $u = 0$ , then $v = e_{0} (ℱ)$ .

Proof. For the first part, use ( 3.2 ). The uniqueness is easy to see.
For the second statement, take the image of $[ℱ] = u [S] + v [P]$ under the isomorphism $θ$ of Theorem 2.5.1 . Take $F$ such that $ℱ = π F$ . We obtain $q_{F} (t) = u q_{L} (t) + v q_{P} (t) + f (t) q_{k} (t)$ for a suitable $f (t) \in Z [t, t^{- 1}]$ . Multiplying both sides with $h_{A} (t) = q_{k} (t)^{- 1}$ yields (see ( 2.3 )) $h_{F} (t) = u h_{L} (t) + v h_{P} (t) + f (t)$ We find $e_{2} (F) = {lim}_{t \to 1} (1 - t)^{2} h_{F} (t) = u$ and if $u = 0$ then $e_{1} (F) = {lim}_{t \to 1} (1 - t) h_{F} (t) = v$ . □

We call the integer

n

appearing in Proposition 3.2 the “invariant” of

ℐ

(or of the corresponding torsion-free rank one graded

A

-module

I

such that

ℐ = π I

). Note that two torsion-free rank one graded

A

-modules

I, J

have the same invariant if and only if

{dim}_{k} I_{i} = {dim}_{k} J (d)_{i}

for

i ≫ 0

and for a fixed integer

d

We will call a torsion-free rank one coherent sheaf

ℐ

P_{q}^{2}

normalized if

[ℐ] = [O] - n [P]

for an integer

n

. We will prove later that this

n

is actually positive.

We will call a torsion-free reflexive rank one sheaf on

P_{q}^{2}

a line bundle. Our aim is to classify line bundles on

P_{q}^{2}

up to shift. By the above discussion this is equivalent to classifying normalized line bundles up to isomorphism.

It is also easy to see that through the functors

π

and

ω

classifying line bundles up to shift is equivalent to classifying reflexive torsion-free rank one graded

A

-modules, also up to shift.

We recall two elementary lemmas.

Lemma 3.3. Let

ℳ, N

be torsion-free coherent sheaves on

P_{q}^{2}

of rank one. Then every nonzero morphism in

Hom (ℳ, N)

is injective.

Proof. $ℳ$ and $N$ are critical of the same dimension. It is well-known that this implies that any map between them must be injective [3] . □

Lemma 3.4. Let

ℳ \in coh (P_{q}^{2})

. Then

ℳ

is reflexive if and only if

ℳ

is torsion-free and

{Ext}^{1} (N, ℳ) = 0

for all

N \in coh (P_{q}^{2})

of dimension zero.

Proof. Assume that $ℳ$ is a reflexive coherent sheaf on $P_{q}^{2}$ . By 2.1 we need to prove the corresponding statement for $grmod (A)$ . Thus assume that $M$ is a reflexive $A$ -module and $GKdim N \leq 1$ . Assume that there is a non-split exact sequence
$\begin{matrix} 0 \to M \to M^{'} \to N \to 0 \end{matrix}$ (3.3)
By [3,Theorem4.1] one has ${\underset{̲}{Ext}}^{1} (N, A) = 0$ . Hence we obtain $M^{' *} = M^{*}$ and thus $M = M^{* *} = M^{' * *}$ . Thus the composition of $M \to M^{'} \to M^{' * *}$ is an isomorphism, implying that the first map splits. This contradicts the non-triviality of the extention 3.3 .
For the other implication, let $ℳ \in coh (P_{q}^{2})$ be torsion-free and ${Ext}^{1} (N, ℳ) = 0$ for all $N \in coh (P_{q}^{2})$ of dimension zero. $M = ω ℳ$ is pure and $GKdim M = 3$ since $ℳ$ is pure two dimensional. By [3,Corollary4.2] there is a canonical map $μ : M \to M^{* *}$ and $ker μ$ is the maximal submodule of $M$ which has $GKdim < 3$ . Hence $μ$ is injective, and we have an exact sequence
$\begin{matrix} 0 \to M \to M^{* *} \to coker μ \to 0 \end{matrix}$ (3.4)
where $GKdim (coker μ) \leq 1$ . Applying $π$ on ( 3.4 ) yields
$\begin{matrix} 0 \to ℳ \to π M^{* *} \to N \to 0 \end{matrix}$ (3.5)
where $N = π coker μ$ . Now $N$ must be zero, otherwise $dim N = 0$ and since ${Ext}^{1} (N, ℳ) = 0$ the sequence ( 3.5 ) would split, which is impossible because $M^{* *}$ is pure three dimensional. Hence $ℳ = π M = π M^{* *}$ and thus $ℳ$ is reflexive. □

Now we can partially compute the cohomology of line bundles on

P_{q}^{2}

. This computation is similar to the one for the homogenized Weyl algebra in [20] . However the computations for the homogenized Weyl algebra rely on the existense of a central element in degree one. So they do not apply in a straightforward way to the case we consider.

Theorem 3.5. Let

ℳ

be a rank one torsion-free coherent sheaf on

P_{q}^{2}

where

[ℳ] = [O] - n [P]

. Assume that

ℳ ≇ O

. Then

(1) $H^{0} (P_{q}^{2}, ℳ (l)) = 0 for l \leq 0,$ $H^{2} (P_{q}^{2}, ℳ (l)) = 0 for l \geq - 2;$
(2) $χ (O, ℳ (l)) = \frac{1}{2} (l + 1) (l + 2) - n for all l \in Z;$
(3) ${dim}_{k} H^{1} (P_{q}^{2}, ℳ) = n - 1$ ${dim}_{k} H^{1} (P_{q}^{2}, ℳ (- 1)) = n$ ${dim}_{k} H^{1} (P_{q}^{2}, ℳ (- 2)) = n$
(4) $H^{j} (P_{q}^{2}, ℳ) = 0$ for $j \geq 3$ .

As a consequence,

n

is positive and nonzero.

ℳ

is a line bundle then we have in addition:

H^{2} (P_{q}^{2}, ℳ (l)) = 0 for l = - 3

and

{dim}_{k} H^{1} (P_{q}^{2}, ℳ (- 3)) = n - 1

Proof. That

H^{j} (P_{q}^{2}, ℳ) = 0

for

j \geq 3

is part of Theorem 2.6.2 .

To prove the rest of the current theorem we first let

l \leq 0

. Suppose

f

is a nonzero morphism in

Hom (O, ℳ (l))

. By lemma 3.3

f

is injective and from the exact sequence

\begin{matrix} 0 \to O \to ℳ (l) \to coker f \to 0 \end{matrix}

(3.6)

we get

[coker f] = l [S] + (l (l + 1) / 2 - n) [P]

. Using Proposition 3.2 gives

l \geq 0

, thus

l = 0

and

[coker f] = - n [P]

. Hence by the discussion in § 2.7 together with Proposition 3.2 we obtain

dim coker f = 0

. By lemma 3.4

{Ext}^{1} (coker f, O) = 0

This means that the exact sequence ( 3.6 ) splits hence

ℳ

is not torsion-free. A contradiction. We conclude that

Hom (O, ℳ (l)) = 0

for

l \leq 0

Second, let

l \geq - 2

. Serre duality (Theorem 2.9.1 ) yields

{Ext}^{2} (O, ℳ (l))^{*} \sim = Hom (ℳ (l + 3), O) .

g

is a nonzero morphism in

Hom (ℳ (l + 3), O)

then

g

is injective, and from the exact sequence

\begin{matrix} 0 \to ℳ (l + 3) \to O \to coker g \to 0 \end{matrix}

(3.7)

we get

[coker g] = u [S] + v [P]

where

u = - (l + 3)

and

v = n - (l + 3) (l + 4) / 2

By Proposition 3.2

u \geq 0

but

l \geq - 2

implies

u < 0

. This yields a contradiction.

Assume now

l \geq - 3

and

ℳ

reflexive. By the same reasoning as above we obtain

l = - 3

and thus the dimension of

ω coker g

is zero. By lemma 3.4 it follows that 3.7 splits. But this contradicts the fact that

O

is torsion-free.

For the second part we use Theorem 2.6.2 to obtain

χ (O, O (l)) = \frac{1}{2} (l + 1) (l + 2) for all l \in Z

and from Proposition 3.1 we deduce

\begin{matrix} [S] & = & - [O (2)] + 3 [O (1)] - 2 [O] \end{matrix}

\begin{matrix} [P] & = & [O (2)] - 2 [O (1)] + [O] \end{matrix}

Combining these results yields

χ (O, S) = 1

and

χ (O, P) = 1

. Now we use ( 3.2 ) to obtain

\begin{matrix} χ (O, ℳ (l)) & = χ (O, O) + l χ (O, S) + (\frac{1}{2} l (l + 1) - n) χ (O, P) & = \frac{1}{2} (l + 1) (l + 2) - n \end{matrix}

Finally, we combine the first two results of the theorem. If

- 2 \leq l \leq 0

(or

- 3 \leq l \leq 0

ℳ

is reflexive) the first statement gives

\begin{matrix} χ (O, ℳ (l)) & = {dim}_{k} H^{0} (P_{q}^{2}, ℳ (l)) - {dim}_{k} H^{1} (P_{q}^{2}, ℳ (l)) + {dim}_{k} H^{2} (P_{q}^{2}, ℳ (l)) & = - {dim}_{k} H^{1} (P_{q}^{2}, ℳ (l)) \end{matrix}

and comparing with the expression

χ (O, ℳ (l)) = \frac{1}{2} (l + 1) (l + 2) - n

completes the proof. □

Using Theorem 3.5 the torsion-free rank one graded

A

-modules having invariant zero are easy to determine.

Corollary 3.6. Let

ℐ

be a torsion-free coherent sheaf of rank one on

P_{q}^{2}

with invariant

n

. Then

n = 0 \Leftrightarrow ℐ \sim = O (d)

for some integer

d

Proof. If $ℐ \sim = O (d)$ then clearly $n = 0$ . Assume conversely $n = 0$ . We may assume that $ℐ$ is normalized. If $ℐ ≁ = O$ then by Theorem 3.5 $n > 0$ . Since $n = 0$ we obtain $ℐ \sim = O$ by contraposition. □

4 Restriction of coherent sheaves

In this section,

A

will be a Sklyanin algebra

{Skl}_{3} (a, b, c)

as defined in § 2.9 . We recycle the notations of sections § 2.6 -§ 2.9 . In particular the symbols

O, E, σ, ℒ, B, i

have their usual meaning.

Note that

E

is a smooth elliptic curve. We fix a grouplaw on

E

. Then

σ

is a translation by some element

ξ \in E

The dimension of objects in

grmod (B)

tails (B)

will be computed in

grmod (A)

tails (A)

. The dimension of objects in

coh (E)

is the dimension of their support.

There is a group homomorphism

K_{0} (P_{q}^{2}) \to K_{0} (E) : [ℳ] \mapsto [i^{*} ℳ] - [L_{1} i^{*} ℳ]

which as usual is also denoted by

i^{*}

Lemma 4.1. We have

\begin{matrix} i^{*} [O] & = [O_{E}] \end{matrix}

\begin{matrix} i^{*} [S] & = [O_{u}] + [O_{v}] + [O_{w}] & u, v, w arbitrary but colinear \end{matrix}

\begin{matrix} i^{*} [P] & = [O_{p}] - [O_{p^{σ^{- 3}}}] & p arbitrary \end{matrix}

\begin{matrix}  \end{matrix}

Proof. This follows easily from lemma 2.8.3 □

According to [15,ExII.6.11] we have

K_{0} (E) \sim = Z \oplus Pic (E)

. The projection

K_{0} (E) \to Z

is given by the rank and the projection

K_{0} (E) \to Pic (E)

is given the first Chern class. If

ℰ

is a vector bundle on

E

then

c_{1} (ℰ) = \land^{rk ℰ} ℰ

. We also have for

q \in E

c_{1} (O_{q}) = O_{E} (q)

There is a homomorphism

deg : Pic (E) \to Z

which assigns to a line bundle its degree. For simplicity we will denote the composition

deg \circ c_{1}

also by

deg

. If

U

is a line bundle then

deg [U] = deg U

. If

F \in coh (E)

has finite length then

deg [F] = length F

[15,Ex.6.12] . From lemma 4.1 we deduce that if

[ℳ] = a [O] + b [S] + c [P]

then

\begin{matrix} \begin{matrix} (4.1) & rk i^{*} [ℳ] = a = rk ℳ deg i^{*} [ℳ] = 3 b \end{matrix} \end{matrix}

(4.2)

Lemma 4.2.

(1) If $M \in grmod (B)$ is pure two dimensional then $\tilde{M} \in coh (E)$ is pure one dimensional.
(2) If $N \in coh (E)$ is pure one dimensional then $Γ_{*} (N)$ is pure two dimensional.

Proof. The indecomposable objects in $coh (E)$ are vector bundles and finite length objects. Using Riemann-Roch it is easy to see that if $0 \neq U \in coh (E)$ then $GKdim Γ_{*} (U) = dim U + 1$ . From this we deduce that if $V \in grmod (B)$ is not in $tors (B)$ then $GKdim V = dim \tilde{V} + 1$ . The lemma now easily follows. □

We deduce

Proposition 4.3.

(1) If $ℳ \in coh (P_{q}^{2})$ is reflexive then $i^{*} ℳ$ is a vector bundle on $E$ and $L_{j} i^{*} ℳ = 0$ for $j > 0$ .
(2) If $ℳ$ is a line bundle then so is $i^{*} ℳ$ .
(3) If $ℳ$ is a line bundle then $ℳ$ is normalized if and only if $deg i^{*} ℳ = 0$ .
(4) If $ℳ$ is a normalized line bundle with invariant $n$ then $c_{1} (i^{*} ℳ) = O ((o) - (3 n ξ))$ where “ $o$ ” is the origin for the group law.

Proof.
- (1) We have $ℳ = π M$ where $M$ is reflexive. In particular $M$ is torsion-free.
  By lemma 2.8.3 it follows that $L_{j} i^{*} ℳ = 0$ for $j > 0$ and $i^{*} ℳ = (M / M g) \tilde{}$ .
  If $M / M g$ contains a nonzero submodule $N / M g$ of GK-dimension $\leq 1$ then $N$ represents an element of ${Ext}^{1} (N / M g, M g)$ which must be zero by lemma 3.4 (or rather its proof ). Thus $N / M g \subset N \subset M$ . This is impossible since $M$ is torsion-free.
  Hence $M / M g$ is pure of GK-dimension 2. By the previous lemma it follows that $(M / M g) \tilde{}$ is a vector bundle.
- (2) This follows from 4.1 .
- (3) This follows from 4.2 .
- (4) We have $[ℳ] = [O] - n [P]$ . By lemma 4.1 we obtain $[i^{*} ℳ] = [O_{E}] - n [O_{p}] + n [O_{p^{σ^{- 3}}}]$ . Hence $c_{1} (i^{*} ℳ) = O (n (p^{σ^{- 3}}) - n (p))$ . Now $n (p^{σ^{- 3}}) - n (p)$ and $(o) - (3 n ξ)$ are both divisors of degree zero which have the same sum for the group law. Hence they are linearly equivalent by [15,IVThm4.13B] . This finishes the proof. □

Now we prove a converse of Proposition 4.3 .

Proposition 4.4. Assume that

σ

has infinite order and that

ℳ \in D^{b} (coh (P_{q}^{2}))

is such that

L i^{*} ℳ

is a vector bundle on

E

. Then

ℳ

is a reflexive object in

coh (P_{q}^{2})

Proof. It follows from lemma 2.8.4 that $i^{*} H^{j} (ℳ) = 0$ for $j \neq 0$ . Then it follows from lemma 2.8.2 that $ℳ \in coh (P_{q}^{2})$ and $L_{1} i^{*} ℳ = 0$ , using lemma 2.8.4 again.
Pick an object $M$ in $grmod (A)$ such that $π M = ℳ$ . We may assume that $M$ contains no subobject in $tors (A)$ . By lemma 2.8.3 we have $L_{1} i^{*} ℳ = ker (M (- 3) - - \to \times g M) \tilde{}$ . Thus $ker (M (- 3) - - \to \times g M) \in tors (A)$ . Since $M$ contains no subobject in $tors (A)$ it follows that $M$ is $g$ -torsion free. Furthermore by lemma 4.2 $Γ_{*} (i^{*} ℳ) = Γ_{*} ((M / M g) \tilde{})$ is pure two dimensional. If $T$ is the maximal submodule of $M / M g$ which is in $tails (A)$ then since $(M / M g) / T \subset Γ_{*} ((M / M g) \tilde{})$ we obtain that $(M / M g) / T$ is pure two dimensional.
We now claim that $M$ is pure three dimensional. Let $N$ be the maximal submodule of $M$ of dimension $\leq 2$ . Then $C = M / N$ is pure three dimensional and in particular $g$ -torsion-free. Hence we have a short exact sequence $0 \to N / N g \to M / M g \to C / C g \to 0$ By the purity of $(M / M g) / T$ it follows that $N / N g \subset T$ and hence $N / N g \in tails (A)$ .
It follows from lemma 2.8.2 than $N \in tails (A)$ and hence $N = 0$ . This shows that $M$ is pure.
Put $Q = M^{* *} / M$ . Thus we obtain an exact sequence $0 \to {Tor}_{1}^{A} (Q, B) \tilde{} \to (M \otimes_{A} B) \tilde{} \to (M^{* *} \otimes_{A} B) \tilde{} \to (Q \otimes_{A} B) \tilde{} \to 0$ By [3] we have $GKdim Q \leq 1$ . Thus we have $GKdim {Tor}_{1}^{A} (Q, B) \leq GKdim Q \leq 1$ .
So by the proof of lemma 4.2 , $dim {Tor}_{1}^{A} (Q, B) \tilde{} \leq 0$ . Since $(M \otimes_{A} B) \tilde{}$ is a vector bundle by hypotheses it contains no finite dimensional subobjects and we obtain ${Tor}_{1}^{A} (Q, B) \tilde{} = 0$ . Thus ${Tor}_{1}^{A} (Q, B) \in tors (A)$ . Thus, in high degree, multiplication by $g$ is an isomorphism on $Q$ . But then by lemma 2.8.2 $Q \in tors (A)$ . Hence $ℳ = π M = π M^{* *}$ and thus $ℳ$ is reflexive. □

5 Elliptic quantumspaces

5.1 Generalities

Let

A

be a Sklyanin algebra

{Skl}_{3} (a, b, c)

. We use again our standard notations as in the previous section.

We set

ℰ = O (2) \oplus O (1) \oplus O

and

D = {Hom}_{P_{q}^{2}} (ℰ, ℰ) = \oplus_{i, j = 0}^{2} {Hom}_{P_{q}^{2}} (O (i), O (j))

the algebra of endomorphisms of

ℰ

. We consider the left exact functor

{Hom}_{P_{q}^{2}} (ℰ, -)

which takes coherent sheaves on

P_{q}^{2}

to right

D

-modules.

{Hom}_{P_{q}^{2}} (ℰ, -)

extends to a functor

{RHom}_{P_{q}^{2}} (ℰ, -)

on bounded derived categories

\begin{matrix} {RHom}_{P_{q}^{2}} (ℰ, -) : D^{b} (coh (P_{q}^{2})) \to D^{b} (mod (D)) . \end{matrix}

(5.1)

This is done as follows:

Qcoh (P_{q}^{2})

has enough injectives and this yields a functor

{RHom}_{P_{q}^{2}} (ℰ, -) : D_{coh (P_{q}^{2})}^{b} (Qcoh (P_{q}^{2})) \to D_{mod (D)}^{b} (Mod (D))

. Now

coh (P_{q}^{2})

and

mod (D)

are noetherian abelian categories and this yields equivalences

D^{b} (coh (P_{q}^{2})) \sim = D_{coh (P_{q}^{2})}^{b} (Qcoh (P_{q}^{2}))

and

D^{b} (mod (D)) \sim = D_{mod (D)}^{b} (Mod (D))

(lemma 2.2 ). The functor 5.1 is obtained by composing with these equivalences.

In a similar way as in [9,Theorem6.2] one shows that

{RHom}_{P_{q}^{2}} (ℰ, -)

is an equivalence of derived categories. The inverse functor is given by

- \otimes L_{D} ℰ

. For a non-negative integer

i

the equivalence restricts to an equivalence between

X_{i}

and

Y_{i}

where

X_{i} \subset coh (P_{q}^{2})

is the full subcategory with objects

X_{i} = {ℳ \in coh (P_{q}^{2}) | {Ext}_{P_{q}^{2}}^{j} (ℰ, ℳ) = 0 for j \neq i}

and

Y_{i} \subset mod (D)

the full subcategory with objects

Y_{i} = {M \in mod (D) | {Tor}_{j}^{D} (M, ℰ) = 0 for j \neq i} .

The inverse equivalences between these categories are given by

{Ext}_{P_{q}^{2}}^{i} (ℰ, -)

and

{Tor}_{i}^{D} (-, ℰ)

Let

(Δ, R)

be the quiver

\begin{matrix} \begin{matrix} \overset{X_{- 2}}{⟶} & \overset{X_{- 1}}{⟶} \\ - 2 & \overset{Y_{- 2}}{⟶} & - 1 & \overset{Y_{- 1}}{⟶} & 0 \\ \overset{Z_{- 2}}{⟶} & \overset{Z_{- 1}}{⟶} \end{matrix} \end{matrix}

(5.2)

with relations

\begin{matrix} {\begin{matrix} a Y_{- 2} Z_{- 1} + b Z_{- 2} Y_{- 1} + c X_{- 2} X_{- 1} = 0 \\ a Z_{- 2} X_{- 1} + b X_{- 2} Z_{- 1} + c Y_{- 2} Y_{- 1} = 0 \\ a X_{- 2} Y_{- 1} + b Y_{- 2} X_{- 1} + c Z_{- 2} Z_{- 1} = 0 \end{matrix} \end{matrix}

(5.3)

We write

Mod (Δ)

for the category of representations of the quiver

Δ

(representations are always assumed to satisfy the relations 5.3 ). If

i = - 2, - 1, 0

then we denote by

P_{i}, S_{i}

respectively the projective representation and the simple representation corresponding to

i

It is easy to see that

D \sim = k Δ / (R)

. Since the category

Mod (Δ)

of representations of

Δ

is equivalent to the category of right

k Δ / (R)

-modules we deduce

Mod (Δ) \sim = Mod (D)

Let

ℳ \in X_{1}

and

M = {Ext}_{P_{q}^{2}}^{1} (ℰ, ℳ)

. By functoriality, multiplication by

x, y, z \in A

induces linear maps

M (λ_{- 1}) : {Ext}^{1} (O (1), ℳ) \to {Ext}^{1} (O, ℳ)

and

M (λ_{- 2}) : {Ext}^{1} (O (2), ℳ) \to {Ext}^{1} (O (1), ℳ)

(

λ = X, Y, Z

). Hence

M

is determined by the following representation of

Δ

\begin{matrix} \begin{matrix} \overset{M (X_{- 2})}{⟶} & \overset{M (X_{- 1})}{⟶} \\ H^{1} (P_{q}^{2}, ℳ (- 2)) & \overset{M (Y_{- 2})}{⟶} & H^{1} (P_{q}^{2}, ℳ (- 1)) & \overset{M (Y_{- 1})}{⟶} & H^{1} (P_{q}^{2}, ℳ) \\ \overset{M (Z_{- 2})}{⟶} & \overset{M (Z_{- 1})}{⟶} \end{matrix} \end{matrix}

For further use we note that the Euler form

χ (S_{i}, S_{j})

is given by the following matrix

\begin{matrix} (\begin{matrix} 1 & - 3 & 3 \\ 0 & 1 & - 3 \\ 0 & 0 & 1 \end{matrix}) \end{matrix}

(5.4)

where

i

refers to the columns and

j

refers to the rows.

Let

Δ^{0}

be the full subquiver of

Δ

consisting of the vertices

- 2, - 1

and let

Res : Mod (Δ) \to Mod (Δ^{0})

be the obvious restriction functor.

Res

has a left adjoint which we denote by

Ind

. If

e

is the sum of the vertices of

Δ^{0}

then

Ind = - \otimes_{k Δ^{0}} e k Δ

Note that

Res \circ Ind = id

The following was already observed by Le Bruyn in the case of the homogenized Weyl algebra.

Lemma 5.1.1. If

ℳ \neq O

is a normalized line bundle on

P_{q}^{2}

and

M = {Ext}^{1} (ℰ, ℳ)

then

M = Ind Res M

Proof. This follows from an argument by Baer [5,Corollary7.2] . For the convenience of the reader we repeat this argument.
We say that two objects $A$ , $B$ in an abelian category are orthogonal ( $A ⊥ B$ ) if $Hom (A, B) = {Ext}^{1} (A, B) = 0$ .
We have $RHom (ℰ, ℳ) = M [- 1]$ , $RHom (ℰ, O) = S_{0}$ . Thus ${Ext}^{i} (ℳ, O) = {Ext}^{i} (M [- 1], S_{0}) = {Ext}^{i + 1} (M, S_{0})$ . In particular $Hom (M, S_{0}) = 0$ and ${Ext}^{1} (M, S_{0}) = Hom (ℳ, O) = H^{2} (P_{q}^{2}, ℳ (- 3))^{*} = 0$ where we have used Serre duality and Theorem 3.5 . We conclude by lemma 5.1.2 below. □

Lemma 5.1.2. Let

M \in mod Δ

. Then

M = Ind Res M

if and only if

M ⊥ S_{0}

Proof. First assume $M = Ind Res M$ . Put $M^{0} = Res M$ and take a projective resolution $0 \to F_{1}^{0} \to F_{0}^{0} \to M^{0} \to 0$ Applying $Ind$ we get a projective resolution of $M$ of the form $0 \to S_{0}^{a} \to F_{1} \to F_{0} \to M \to 0$ for some $a \in N$ where $F_{i} = Ind F_{i}^{0}$ . The fact that $Hom (F_{1}, S_{0}) = Hom (F_{0}, S_{0}) = 0$ (by adjointness) implies $Hom (M, S_{0}) = 0$ and ${Ext}^{1} (M, S_{0}) = 0$ .
To prove the converse let $N = Ind Res M$ . By adjointness we have a map $p : N \to M$ whose kernel $K$ and cokernel $C$ are direct sums of $S_{0}$ . We have $Hom (M, S_{0}) = 0$ and hence $Hom (C, S_{0}) = 0$ . Thus $C = 0$ and $p$ is surjective.
Applying $Hom (-, S_{0})$ to the short exact sequence $0 \to K \to N \to M \to 0$ and using $Hom (N, S_{0}) = 0$ (by adjointness) yields $Hom (K, S_{0}) = 0$ and hence $K = 0$ . Thus $p$ is an isomorphism and we are done. □

Lemma 5.1.3. Let

p = (α, β, γ) \in E

and put

(α_{i}, β_{i}, γ_{i}) = p^{σ^{i}}

p

corresponds to a point module

P

A

. Put

P = π P

and

\bar{P} = ω P

(1) $H^{i} (P_{q}^{2}, P (n)) = 0$ for all $n$ and $i > 0$ . In particular $P \in X_{0}$ .
(2) $dim {\bar{P}}_{n} = 1$ for all $n$ and ${\bar{P}}_{\geq n}$ is a shifted point module for all $n$ . In particular ${\bar{P}}_{\geq 0} = P$ .
(3) $H^{0} (P_{q}^{2}, P (n)) = {\bar{P}}_{n}$ .

(4) The representation of

Δ

corresponding to

P

\begin{matrix} \begin{matrix} \overset{α_{- 2}}{⟶} & \overset{α_{- 1}}{⟶} \\ k & \overset{β_{- 2}}{⟶} & k & \overset{β_{- 1}}{⟶} & k \\ \overset{γ_{- 2}}{⟶} & \overset{γ_{- 1}}{⟶} \end{matrix} \end{matrix}

(5) Denote the representation in the previous diagram also by $p$ .
We have $p = Ind Res p$ .

Proof.
- (1) Since the $P (n)$ are all obtained from point modules, it suffices to treat the case $n = 0$ . We use lemma 2.8.3 and the discussion before that. We have $P = i_{*} O_{p}$ and hence ${Ext}^{j} (O, P) = {Ext}_{E}^{j} (L i^{*} O, O_{p}) = {Ext}_{E}^{j} (O_{E}, O_{p}) = 0$ for $j > 0$ .
- (2) This is easy to check.
- (3) Use $ω = {\underset{̲}{Hom}}_{Tails} (O, -)$ .
- (4) This follows from the previous step.
- (5) According to lemma 5.1.2 we need ${Ext}^{i} (p, S_{0}) = 0$ for $i \leq 1$ . This follows from the fact that we have ${Ext}^{i} (p, S_{0}) = {Ext}^{i} (P, O) = 0$ for $i \leq 1$ by lemma 3.4 . □

To simplify the discussion below we define

ℛ_{n}

(for

n \geq 1

) as the category in which the objects are the normalized line bundles on

P_{q}^{2}

with invariant

n

and the morphisms are the isomorphisms in

coh (P_{q}^{2})

. Thus

ℛ_{n}

is a groupoid. Note that we do not know yet if

ℛ_{n} \neq \emptyset

. This question will be addressed below.

5.2 $ℛ_{n}$ is non-empty

To simplify things we assume that

σ

has infinite order. The following result is necessary for the dimension computations in § 5.5 .

Lemma 5.2.1. The set

ℛ_{n}

is not empty.

Proof. Let $S$ be a line object on $P_{q}^{2}$ . Writing $S$ as the cokernel of a map $O (- 1) \to O$ we find by Theorem 2.6.2 that if $n \geq - 1$ then $H^{0} (P_{q}^{2}, S (n))$ has dimension $n + 1$ .
By [3] there exist at most three line objects $S^{'}$ such that $S^{'} (- 1)$ is a subobject of $S$ and furthermore these three line objects contain in turn any other object contained in $S$ .
Hence if $n \geq 0$ then we may pick an epimorphism $f : O \to S (n)$ (a generic $f$ will do). Put $ℐ = (ker f) (1)$ . Using Proposition 3.1 we find $[ℐ (- 1)] = [O] - ([S] + n [P])$ and hence $[ℐ] = [O] - n [P]$ . It is easy to see that $ℐ$ is reflexive. Thus $ℐ \in ℛ_{n}$ . □

Below we will show that

ℛ_{n}

is parametrized by an algebraic variety of dimension

2 n

. The amount of freedom in the construction exhibited in the proof of lemma 5.2.1 is less than or equal to

2 (choice of S) + n (choice of f)

parameters, hence for

n > 2

this construction can not possibly yield all elements of

ℛ_{n}

. In § 5.6 we will exhibit a related construction which works for all

n

5.3 First description of $ℛ_{n}$

Let

C_{n}

be the image of

ℛ_{n}

under the equivalence

X_{1} \sim = Y_{1}

Theorem 5.3.1. Let

n \geq 1

. There is an equivalence of categories

\begin{matrix} ℛ_{n} \binom{\binom{{Ext}_{P_{q}^{2}}^{1} (ℰ, -)}{⟶}}{\binom{⟵}{{Tor}_{1}^{D} (-, ℰ)}} C_{n} \end{matrix}

(5.5)

where

C_{n} = {M \in mod (Δ) | \underset{̲}{dim} M = (n, n, n - 1) and {Hom}_{Δ} (M, p) = 0, {Hom}_{Δ} (p, M) = 0 for all p \in E} .

Proof. First, let $ℳ$ be an object of $ℛ_{n}$ . By Proposition 4.3 we have $i^{*} ℳ \sim = N$ for a line bundle $N$ of degree zero on $E$ . Hence (for all $p \in E$ ) we have ${RHom}_{E} (L i^{*} ℳ, O_{p}) = k$ . Since ${RHom}_{E} (L i^{*} ℳ, O_{p}) \sim = {RHom}_{P_{q}^{2}} (ℳ, i_{*} O_{p})$ we obtain ${RHom}_{P_{q}^{2}} (ℳ, P) = k$ where $P = i_{*} O_{p}$ is the corresponding point object on $P_{q}^{2}$ . Writing $M = {Ext}_{P_{q}^{2}}^{1} (ℰ, ℳ)$ this means that ${RHom}_{D} (M [- 1], p) = k$ , proving that ${Hom}_{D} (M, p) = 0$ and ${Ext}_{D}^{2} (M, p) = 0$ . By BKS-duality (Theorem 2.9.1 ) we obtain ${Hom}_{D} (p^{'}, M) = 0$ for some other point $p^{'}$ determined by $p$ (and determining $p$ ). Hence $M \in C_{n}$ .
Conversely, let $M$ be an object of $C_{n}$ . Thus (using Serre duality on $P_{q}^{2}$ again) ${Hom}_{D} (M, p) = {Ext}_{D}^{2} (M, p) = 0$ for all $p \in E$ . Now $gldim D = 2$ so we may compute $dim {Ext}_{D}^{1} (M, p)$ using the Euler form 5.4 on $mod (D)$ . We obtain ${Ext}_{D}^{1} (M, p) = k$ . In other words ${RHom}_{D} (M [- 1], p) = k$ .
Put $ℳ = M [- 1] \otimes L_{D} ℰ$ . By the category equivalence between $D^{b} (coh (P_{q}^{2}))$ and $D^{b} (mod (D))$ we obtain ${RHom}_{P_{q}^{2}} (ℳ, P) = k$ , giving (by adjointness) ${RHom}_{E} (L i^{*} ℳ, O_{p}) = k$ . Since $E$ is a smooth elliptic curve it is easy to see that this implies that $L i^{*} ℳ$ is a line bundle. Hence by Propositions 4.3 and 4.4 the same is true for $ℳ$ . □

5.4 Application

Using the material in the previous sections it is now easy to parametrize the line bundles on

P_{q}^{2}

with invariant one.

Theorem 5.4.1. The representations in

C_{1}

are the representations

\begin{matrix} \begin{matrix} \overset{α}{⟶} & \overset{0}{⟶} \\ k & \overset{β}{⟶} & k & \overset{0}{⟶} & 0 \\ \overset{γ}{⟶} & \overset{0}{⟶} \end{matrix} \end{matrix}

(5.6)

for some

(α, β, γ) \in P^{2} - E

Proof. First let $F \in C_{1}$ . $F$ is given by a representation as in 5.6 . Then the condition ${Hom}_{Δ} (p, F) = 0$ for $p \in E$ implies $(α, β, γ) \notin E$ .
Conversely let $F$ be as in 5.6 with $(α, β, γ) \notin E$ . Then we immediately have ${Hom}_{Δ} (p, F) = {Hom}_{Δ} (F, p) = 0$ for $p \in E$ . □

5.5 Second description of $ℛ_{n}$

Although the category

C_{n}

has a fairly elementary description, it is not so easy to handle. In particular the analogy with the Weyl algebra case is not obvious. We will now give another description of

ℛ

which is more similar to the one used for the Weyl algebra. In particular it will follow that the isomorphism classes of objects in

ℛ_{n}

are parametrized by smooth affine varieties of dimension

2 n

In general, let

Q

be a quiver without oriented cycles and write

Q_{0}

Q_{1}

for respectively the set of vertices and edges of

Q

. Let “

\cdot

” be the standard scalar product on

Z^{Q_{0}}

(α_{v})_{v} \cdot (β_{v})_{v} = \sum_{v} α_{v} β_{v}

Let

θ \in Z^{Q_{0}}

. A representation

F

Q

is called

θ

-semistable (resp. stable) if

θ \cdot \underset{̲}{dim} F = 0

and

θ \cdot \underset{̲}{dim} N \geq 0

(resp.

> 0

) for every non-trivial subrepresentation

N

F

We have

K_{0} (mod Q) = Z^{Q_{0}}

, canonically. It is a fundamental fact [25] that

F

is semistable for some

θ

if and only there exists

G \in mod (Q)

such that

F ⊥ G

. The relation between

θ

and

\underset{̲}{dim} G

is such that the forms

- \cdot θ

and

χ (-, \underset{̲}{dim} G)

are proportional.

Fix a dimension vector

α \in Z^{Q_{0}}

and let

Rep (Q, α)

be the corresponding representation space, i.e.

Rep (Q, α) = \prod_{i \in Q_{1}} M_{α_{h (i)} \times α_{t (i)}} (k)

where the maps

h, t : Q_{1} \to Q_{0}

associate to an arrow its begin and end vertex. The isomorphism class of representations of dimension vector

α

are in one-one correspondence with the orbits of the group

Gl (α) = \prod_{v \in Q_{0}} {Gl}_{α_{v}} (k)

acting on

Rep (Q, α)

by conjugation.

Associated to

G \in mod (Q)

there is a semi-invariant function

φ_{G}

Rep (Q, α)

such that the set

\begin{matrix} ^{⊥} G = {F \in Rep (Q, α) | F ⊥ G} \end{matrix}

(5.7)

coincides with

{φ_{G} \neq 0}

. In particular 5.7 is affine.

Lemma 5.5.1. There exists

V \in mod (Δ^{0})

with

\underset{̲}{dim} V = (6, 3)

such that

(1) for all $M \in C_{n}$ we have $M^{0} ⊥ V$ where $M^{0} = Res M$ and
(2) if $p \in E$ then ${RHom}_{Δ^{0}} (Res p, V) \neq 0$ .

Proof.
- (1) Pick a degree zero line bundle $U$ on $E$ which is not of the form $O ((o) - (3 n ξ))$ for $n \in N$ (where $o$ , $ξ$ are as in Proposition 4.3 ).
  Let $ℳ \in ℛ_{n}$ . Then we have by adjointness $RHom (ℳ, i_{*} U) = {RHom}_{E} (L i^{*} ℳ, U)$ . By Proposition 4.3 we have $L i^{*} ℳ = O ((o) - (3 n ξ))$ .
  We conclude by Serre duality for $E$ that $RHom (ℳ, i_{*} U) = 0$ . Now put $M = {Ext}^{1} (ℰ, ℳ)$ and $U^{'} = RHom (ℰ, i_{*} U)$ . We obtain ${RHom}_{D} (M [- 1], U^{'}) = 0$ .
  What is $U^{'}$ ? By adjointness we have $RHom (ℰ, i_{*} U) = {RHom}_{E} (L i^{*} ℰ, U)$ .
  An easy verification shows that $L i^{*} ℰ = σ_{*}^{2} (ℒ) \otimes σ_{*} (ℒ) \oplus σ_{*} ℒ \oplus O_{E}$ . Thus by Riemann-Roch and Serre duality $U^{'} = U [- 1]$ where $\underset{̲}{dim} U = (6, 3, 0)$ .
  Put $V = Res U$ . Thus $\underset{̲}{dim} V = (6, 3)$ .
  Replacing $M$ with a projective resolution it is easy to see that ${RHom}_{Δ} (M, U) = {RHom}_{Δ^{0}} (M^{0}, V)$ . It follows that ${Hom}_{Δ^{0}} (M^{0}, V) = 0$ and ${Ext}_{Δ^{0}}^{1} (M^{0}, V) = 0$ .
- (2) Put $Q = Res p$ for $p \in E$ . Then ${RHom}_{Δ^{0}} (Q, V) = {RHom}_{Δ} (p, U) = {RHom}_{Δ} (p [- 1], U^{'}) = {RHom}_{P_{q}^{2}} (i_{*} O_{p} [- 1], i_{*} U) = {RHom}_{E} (L i^{*} i_{*} O_{p} [- 1], U)$ .
  Now $L i^{*} i_{*} O_{p} [- 1]$ is a nonzero complex whose homology has finite length. It is easy to deduce from this ${RHom}_{E} (L i^{*} i_{*} O_{p} [- 1], U) \neq 0$ . Hence we are done □

We obtain the following consequence.

Lemma 5.5.2. If

M \in C_{n}

and

M^{0} = Res M

then

M^{0}

θ

-semistable for

θ = (- 1, 1)

Proof. This is a straightforward verification. □

Lemma 5.5.3. Assume that

σ

has infinite order. Let

N

be a representation of

Δ^{0}

of dimension vector

(n, n)

n \geq 1

. If

{Hom}_{Δ^{0}} (N, Res p) = {Hom}_{Δ^{0}} (Res p, N) = 0

for all

p \in E

then

dim (Ind N)_{0} \leq n - 1

Proof. Assume the lemma is false. Thus

dim (Ind N)_{0} \geq n

. Then we may construct a surjective map

Ind N \to W

where

\underset{̲}{dim} W = (n, n, n)

. We will consider

W \otimes L_{D} ℰ

and

L i^{*} (W \otimes L_{D} ℰ)

. Note that since

E

is smooth,

L i^{*} (W \otimes L_{D} ℰ)

is the sum of its homology.

We have for

p \in E

\begin{matrix} \begin{matrix} {Ext}_{E}^{j} (L i^{*} (W \otimes L_{D} ℰ), O_{p}) & = {Ext}_{P_{q}^{2}}^{j} (W \otimes L_{D} ℰ, i_{*} O_{p}) & = {Ext}_{Δ}^{j} (W, p) \end{matrix} \end{matrix}

(5.8)

Now a simple computation shows that

χ (W, p) = 0

. Furthermore we have

{Hom}_{Δ} (W, p) \subset {Hom}_{Δ} (Ind N, p) = {Hom}_{Δ^{0}} (N, Res p) = 0

. Finally by Serre duality on

P_{q}^{2}

(see Theorem 2.9.1 ) we have

{Ext}_{Δ}^{2} (W, p) = {Hom}_{Δ} (p^{'}, W)^{*} = 0

. We conclude that also

{Ext}_{Δ}^{1} (W, p) = 0

. It follows from 5.8 that

L i^{*} (W \otimes L_{D} ℰ) = 0

. Hence by lemmas 2.8.4 and 2.8.2 we deduce

W \otimes L_{D} ℰ = 0

and hence

W = 0

which is a contradiction. □

We can now prove our main result.

Theorem 5.5.4. Assume that

σ

has infinite order. Let

V \in mod (Δ^{0})

be as in lemma 5.5.1 .

(1) The functors $Res$ and $Ind$ define inverse equivalences between $C_{n}$ and the following category $D_{n} = {F \in mod (Δ^{0}) | \underset{̲}{dim} F = (n, n), F ⊥ V, dim (Ind F)_{0} \geq n - 1}$
(2) The representations in $D_{n}$ are $θ$ -stable for $θ = (- 1, 1)$ .

Proof. Below we use often implicitly the already proved equivalence $C_{n} \sim = ℛ_{n}$ (Theorem 5.3.1 ).
Step 1. $Res (C_{n}) \subset D_{n}$ . This follows from lemmas 5.1.1 and 5.5.1 .

Step 2. Let $F \in D_{n}$ . If $F^{'} \subset F$ is such that $\underset{̲}{dim} F^{'} = (m, m)$ then ${Hom}_{Δ^{0}} (F^{'}, V) = {Ext}_{Δ^{0}}^{1} (F^{'}, V) = {Hom}_{Δ^{0}} (F / F^{'}, V) = {Ext}_{Δ^{0}}^{1} (F / F^{'}, V) = 0$ . This follows easily by using the Euler form.

Step 3. Let $F \in D_{n}$ and $p \in E$ . Then ${Hom}_{Δ^{0}} (F, Res p) = {Hom}_{Δ^{0}} (Res p, F) = 0$ . Both claims are similar so we only consider the first one. We have $F ⊥ V$ hence $F$ is $θ$ -semistable and $Res p$ is obviously stable. So if ${Hom}_{Δ^{0}} (F, Res p) \neq 0$ then there is an epimorphism $F \to Res p$ . By Step 2 we obtain ${RHom}_{Δ^{0}} (Res p, V) = 0$ . But this contradicts the choice of $V$ , finishing the argument.

Step 4. $Ind (D_{n}) \subset C_{n}$ . Let $F \in D_{n}$ . By Step 3 and lemma 5.5.3 we obtain $dim (Ind F)_{0} = n - 1$ .
It remains to show that for $p \in E$ we have ${Hom}_{Δ} (Ind F, p) = {Hom}_{Δ} (p, Ind F) = 0$ . By lemma 5.1.3 we have $p = Ind Res p$ . Thus ${Hom}_{Δ} (Ind F, p) = {Hom}_{Δ^{0}} (F, Res p) = 0$ and similarly ${Hom}_{Δ} (p, Ind F) = {Hom}_{Δ^{0}} (Res p, Res Ind F) = {Hom}_{Δ^{0}} (Res p, F) = 0$ where we have used Step 3 again.

Step 5. $Ind$ and $Res$ are inverses to each other. To prove this we only need to show $Ind \circ Res (F) = F$ for $F \in C_{n}$ . This follows from lemma 5.1.1 .

Step 6. Let $F \in D_{n}$ . Then $F$ is $θ$ -stable. Put a filtration $0 = F_{0} ⊊ F_{1} ⊊ \dots ⊊ F_{m - 1} ⊊ F_{m} = F$ on $F$ such that $F_{i} / F_{i - 1}$ is $θ$ -stable.
With the same proof as Step 3 it follows that ${Hom}_{Δ^{0}} (F_{i} / F_{i - 1}, Res p) = {Hom}_{Δ^{0}} (Res p, F_{i} / F_{i - 1}) = 0$ for $p \in E$ . Assume $\underset{̲}{dim} F_{i} / F_{i - 1} = (d_{i}, d_{i})$ . Then by lemma 5.5.3 we have $dim (Ind (F_{i} / F_{i - 1}))_{0} \leq d_{i} - 1$ . From the right exactness of $Ind$ we deduce $dim (Ind F)_{0} \leq n - m$ . Hence $m = 1$ and thus $F$ is stable. □

Below we will define some varieties. We take the classical viewpoint. So they are always reduced.

Let

V

as in lemma 5.5.1 . Let

α = (n, n)

and put

\begin{matrix} \begin{matrix} {\tilde{D}}_{n} & = {F \in Rep (Δ^{0}, α) | F \in D_{n}} & = {F \in Rep (Δ^{0}, α) | φ_{V} (F) \neq 0, dim (Ind F)_{0} \geq n - 1} . \end{matrix} \end{matrix}

(5.9)

It is clear that

{\tilde{D}}_{n}

is a closed subset of

{φ_{V} \neq 0}

so in particular

{\tilde{D}}_{n}

is affine.

Put

D_{n} = {\tilde{D}}_{n} / / Gl (α)

Theorem 5.5.5. The affine variety

D_{n}

is smooth of dimension

2 n

The isomorphism classes in

D_{n}

(and hence in

C_{n}

and

ℛ_{n}

) are in natural bijection with the points in

D_{n}

Proof. Since all representions in ${\tilde{D}}_{n}$ are stable by Theorem 5.5.4 , all $Gl (α)$ -orbits on ${\tilde{D}}_{n}$ are closed and so $D_{n}$ is really the orbit space for the $Gl (α)$ action on ${\tilde{D}}_{n}$ .
This proves that the isomorphism classes in $D_{n}$ are in natural bijection with the points in $D_{n}$ .
To prove that $D_{n}$ is smooth it suffices to prove that ${\tilde{D}}_{n}$ is smooth (this follows for example using the Luna slice theorem [21] ).
We first estimate the dimension of ${\tilde{D}}_{n}$ . We write the equations of $A$ in the usual form $M (x y z)^{t}$ . Given $n \times n$ -matrices $X$ , $Y$ , $Z$ let $M (X, Y, Z)$ be obtained from $M$ by replacing $(x, y, z)$ by $X, Y, Z$ (thus $M (X, Y, Z)$ is a $3 n \times 3 n$ -matrix). Then ${\tilde{D}}_{n}$ has the following alternative description: ${\tilde{D}}_{n} = {(X, Y, Z) \in M_{n} (k)^{3} | φ_{V} (X, Y, Z) \neq 0 and rk M (X, Y, Z) \leq 3 n - (n - 1)} .$ By § 5.2 ${\tilde{D}}_{n}$ is non-empty. The triples satifying $φ_{V} (X, Y, Z) \neq 0$ are a dense open subset of $M_{n} (k)^{3}$ and hence they represent a variety of dimension $3 n^{2}$ . Imposing that $M (X, Y, Z)$ should have corank $\geq n - 1$ represents $(n - 1)^{2}$ independent conditions. So the irreducible components of ${\tilde{D}}_{n}$ have dimension $\geq 3 n^{2} - (n - 1)^{2}$ .
Define ${\tilde{C}}_{n}$ by ${G \in Rep (Δ, \tilde{α}) | G \sim = Ind Res G, Res G \in {\tilde{D}}_{n}}$ where $\tilde{α} = (n, n, n - 1)$ (as usual we assume the points of $Rep (Δ, \tilde{α})$ to satisfy the relation imposed on $Δ$ ).
To extend $F \in {\tilde{D}}_{n}$ to a point in ${\tilde{C}}_{n}$ we need to choose a basis in $(Ind F)_{0}$ .
Thus ${\tilde{C}}_{n}$ is a principal ${Gl}_{n - 1} (k)$ fiber bundle over ${\tilde{D}}_{n}$ . In particular ${\tilde{C}}_{n}$ is smooth if and only ${\tilde{D}}_{n}$ is smooth and the irreducible components of ${\tilde{C}}_{n}$ have dimension $\geq 3 n^{2} - (n - 1)^{2} + (n - 1)^{2} = 3 n^{2}$ . Note that by the description of $C_{n}$ in Theorem 5.3.1 it follows that ${\tilde{C}}_{n}$ is an open subset of $Rep (Δ, \tilde{α})$ .
Let $x \in {\tilde{C}}_{n}$ . The stabilizer of $x$ consists of scalars thus if we put $G = Gl (\tilde{α}) / k^{*}$ then we have inclusions $Lie (G) \subset T_{x} ({\tilde{C}}_{n}) = T_{x} (Rep (Δ, \tilde{α}))$ . Voigt in [35,Ch.2,§3.4] has shown that there is a natural inclusion $T_{x} (Rep (Δ, \tilde{α})) / Lie (G) ↪ {Ext}_{Δ}^{1} (x, x)$ (Voigt actually obtains an isomorphism since he is not assuming his representation spaces to be reduced). Now $x$ corresponds to some line bundle $ℋ$ on $P_{q}^{2}$ and we have ${Ext}_{Δ}^{1} (x, x) = {Ext}^{1} (ℋ, ℋ)$ . An easy computation shows $χ (ℋ, ℋ) = χ (x, x) = 1 - 2 n$ .
We have $Hom (ℋ, ℋ) = k$ and by Serre duality ${Ext}^{2} (ℋ, ℋ) = Hom (ℋ, ℋ (- 3)) = 0$ .
Thus $dim {Ext}^{1} (ℋ, ℋ) = 2 n$ .
Hence we obtain $3 n^{2} \leq dim T_{x} ({\tilde{C}}_{n}) \leq 2 n + dim G = 2 n + 2 n^{2} + (n - 1)^{2} - 1 = 3 n^{2}$ .
Thus $dim T_{x} ({\tilde{C}}_{n}) = 3 n^{2}$ is constant and hence ${\tilde{C}}_{n}$ is smooth. We also obtain $dim {\tilde{D}}_{n} = 3 n^{2} - (n - 1)^{2}$ .
The dimension of $D_{n}$ is equal to $dim {\tilde{D}}_{n} - dim Gl (α) + 1 = 3 n^{2} - (n - 1)^{2} - 2 n^{2} + 1 = 2 n$ . This finishes the proof. □

5.6 Explicit construction of the elements in $ℛ_{n}$

For simplicity we assume throughout that

σ

has infinite order.

In the discussion below we have to compute the cohomology of a line object.

Lemma 5.6.1. Let

S = π (A / u A)

be a line object on

P_{q}^{2}

. Let

m \leq - 1

. Then

H^{1} (P_{q}^{2}, S (m)) \sim = (A / A u)_{- m - 2}^{*}

and

H^{i} (P_{q}^{2}, S (m)) = 0

for

i \neq 1

. Furthermore if

η \in A_{1}

then the induced linear map

H^{1} (P_{q}^{2}, S (m)) - \to \cdot η H^{1} (P_{q}^{2}, S (m + 1))

corresponds to

(η \cdot)^{*}

(A / A u)^{*}

Proof. That $H^{0} (P_{q}^{2}, S (m)) = 0$ follows by writing $S$ as the cokernel of a map $O (- 1) \to O$ and invoking Theorem 2.6.2 . That $H^{2} (P_{q}^{2}, S (m)) = 0$ follows by Serre duality (Theorem 2.9.1 ).
Using Theorem 2.6.2 we find $H^{1} (P_{q}^{2}, S (m)) = ker ({Ext}^{2} (O (- m), O (- 1)) - - - - \to (-, u \cdot) {Ext}^{2} (O (- m), O))$ . Using Serre duality (Theorem 2.9.1 ) this translates into $H^{1} (P_{q}^{2}, S (m)) = coker (Hom (O (- 1), O (- m - 3))^{*} - - - - - \to (u \cdot, -)^{*} Hom (O, O (- m - 3))^{*})$ . Dualizing yields that indeed $H^{1} (P_{q}^{2}, S (m)) \sim = (A / A u)_{- m - 2}^{*}$ . That $η$ acts in the indicated way follows by inspecting the appropriate commutative diagram. □

Corollary 5.6.2. Let

S = π (A / u A)

be a line object on

P_{q}^{2}

. Then

S (- 1)

corresponds to

S [- 1]

where

S

is the representation of

Δ

given by

\begin{matrix} \begin{matrix} \overset{(x \cdot)^{*}}{⟶} & \overset{}{⟶} \\ (A / A u)_{1}^{*} & \overset{(y \cdot)^{*}}{⟶} & k & \overset{}{⟶} & 0 \\ \overset{(z \cdot)^{*}}{⟶} & \overset{}{⟶} \end{matrix} \end{matrix}

(5.10)

Since line objects on

P_{q}^{2}

are of the form

π (A / u A)

they are naturally parametrized by points in

P (A_{1})

Proposition 5.6.3. Let

ℐ

be a normalized line bundle on

P_{q}^{2}

with invariant

n > 0

. Then the set of line objects

S

such that

Hom (ℐ, S (- 1)) \neq 0

is a curve of degree

n

P (A_{1})

. In particular this set is non-empty.

Proof. Let $S = π (A / u A)$ with $u = α x + β y + γ z$ . Put $I = {Ext}^{1} (ℰ, ℐ)$ , $S = {Ext}^{1} (ℰ, S (- 1))$ . Then $Hom (ℐ, S (- 1)) = {Hom}_{Δ} (I, S) = {Hom}_{Δ^{0}} (Ind I^{0}, S) = {Hom}_{Δ^{0}} (I^{0}, S^{0})$ where $I^{0}$ , $S^{0}$ are the restrictions of $I$ and $S$ to $Δ^{0}$ .
Assume that $I^{0}$ is given by matrices $X, Y, Z \in M_{n} (k)$ . Then an easy verification shows that ${Hom}_{Δ^{0}} (I^{0}, S^{0}) \neq 0$ if and only if $det (α X + β Y + γ Z) = 0$ . This is a homogeneous equation in $(α, β, γ)$ and we have to show that it is not identically zero, i.e. we have to show that there is at least one $S$ such that $Hom (ℐ, S (- 1)) = 0$ .
This follows from lemma 5.6.4 below. □

Lemma 5.6.4. Let

ℐ

be a normalized line bundle on

P_{q}^{2}

with invariant

n

and let

P

be a point object on

P_{q}^{2}

. Then, modulo zero dimensional objects, there exist at most

n

different line objects

S

such that

Hom (ℐ, S (- 1)) \neq 0

and such that

Hom (S, P) \neq 0

Proof. We use induction on $n$ . Writing $S$ as the cokernel of a map $O (- 1) \to O$ we deduce by Theorem 2.6.2 that $Hom (O, S (- 1)) = 0$ . So the case $n = 0$ is clear by Corollary 3.6 . Assume $n > 0$ . Let $(S_{i})_{i = 1, \dots, m}$ be the different line objects (modulo zero dimensional objects) satisfying $Hom (ℐ, S_{i} (- 1)) \neq 0$ and $Hom (S_{i}, P) \neq 0$ . If $m = 0$ then we are done. So assume $m > 0$ . Let $S_{i}^{'} (- 1)$ be the kernel of a non-trivial map $S_{i} \to P$ . It is proved in [3] that there is some different point object $P^{'}$ such that for all $i$ : $Hom (S_{i}, P^{'}) \neq 0$ . Let $ℐ^{'} (- 1)$ be the kernel of a non-trivial map $ℐ \to S_{1} (- 1)$ . The subobjects of line objects are shifted line objects and hence the image of $ℐ$ in $S_{1} (- 1)$ is a shifted line object. We find by 3.1 : $[ℐ^{'}] = [O] - (n - b) [P]$ with $b \geq 1$ . From this we deduce that the invariant of $ℐ^{'}$ is $\leq n - 1$ .
Since $P = i_{*} O_{p}$ for some point $p \in E$ it follows by adjointness and by Proposition 4.3 that $dim Hom (ℐ, P (- 1)) = 1$ . Hence the composition $ℐ^{'} (- 1) \to ℐ \to S_{i} (- 1)$ maps $ℐ^{'} (- 1)$ to $S_{i}^{'} (- 2)$ . We claim that for $i > 1$ this map must be nonzero. If not then there is a non-trivial map $ℐ / ℐ^{'} (- 1) \to S_{i} (- 1)$ and since $ℐ / ℐ^{'} (- 1)$ is also subobject of $S_{1} (- 1)$ it follows that $S_{1}$ and $S_{i}$ have a common subobject. But this is impossible since $S_{1}$ and $S_{i}$ are different modulo zero dimensional objects.
Hence $Hom (ℐ^{'}, S_{i}^{'} (- 1)) \neq 0$ for $i = 2, \dots, m$ . Since the $S_{i}^{'}$ are still different modulo zero dimensional objects, we obtain $m - 1 \leq n - 1$ and hence $m \leq n$ . □

The following lemma shows how to reduce the invariant of a line bundle.

Lemma 5.6.5. Let

ℐ

be a normalized line bundle on

P_{q}^{2}

with invariant

n > 0

. Then there exists a line object

S

P_{q}^{2}

such that

{Ext}^{1} (S (1), ℐ (- 1)) \neq 0

. If

J = π J

is the middle term of a corresponding non-trivial extension and

J^{* *} = π J^{* *}

then

J^{* *}

is a normalized line bundle with invariant

\leq n - 1

. Furthermore

J^{* *} / ℐ (- 1)

is a shifted line object.

Proof. Using Serre duality we have ${Ext}^{1} (S (1), ℐ (- 1)) = {Ext}^{1} (ℐ (- 1), S (- 2))^{*} = {Ext}^{1} (ℐ, S (- 1))^{*}$ . Also using Serre duality we deduce ${Ext}^{2} (ℐ, S (- 1)) = 0$ . Then a simple computation using the Euler form shows that $dim Hom (ℐ, S (- 1)) = dim {Ext}^{1} (ℐ, S (- 1))$ . Hence it follows from Proposition 5.6.3 that there exist $S$ such that ${Ext}^{1} (S (1), ℐ (- 1)) \neq 0$ .
Now let $J = π J$ be the middle term of a non-trivial extension of $ℐ (- 1)$ by $S (1)$ .
Then we have $[J] = [O] - [S] - n [P] + [S] + [P] = [O] - (n - 1) [P]$ .
We claim that $J$ is torsion-free. Assume this is not the case and let $ℱ \subset J$ be a maximal subobject of $J$ of dimension $\leq 1$ . So $ℱ \neq 0$ . Since $ℐ$ is torsion-free we have $ℱ \cap ℐ (- 1) = 0$ . So we may consider $ℱ$ as a subobject of $S (1)$ . Hence we obtain an extention
$\begin{matrix} 0 \to ℐ (- 1) \to J / ℱ \to S (1) / ℱ \to 0 \end{matrix}$ (5.11)
According to lemma 3.4 this extension is split. But this means that $S (1) / ℱ$ is a subobject of $J / ℱ$ of dimension $\leq 1$ , contradicting the maximality of $ℱ$ .
It follows from [3] that $GKdim J^{* *} / J \leq 1$ . Thus $J^{* *} / J = b [P]$ for some $b \geq 0$ by Proposition 3.2 . Hence $[J^{* *}] = [O] - (n - 1 - b) [P]$ .
Let $S^{'} = J^{* *} / ℐ (- 1)$ . Then by lemma 3.4 $S^{'}$ is pure and furthermore we have $e (S^{'}) = 1$ . It then follows easily using the methods of [3] that $S^{'}$ is a shifted line object. This finishes the proof. □

We can now prove another main result.

Theorem 5.6.6. Let

ℐ

be a normalized line bundle on

P_{q}^{2}

. Then there exists an

m \in N

together with a monomorphism

ℐ (- m) ↪ O

such that there exists a filtration of line bundles

O = ℳ_{0} \supset ℳ_{1} \supset \dots \supset ℳ_{u} = ℐ (- m)

P_{q}^{2}

with the property that the

ℳ_{i} / ℳ_{i + 1}

are shifted line objects.

Proof. This follows easily from the lemma 5.6.5 and Corollary 3.6 . □

Remark 5.6.7. There is some freedom in choosing the line objects occurring in Theorem 5.6.6 . We may assume for example that they all map to the same point object.

A Serre duality for graded rings

In this section we prove that (a generalization of ) BKS-duality holds for graded rings. For the convenience of the reader we restate some definitions so that this appendix can be read independently of the rest of this paper.

Let

A

be a

k

-linear

Ext

finite triangulated category. By this we mean that for all

ℳ, N \in A

we have

\sum_{n} {dim}_{k} Hom (ℳ, N [n]) < \infty

. The category

A

is said to satisfy Bondal-Kapranov-Serre (BKS) duality if there is an autoequivalence

F : A \to A

together with for all

A, B \in A

natural isomorphisms

Hom (A, B) \to Hom (B, F A)^{'}

(where

(-)^{'}

denotes the

k

-dual).

Let

C

be an abelian category. An object

O

D^{b} (C)

is said to have finite projective (injective) dimension if

{Ext}^{i} (O, C) = 0

(

{Ext}^{i} (C, O) = 0

) for

| i | > u

for some

u \geq 0

. The minimal such

u

we call the projective (injective) dimension of

O

In this appendix we assume that

A

is a connected graded noetherian ring over a

k

. By

(-)^{'}

we denote the functor on graded vectorspaces which sends

M

\oplus_{n} M_{- n}^{*}

. If we use notations which refer to the left structure of

A

then we adorn them with a superscript “

\circ

”.

We make the following additional assumptions on

A

(1) $A$ satisfies $χ$ and the functor $τ$ has finite cohomogical dimension.
(2) $A$ satisfies $χ^{\circ}$ and the functor $τ^{\circ}$ has finite cohomogical dimension.

These conditions imply that

A

has a balanced dualizing complex [37] given by

R = R τ (A)^{'} = R τ^{\circ} (A)^{'}

[32, 37] . Below we freely use the properties of such dualizing complexes.

We let

D (A)

be the derived category of graded right

A

-modules.

D_{f}^{b} (A)

will be the full subcategory of objects in

D^{b} (A)

with finitely generated homology.

The category

D_{f}^{b} (Tails (A))

is the full subcategory of

D^{b} (Tails (A))

consisting of complexes with homology in

tails (A)

We let

D_{f}^{b} (A)_{fpd}

(

D_{f}^{b} (A)_{fid}

) be the full category of

D_{f}^{b} (A)

consisting of objects of finite projective (injective) dimension. The categories

D_{f}^{b} (Tails (A))_{fpd}

and

D_{f}^{b} (Tails (A))_{fid}

are defined in a similar way. The fact that

τ

has finite cohomological dimension implies

π A (n) \in D_{f}^{b} (Tails (A))_{fpd}

We will denote the functors

R {\underset{̲}{Hom}}_{A} (-, R)

and

R {\underset{̲}{Hom}}_{A^{\circ}} (-, R)

D

. Since they define a duality between

D_{f}^{b} (A)

and

D_{f}^{b} (A^{\circ})

it is clear that they define a duality between

D_{f}^{b} (A)_{fid}

and

D_{f}^{b} (A^{\circ})_{fpd}

and between

D_{f}^{b} (A)_{fpd}

and

D_{f}^{b} (A^{\circ})_{fid}

It is also clear that these functors induce a duality between

D_{f}^{b} (Tails (A))

and

D_{f}^{b} (Tails (A^{opp}))

. We denote these induced functors also by

D

. Again they define a duality between

D_{f}^{b} (Tails (A))_{fid}

and

D_{f}^{b} (Tails (A^{\circ}))_{fpd}

and between

D_{f}^{b} (Tails (A))_{fpd}

and

D_{f}^{b} (Tails (A^{\circ}))_{fid}

. Recall the following:

Lemma A.1. Let

P \in D_{f}^{b} (Tails (A))_{fpd}

. Then there exists an object

P \in D_{f}^{b} (A)_{fpd}

such that

P

is a direct summand of

π P

Proof. This can be deduced from general results about compact objects in triangulated categories. For simplicity we give a direct proof based on a trick which the authors learned from Maxim Kontsevich. Take $M$ arbitrary such that $π M = P$ . Take a quasi-isomorphism $Q \to M$ where $Q$ is a right bounded complex of finitely generated projective modules. This yields a triangle: $(π Z) [a] \to σ_{\geq - a} π Q \to P$ where $Z = ker (Q_{- a} \to Q_{- a + 1})$ . This triangle corresponds to an element of ${Ext}^{a + 1} (P, π Z)$ which must be zero for large $a$ . Hence $σ_{\geq a} π Q = P \oplus (π Z) [a]$ . This proves the lemma. □

We recall the following fact.

Proposition A.2. The functors

- \otimes L_{A} R

and

R {\underset{̲}{Hom}}_{A} (R, -)

induce inverse equivalences between

D_{f}^{b} (A)_{fpd}

and

D_{f}^{b} (A)_{fid}

Proof. If $P \in D_{f}^{b} (A)_{fpd}$ then it is quasi-isomorphic to a bounded complex of finitely generated projective $A$ -modules. For such such a complex it is clear that $P \otimes_{A} R$ has finite injective dimension. There is a canonical map $P \to R \underset{̲}{Hom} (R, P \otimes_{A} R)$ which is an isomorphism for $P = A$ . By induction over triangles one shows that it is an isomorphism for all $P$ .
Conversely assume $I \in D_{f}^{b} (A)_{fid}$ . Then by duality $R \underset{̲}{Hom} (R, I) = R \underset{̲}{Hom} (D I, A)$ .
By the above discussion $D I \in D_{f}^{b} (A^{\circ})_{fpd}$ . Hence $R {\underset{̲}{Hom}}_{A} (D I, A) \in D_{f}^{b} (A)_{fpd}$ .
We also find $R {\underset{̲}{Hom}}_{A} (D I, A) \otimes_{A} R = R {\underset{̲}{Hom}}_{A} (D I, R) = I$ . □

The functor

- \otimes_{A} R

induces a functor

D^{-} (Tails (A)) \to D^{-} (Tails (A))

which we denote by

- \otimes ℛ

. Similarly the functor

{RHom}_{A} (R, -)

induces a functor

D^{+} (Tails (A)) \to D^{+} (Tails (A))

which we denote by

Rℋom (ℛ, -)

Proposition A.3. The functors

- \otimes ℛ

and

Rℋom (ℛ, -)

induces inverse equivalences between

D_{f}^{b} (Tails (A))_{fpd}

and

D_{f}^{b} (Tails (A))_{fid}

Proof. If $P \in D_{f}^{b} (Tails (A))_{fpd}$ then by lemma A.1 $P$ is direct summand of some $π P$ with $P \in D_{f}^{b} (A)_{fpd}$ .
Using the proof of the previous proposition this easily implies that $P \otimes ℛ \in D_{f}^{b} (Tails (A))_{fid}$ and $Rℋom (ℛ, P \otimes ℛ) = P$ (essentially because we may reduce to $P = π A (n)$ for some $n$ ).
Conversely assume $ℐ = π I \in D_{f}^{b} (Tails (A))_{fid}$ . Then $Rℋom (ℛ, ℐ) = π R \underset{̲}{Hom} (R, I) = π R \underset{̲}{Hom} (D I, A)$ .
We have by definition $π D I = D π I$ , and hence $π D I \in D_{f}^{b} (A)_{fpd}$ . Then it follows from lemma A.1 that $π D I$ is a direct summand of some $π Q$ with $Q \in D_{f}^{b} (Q)_{fpd}$ .
We easily deduce from this that $π R \underset{̲}{Hom} (D I, A)$ is a direct summand of $π R \underset{̲}{Hom} (Q, A)$ and hence $Rℋom (ℛ, ℐ) = π R \underset{̲}{Hom} (D I, A) \in D_{f}^{b} (Tails (A))_{fpd}$ .
The proof now continous as the proof of the previous proposition. □

Theorem A.4. (Serre duality) For all

ℳ \in D_{f}^{b} (Tails (A))_{fpd}

N \in D_{f}^{b} (Tails (A))

there are natural isomorphisms

Hom (ℳ, N) \sim = Hom (N, F ℳ)^{'}

where

\begin{matrix} F ℳ = (ℳ \otimes L ℛ) [- 1] \end{matrix}

(A.1)

Furthermore the functor

F

defines an equivalence between

D_{f}^{b} (Tails (A))_{fpd}

and

D_{f}^{b} (Tails (A))_{fid}

Proof. As in [38] our proof of Serre duality is based on the local duality formula [32, 37] . The formulation of local duality in [32] used the functor

R τ

but the same proof works for the functor

R Q

where

Q = ω \circ π

. Furthermore it is possible to throw an extra perfect complex into the bargain. If we do this we obtain canonical isomorphisms

\begin{matrix} {\underset{̲}{Hom}}_{A} (N, P \otimes_{A} (R Q A)^{'}) \sim = {\underset{̲}{Hom}}_{A} (P, R Q N)^{'} \end{matrix}

(A.2)

for

N \in D (A)

and

P \in D_{f}^{b} (A)_{fpd}

. By adjointness

{\underset{̲}{Hom}}_{A} (P, R Q N)_{0} = {Hom}_{Tails (A)} (π P, π N)

. In addition, if we apply A.2 with

N

finite dimensional then we find

{\underset{̲}{Hom}}_{A} (N, P \otimes_{A} (R Q A)^{'}) = 0

. Thus using lemma A.1 we obtain for

N \in D_{f}^{b} (A)

{\underset{̲}{Hom}}_{A} (N, P \otimes_{A} (R Q A)^{'})_{0} = {Hom}_{Tails (A)} (π N, π (P \otimes_{A} (R Q A)^{'})

Now the standard triangle for local cohomology yields

R Q A = cone (R τ A \to A)

and thus

(R Q A)^{'} = cone (A^{'} \to R) [- 1]

. Using the fact that

A^{'}

is torsion we easily obtain from this:

π (P \otimes_{A} (R Q A)^{'}) = F (π P)

where

F

is defined as in the statement of the theorem. So now we have shown

\begin{matrix} {Hom}_{Tails (A)} (π N, F (π P)) \sim = {Hom}_{Tails (A)} (π P, π N)^{'} \end{matrix}

(A.3)

Now we obtain from lemma A.1 that

ℳ

is a direct summand of a complex

π P

with

P \in D_{f}^{b} (A)_{fpd}

. Thus A.3 is true for

ℳ

and this finishes of the the first part of the theorem. The last part is Proposition A.3 . □

Corollary A.5. If

Tails (A)

has finite global dimension then

D_{f}^{b} (Tails (A))

satisfies BKS-duality.

References

M. Artin and J. J. Zhang, Noncommutative projective schemes, Adv. in Math. 109 (1994), no. 2, 228–287.
M. Artin, J. Tate, and M. Van den Bergh, Some algebras associated to automorphisms of elliptic curves, The Grothendieck Festschrift, vol. 1, Birkhäuser, 1990, pp. 33–85.
, Modules over regular algebras of dimension 3, Invent. Math. 106 (1991), 335–388.
M. Artin and M. Van den Bergh, Twisted homogeneous coordinate rings, J. Algebra 133 (1990), 249–271.
D. Baer, Tilting sheaves in representation theory of algebras, Manuscripta Math. 60 (1988), no. 3, 323–347.
V. Baranovsky, V. Ginzburg, and A. Kuznetsov, Quiver varieties and a non-commutative $P^{2}$ , Compositio Math. 134 (2002), no. 3, 283–318.
Y. Berest and G. Wilson, Ideal classes of the Weyl algebra and noncommutative projective geometry (with an appendix by Michel Van den Bergh), Int. Math. Res. Not. (2002), no. 26, 1347–1396.
, Automorphisms and ideals of the Weyl algebra, Math. Ann. 318 (2000), no. 1, 127–147.
A. I. Bondal, Helices, representions of quivers and Koszul algebras, Helices and Vector bundles, 75–95, London Math. Soc. Lecture Note Ser., 148, Cambridge Univ. Press, Cambridge, (1990).
A. I. Bondal and M. M. Kapranov, Representable functors, Serre functors, and reconstructions, Izv. Akad. Nauk SSSR Ser. Mat. 53 (1989), no. 6, 1183–1205, 1337.
A. I. Bondal and A. E. Polishchuk, Homological properties of associative algebras: the method of helices, Russian Acad. Sci. Izv. Math 42 (1994), 219–260.
R. C. Cannings and M. P. Holland, Right ideals of rings of differential operators, J. Algebra 167 (1994), no. 1, 116–141.
K. De Naeghel and M. Van den Bergh, Ideal classes of three dimensional Sklyanin algebras 2, in preparation.
R. Hartshorne, Residues and duality, Lecture notes in mathematics, vol. 20, Springer Verlag, Berlin, 1966.
, Algebraic geometry, Springer-Verlag, 1977.
A. Kapustin, A. Kuznetsov, and D. Orlov, Noncommutative instantons and twistor transform, Comm. Math. Phys. 221 (2001), no. 2, 385–432.
M. Kashiwara and P. Schapira, Sheaves on manifolds, Die Grundlehren der Mathematischen Wissenschaften, vol. 292, Springer Verlag, 1994.
K. M. Kouakou, Isomorphismes entre algèbres d'opérateurs différentiels sur les courbes algébriques affines, Ph.D. thesis, Université Claude Bernard-Lyon 1, 1994.
G. R. Krause and T. H. Lenagan, Growth of algebras and Gelfand-Kirillov dimension, Research Notes in Mathematics, vol. 116, Pitman, Boston, 1985.
L. Le Bruyn, Moduli spaces for right ideals of the Weyl algebra, J. Algebra 172 (1995), 32–48.
D. Luna, Slices étales, Bull. Soc. Math. France 33 (1973), 81–105.
I. Mori and S. P. Smith, Bézout's theorem for non-commutative projective spaces, J. Pure Appl. Algebra 157 (2001), no. 2-3, 279–299.
C. Nastasescu and F. Van Oystaeyen, Graded and filtered rings and modules, Springer, Berlin, 1979.
T. A. Nevins and J. T. Stafford, Sklyanin algebras and Hilbert schemes of points, to appear.
A. Schofield, Semi-invariants of quivers, J. London Math. Soc. (2) 43 (1991), no. 3, 385–395.
S. P. Smith, Subspaces of non-commutative spaces, Trans. Amer. Math. Soc. 354 (2002), no. 6, 2131–2171 (electronic).
J. T. Stafford and M. Van den Bergh, Noncommutative curves and noncommutative surfaces, Bull. Amer. Math. Soc. (N.S.) 38 (2001), no. 2, 171–216 (electronic).
S. T. Stafford, Noncommutative projective geometry (ICM '2002 talk).
B. Stenström, Rings of quotients, Die Grundlehren der Mathematischen Wissenschaften in Einzeldarstellungen, vol. 217, Springer Verlag, Berlin, 1975.
D. R. Stephenson, Artin-Schelter regular algebras of global dimension three, J. Algebra 183 (1996), 55–73.
, Algebras associated to elliptic curves, Trans. Amer. Math. Soc. 349 (1997), 2317–2340.
M. Van den Bergh, Existence theorems for dualizing complexes over non-commutative graded and filtered rings, J. Algebra (1997), 662–679.
, Blowing up of non-commutative smooth surfaces, Mem. Amer. Math. Soc. 154 (2001), no. 734, x+140.
F. Van Oystaeyen and L. Willaert, Grothendieck topology, coherent sheaves and Serre's theorem for schematic algebras, J. Pure Appl. Algebra 104 (1995), no. 1, 109–122.
D. Voigt, Induzierte Darstellungen in der Theorie der endlichen, algebraischen Gruppen, Springer-Verlag, Berlin, 1977, Mit einer englischen Einführung, Lecture Notes in Mathematics, Vol. 592.
G. Wilson, Collisions of Calogero-Moser particles and an adelic Grassmannian, Invent. Math. 133 (1998), no. 1, 1–41, With an appendix by I. G. Macdonald.
A. Yekutieli, Dualizing complexes over noncommutative graded algebras, J. Algebra 153 (1992), 41–84.
A. Yekutieli and J. J. Zhang, Serre duality for non-commutative projective schemes, Proc. Amer. Math. Soc. (1997), no. 125, 697–707.