Ideals of cubic algebras and an
invariant ring of the Weyl
algebra
talk at the Antwerp Mini Workshop on Noncommutative Geometry Koen De Naeghel∗ University of Hasselt January 18, 2006 ∗website: http://alpha.luc.ac.be/ elucp1324
Based on joined work with N. Marconnet math.AG/0601096
Contents:
1. Introduction and main results 2. Noncommutative quadrics
3. From reflexive ideals to line bundles
4. From line bundles to quiver representations 5. Generic type A
6. The enveloping algebra 7. Hilbert series of ideals
1. Introduction and main results
• k = C field of complex numbers • First Weyl algebra
A1 = khx, yi/(xy − yx − 1) Question: describe right ideals of A1
• Cannings and Holland (1994), Wilson (1998) R(A1) = { right A1-ideals}/=∼ ←→ a
n
Cn where
Cn={(X,Y)∈ Mn(k)2| rk(YX−XY−I)≤ 1}/Gln(k) is the n-th Calogero-Moser space
Le Bruyn (1995) proposed an alternative method • Beilinson (1979) Db(coh(P2)) RHomP2(E,–) > < –⊗L∆E Db(mod(∆))
where E = O(2) ⊕ O(1) ⊕ O and ∆ is the quiver −2 X−2 > Y−2 > Z−2 > −1 X−1 > Y−1 > Z−1 > 0
with relations reflecting the relations in k[x, y, z]
Y−1X−2 = X−1Y−2 Z−1Y−2 = Y−1Z−2 X−1Z−2 = Z−1X−2
• Hulek, Barth (1977-1980): (stable) vector bundles on P2 are determined by certain (stable) representations of the quiver ∆
• The picture survives when we replace
– k[x, y, z] by homogenized Weyl algebra
H = khx, y, zi .(zx − xz, zy − yz, yx − xy − z2) a noncommutative analogue of k[x, y, z]
– P2 by Proj H (sense of Artin and Zhang) – relations of ∆ by the ones induced by H
• Equality A1 = H[z−1]0 induces bijection between sets
R(A1) = { right A1-ideals}/=∼
and
R(H) = { reflexive graded right H-ideals }/=,sh∼
They correspond to “line bundles” on P2
q
By derived equivalence line bundles are de-termined by certain stable representations of the quiver ∆
• Berest and Wilson (2002) used these ideas to reprove
Theorem A.
Aut A1 has a natural action on R(A1) – orbits indexed by N,
– the n-th orbit is in bijection with n-th Calogero-Moser space
Cn={(X,Y)∈ Mn(k)2| rk(YX−XY−I)≤ 1}/Gln(k) smooth connected affine variety of
There are many more k-algebras inducing a P2q. Interesting class: by Artin and Schelter (1986)
• Artin-Schelter algebra of dimension 3 is (i) graded k-algebra A = k ⊕ A1 ⊕ A2 ⊕ . . .
global dimension 3
(ii) A has polynomial growth
(iii) A is Gorenstein, i.e. for some l ∈ Z ExtiA(kA, A) ∼=
(
Ak(l) if i = 3,
0 otherwise.
• Classified by Artin, Tate and Van den bergh (1990) and Stephenson (1994).
• They are all noetherian domains GK-dim 3 have all expected nice homological proper-ties
• Assume A generated in degree one. Two possibilities:
– A is quadratic
0 → A(−3) → A(−2)3 → A(−1)3 → A → kA → 0
hA(t) = X
n
dimk Antn = 1
(1 − t)3 – A is cubic
0 → A(−4) → A(−3)2 → A(−1)2 → A → kA → 0
hA(t) = X
n
dimk Antn = 1
(1 − t)2(1 − t2)
• Generic class: called type A algebras quadratic: ayz + bzy + cx2 = 0 azx + bxz + cy2 = 0 axy + byx + cz2 = 0 cubic: ( ay2x + byxy + axy2 + cx3 = 0 ax2y + bxyx + ayx2 + cy3 = 0 where a, b, c ∈ k generic.
• A is determined by a triple (E, σ, j) where either – A is linear: j : E ∼= P2 if A is quadratic j : E ∼= P1 × P1 if A is cubic – A is elliptic: j : E ,→ P2, E divisor degree 3 if A is quadratic j : E ,→ P1×P1, E divisor bidegree (2, 2) if A is cubic
Generic case: A is type A and E is smooth elliptic curve (called generic type A)
• Define the set
• Van den Bergh and De Naeghel (2002) Theorem B.
Let A be elliptic quadratic and o(σ) = ∞.
Then the set R(A) is in bijection with `n∈NDn where Dn is smooth locally closed variety dimension 2n.
If A is of generic type A then Dn is affine. • Similar result by Nevins and Stafford (2002). • Aim of the talk:
analogue of Theorem B for cubic A.
N := {(ne, no) ∈ N2 | ne − (ne − no)2 ≥ 0} Theorem 1.
Let A be elliptic cubic and o(σ) = ∞. Then the set R(A) is in bijection with
`
(ne,no)∈N D(ne,no) where D(ne,no) is smooth
locally closed variety of dimension 2(ne − (ne − no)2).
If A is of generic type A then D(ne,no) is affine.
• Application: enveloping algebra
Hc = khx, y, zi/(yz − zy, xz − zx, xy − yx − z)
= khx, yi/([y, [y, x]], [x, [x, y]])
Let ϕ ∈ Aut(A1), ϕ(x) = −x, ϕ(y) = −y. Equality Ahϕi1 = Hc[z−1]0 induces bijection between sets
R(Ahϕi1 ) = { right Ahϕi1 -ideals}/=∼
and
R(Hc) = { reflexive graded right Hc-ideals }/=,sh∼
By Theorem 1 and further investigation Theorem 2. The set R(Ahϕi1 ) is in bijection with `(n e,no)∈N D(ne,no) where D(ne,no) = {(X,Y,X0,Y0) ∈ Mn e×no(k)2×Mno×ne(k)2 | Y0X − X0Y = I and rank(YX0−XY0−I) ≤ 1} /Gln e(k) × Glno(k)
are smooth affine varieties of dimension 2(ne − (ne − no)2).
2. Noncommutative quadrics • A cubic Artin-Schelter algebra
graded k algebra A = k ⊕ A1 ⊕ A2 ⊕ . . . Hilbert series 1 + 2t + 4t2 + 6t3 + 9t4 + . . . • graded right A-module M = ⊕i∈ZMi then
degree 0 ↓
M = · · · ⊕ M−1⊕M0 ⊕ M1 ⊕ . . . M (1) := · · · ⊕ M0 ⊕M1 ⊕ M2 ⊕ . . . • - GrMod(A) graded right A-modules
- Tors(A) direct limits of fdim modules
- GrMod(A) −→ GrMod(A)/ Tors(A) = Tails(A)π M 7→ M
A 7→ O
- M 7→ M (1) induces sh : M 7→ M(1)
- grmod(A), tors(A), tails(A) noeth. obj. - X := Proj(A) := (tails(A), O, sh)
A is determined by triple (E, σ, j)
• P = ⊕nPn ∈ grmod(A) is a point module if
A P and hP(t) = X n dimk Pntn = 1 1 − t Choosing a k-basis e0, e1, . . . in P0, P1, . . . ( e0 · x = α0e1 e0 · y = β0e1 , ( e1 · x = α1e2 e1 · y = β1e2 , ( e2 · x = α2e3 e2 · y = β2e3 , . . . for some αi, βi ∈ k. If r is relation in A then e0 · r = 0. Leads to equation in (α0, β0), (α1, β1)
E is the zero locus in P1 ×P1 of equation. • Example: if A is type A then E is
(c2−b2)α0β0α1β1+aα20(cα21−bβ12)+aβ02(cβ12−bα21) = 0 • Example: if A = Hc enveloping algebra
then E is
(α0β1 − β0α1)2 = 0
the double diagonal 2D on P1 × P1
• Point module P 7→ p closed point in E P (1)≥0 7→ σ(p)
Artin, Tate and Van den Bergh:
A −→ (E, σ, j) −→ B = B(E, σ, j) • A linear: B ∼= A
A elliptic: B ∼= A/gA with g ∈ A4 normal • Tails B and Qcoh E are equivalent
Tails A −⊗AB −→ ←− (−)A Tails B ˜ (−) −→ ←− Γ∗ Qcoh E > i∗ < i∗
i∗ is right exact (restriction functor) i∗ is exact
3. From reflexive ideals to line bundles
• A cubic Artin-Schelter algebra, X = Proj A • For I ∈ grmod(A) graded right ideal
m 7→ dimk Am − dimk Im
is linear in m.
• For J ∈ grmod(A) of rank one, ∃! d ∈ Z s.t. dimk Am − dimk J(d)m =
(
ne for m 0 even no for m 0 odd for some integers ne, no. We say
- J(d) is normalized
- (ne, no) are the invariants of J
• For ne, no ∈ Z define full subcategory of
grmod(A)
R(ne,no)(A) = {J ∈ grmod(A) | J rank one, reflexive, normalized with invariants (ne, no)}
• Apply exact quotient functor
π : grmod(A) → coh(X) I 7→ I
Let R(ne,no)(X) be image of R(ne,no)(A). Objects in R(ne,no)(X) are called
normalized line bundles.
• R(ne,no)(A) and R(ne,no)(X) are groupoids • Natural bijection between set
R(A) = { reflexive graded right A-ideals }/=,sh∼
and isoclasses of a (ne,no)∈Z2 R(ne,no)(A) and isoclasses of a (ne,no)∈Z2 R(ne,no)(X)
• For I ∈ R(ne,no)(X) its cohomology groups Hi(X, I) := ExtiX(O, I)
are partially computed (for I 6∼= O)
l . . . −5 −4 −3 −2 −1 0 1 . .
dimk H0(X, I(l)) . . . 0 0 0 0 0 0 ∗ . .
dimk H1(X, I(l)) . . . ∗ ne − 1 no ne no ne − 1 ∗ . .
dimk H2(X, I(l)) . . . ∗ 0 0 0 0 0 0 . .
Thus ne > 0, no ≥ 0 and easy to prove
R(0,0) = {O}
• By standard computation on the Euler form dimk Ext1X(I, I) = 2(ne − (ne − no)2)
hence for any integers ne, no
R(ne,no)(X) 6= ∅ ⇒ (ne, no) ∈ N Converse also true! See below.
4. From line bundles to quiver representa-tions
• A cubic Artin-Schelter algebra, X = Proj A • Due to a Theorem of Bondal (1990)
Db(coh X) RHomX(E,–) > < –⊗LΓE Db(mod Γ)
where E = O(3) ⊕ O(2) ⊕ O(1) ⊕ O and Γ is the quiver
−3 X−3 > Y−3 >−2 X−2 > Y−2 >−1 X−1 > Y−1 >0
with relations reflecting the relations in A.
Assume (ne, no) 6= (0, 0). Fix I ∈ R(ne,no)(X).
• Consider I as complex of degree zero
• By previous its image is in degree one RHomX(E, I) = M [−1]
where M = Ext1X(E, I)
• M is given by a representation of Γ H1(X, I(−3)) X > Y >H 1(X, I(−2)) X0 > Y0 >H 1(X, I(−1)) X00 > Y00 >H 1(X, I)
dimM = (no, ne, no, ne − 1) and relations.
For example, if A is type A then
X00 Y 00 · aY 0Y + cX0X bX0Y + aY 0X bY 0X + aX0Y aX0X + cY 0Y
!
How is “I is line bundle” translated? • Let P be point module, P = πP .
Cohomology groups given by
l . . . −5 −4 −3 −2 −1 0 1 . . .
dimk H0(X, P(l)) . . . 1 1 1 1 1 1 1 . . . dimk H1(X, P(l)) . . . 0 0 0 0 0 0 0 . . . dimk H2(X, P(l)) . . . 0 0 0 0 0 0 0 . . .
P determines the representation of Γ k α−3 > β−3 >k α−2 > β−2 >k α−1 > β−1 >k where p = ((α0, β0), (α1, β1)) ∈ E σip = ((αi, βi), (αi+1, βi+1))
• I is line bundle means Ext1X(P, I) = 0 0 = Ext1X(P, I) = H0(RHomX(P, I[1]))
∼
= H0(RHomΓ(p, M )) = HomΓ(p, M )
• Trivially,
Hom∆(M, p) = H0(RHomΓ(M, p))
= H0(RHomX(I[1], P)) = 0
• These properties characterize M . Theorem.
Let A be elliptic cubic and o(σ) = ∞. Then equivalence R(ne,no)(X) Ext1X(E,–) > < TorΓ1(–,E) C(ne,no)(Γ) where C(ne,no)(Γ) = {M ∈ mod(Γ) | dimM = (no, ne, no, ne−1) HomΓ(M, p) = HomΓ(p, M ) = 0 ∀p ∈ E}
Pick up another idea of Le Bruyn. I ∈ R(ne,no)(X) is determined by both H1(X, I(−3)) X > Y >H 1(X, I(−2)) X0 > Y0 >H 1(X, I(−1)) X00 > Y00 >H 1(X, I) H1(X, I(−4)) 0X > 0Y >H 1(X, I(−3)) X > Y >H 1(X, I(−2)) X0 > Y0 >H 1(X, I(−1))
So I is actually determined by repr. M0
H1(X, I(−3)) X > Y >H 1(X, I(−2)) X0 > Y0 >H 1(X, I(−1))
of the full subquiver Γ0 of Γ −3 X−3 > Y−3 >−2 X−2 > Y−2 >−1
Characterizing properties of M0?
• As M0 is the restriction of M :
certain rank condition involving X, Y, X0, Y 0 For example, if A is type A then
X00 Y 00 · aY 0Y + cX0X bX0Y + aY 0X bY 0X + aX0Y aX0X + cY 0Y ! = 0 yields rank aY 0Y + cX0X bX0Y + aY 0X bY 0X + aX0Y aX0X + cY 0Y ! ≤ dim ker X00 Y 00 = 2no − (ne − 1) • M0 is θ-stable for θ = (−1, 0, 1). Indeed:
∀I ∈ R(ne,no)(X) : ∃ v ∈ A2 : HomX(I, π(A/vA)) = 0 ⇒ HomΓ0(M0, Q0) = 0
⇒M0 ⊥ Q0
• These properties characterize M0 for (ne, no) 6= (1, 1)
Theorem.
Let A be elliptic cubic and o(σ) = ∞. Then equivalence C(ne,no)(Γ) Res > < Ind D(ne,no)(Γ0) where D(ne,no)(Γ0) = {F ∈ mod(Γ0) | dimF = (no, ne, no),
F is θ-stable, dimk(Ind F )0 ≥ ne − 1}
• Put α = (no, ne, no) and
D(ne,no) := {F ∈ Repα(Γ0) | F ∈ D(ne,no)(Γ0)}// Glα(k) Locally closed follows.
Smooth of dimension 2(ne − (ne − no)2) is proved.
5. Generic type A
Assume A is generic type A.
`
(ne,no) R(ne,no)(X) equivalent to
{M ∈ coh(X) | i∗M is line bundle on E deg. zero} By picking a suitable line bundle V on E
∀I ∈ R(ne,no)(X) : RHomE(i∗I, V) = 0 ⇒ RHomX(I, i∗V) = 0
⇒ RHomΓ0(M0, V 0) = 0
⇒M0 ⊥ V 0
for some representation V 0 ∈ mod(Γ0). Leads to
D(ne,no)(Γ0) = {F ∈ mod(Γ0) | dimF = (no, ne, no),
F ⊥ V 0, dimk(Ind F )0 ≥ ne − 1} As {F ∈ Repα(Γ0) | F ⊥ V 0} is affine
6. The enveloping algebra
Assume A = Hc is the enveloping algebra
Hc = khx, y, zi/(yz − zy, xz − zx, xy − yx − z) = khx, yi/([y, [y, x]], [x, [x, y]])
Work with Ered = D, the diagonal on P1 × P1.
`
(ne,no) R(ne,no)(X) equivalent to
{M ∈ coh(X) | i∗M line bundle on D deg. zero} ={M ∈ coh(X) | i∗M ∼= OD}
Translates into
∀I ∈ R(ne,no)(X) : HomX(I, π(A/zA)) = 0 ⇒ HomΓ0(M0, Q0) = 0 ⇒M0 ⊥ Q0 Leads to D(ne,no)(Γ0) = {F ∈ mod(Γ0) | dimF = (no, ne, no), F ⊥ Q0, dimk(Ind F )0 ≥ ne − 1} Thus D(ne,no) is affine.
We now simplify this expression. Let F ∈ Repα(Γ0) i.e.
kno X > Y >k ne X0 > Y 0 >k no • F ⊥ Q0 means Y 0X −X0Y is an isomorphism Defining ( X = X Y = Y and ( X0 = (Y 0X − X0Y )−1X0 Y0 = (Y 0X − X0Y )−1Y 0 this means YX0 − X0Y0 = I
• dimk(Ind F )0 ≥ ne − 1 means rank Y 0Y X0Y − 2Y 0X
Y 0X − 2X0Y X0X
!
≤ 2no−(ne−1)
7. Hilbert series of ideals
A is three-dimensional Artin-Schelter algebra. Question.
If I = ⊕nIn ∈ grmod(A) is a graded right ideal, how does hI(t) = Pn dimk Intn look like?
• Macaulay (1927): for A = k[x, y, z] Restrict to pd I ≤ 1.
m 7→ dimk Am − dimk Im is linear in m, i.e. ∃! d ∈ Z s.t.
dimk Am − dimk I(d)m = n for m 0 for some integer n.
In terms of formal power series hI(d)(t) = hk[x,y,z](t) − s(t) 1 − t = 1 (1 − t)3 − s(t) 1 − t
Turns out s(t) is a Castelnuovo polynomial s(t) = 1 + 2t + 3t2 +· · · + utu−1+ sutu +· · · + svtv
u ≥ su ≥ . . . ≥ sv ≥ 0
for some integers u, v ≥ 0. s(1) = n ≥ 0.
Visualized in form of a stair Example:
1 2 3 4 5 5 3 2 1 1 1 1
Answer (commutative case). h(t) ∈ Z((t)) is of the form h
I(d)(t) for some
graded ideal I with pd I ≤ 1 if and only if
h(t) = 1
(1 − t)3 −
s(t) 1 − t
• Van den Bergh and De Naeghel (2004): Answer (quadratic case).
Let A be quadratic Artin-Schelter algebra. h(t) ∈ Z((t)) is of the form h
I(d)(t) for
some graded right ideal I with pd I ≤ 1 iff
h(t) = 1
(1 − t)3 −
s(t) 1 − t
for some Castelnuovo polynomial s(t).
If in addition A is elliptic and o(σ) = ∞ then I may be chosen reflexive.
• What is the answer for cubic A? As indicated above: ∃! d ∈ Z s.t. dimk Am − dimk I(d)m =
(
ne for m 0 even
no for m 0 odd for some integers ne, no ≥ 0.
In terms of formal power series
hI(t) = 1
(1 − t)2(1 − t2) −
s(t) 1 − t2 for some s(t) = Pi siti ∈ Z[t, t−1]
with Pi s2i = ne and Pi s2i+1 = no.
Answer (cubic case).
Let A be cubic Artin-Schelter algebra. h(t) ∈ Z((t)) is of the form h
I(d)(t) for
some graded right ideal I with pd I ≤ 1 iff
h(t) = 1
(1 − t)2(1 − t2) −
s(t) 1 − t2 for some Castelnuovo polynomial s(t).
If in addition A is elliptic and o(σ) = ∞ then I may be chosen reflexive.
Example:
1 2 3 4 5 5 3 2 1 1 1 1
P
i s2i = 14 even weight
P
i s2i+1 = 15 odd weight.
As a consequence, for all integers ne, no
R(ne,no)(X) 6= ∅ ⇔ ∃ Castelnuovo polynomial s(t) with even weight ne and odd weight no But recall
R(ne,no)(X) 6= ∅ ⇒ (ne, no) ∈ N
where
N = {(ne, no) ∈ N2 | ne − (ne − no)2 ≥ 0}
For (ne, no) ∈ N easy to construct s(t), hence R(ne,no)(X) 6= ∅ ⇔ (ne, no) ∈ N
We end with combinatorical by-product. Shift rows of s(t) to left
For any partition λ, put draughts colouring. Number of is called even weight of λ
Number of is called odd weight of λ Bijection between
{Castelnuovo polynomials
with even weight ne and odd weight no} and
{partitions in distinct parts
Theorem. Let ne, no ∈ N. There is a partition
λ in distinct parts with even weight ne and odd
weight no iff
ne − (ne − no)2 ≥ 0
Special case of Chung’s Conjecture (1951) Proved by G. de B. Robinson (1951)
In particular, the number of such partitions λ is the number of partitions of ne − (ne − no)2. We may extend Theorem to all partitions