Classification of Lie algebras of differential operators
Helge Γystein Maakestad
email [email protected]
[email protected]
(Date: January 2019)
Abstract.
In a previous paper we introduced the notion of a D-Lie algebra L~. A D-Lie algebra L~ is an A/k-Lie-Rinehart algebra with a right A-module structure
and a canonical central element D satisfying several conditions. We used this notion to define the universal enveloping algebra of the category of L~-connections
and to define the cohomology and homology of an arbitrary connection. In this note we introduce the canonical quotient L of a D-Lie algebra L~ and use this to classify
D-Lie algebras where L is projective as A-module. We define for any 2-cocycle fβZ2(Derkβ(A),A) a functor FfΞ±β(β) from the category of
A/k-Lie-Rinehart algebras to the category of D-Lie algebras and classify D-Lie algebras with projective canoncial quotient using the functor FfΞ±β(β). We prove a similar classification for
non-abelian extensions of D-Lie algebras. We moreover classify maps of D-Lie algebras and L~-connections (E,Ο) in the case when the canonical quotient L of L~ is projective as A-module.
Key words and phrases:
D-Lie algebra, canonical quotient, Lie-Rinehart algebra, connection, classification, non-abelian extension
Contents
- 1 Introduction
- 2 Classification of morphisms of D-Lie algebras
- 3 Classification of D-Lie algebras with projective canonical quotient
- 4 Classification of connections on D-Lie algebras with projective canonical quotient
1. Introduction
A D-Lie algebra L~ is a refinement of the notion of an A/k-Lie-Rinehart algebra: It is a k-Lie algebra and left PA/k1β-module, where PA/k1β is the first
order module of principal parts of A/k. The Lie-product on L~ satisfies the anchor condition
[TABLE]
for all u,vβL~ and cβA. Here Ο~:L~βDerkβ(A) is a map of k-Lie algebras and left P-modules. Hence if we view L~ as a left A-module and k-Lie algebra, it follows
the pair (L~,Ο~) is an A/k-Lie-Rinehart algebra. There is a canonical central element DβL~ with the property that
[TABLE]
holds for uβL~ and cβA. We may define the notion of a connection Ο:L~βDiff1(E) where E is a left A-module. In the papers [8] and [9] general properties
of the category of D-Lie algebras are introduced and studied: Universal enveloping algebras of D-Lie algebras, cohomology and homology of connections on D-Lie algebras.
Let D-Lieβ denote the category of D-Lie algebras and morphisms. In [8] Theorem 3.5 and 3.18 we prove the following result:
Theorem 1.1**.**
There are covariant functors
[TABLE]
with the following property: For any D-Lie algebra L~ there are exact equivalences of categories
[TABLE]
with the property that F1β and F2β preserves injective and projective objects.
We used the associative unital rings Uβ(L~) and UΟ(L~) to define the cohomology and homology of an arbitrary L~-connection (E,Ο). By Lemma 3.19 in [8] there is for any 2-cocycle fβH2(Derkβ(A),A) an exact equivalence of categories
[TABLE]
preserving injective and projective objetcs. Here L is an A/k-Lie-Rinehart algebra and Connβ(L) is the category of L-connections and morphisms of connections. Hence we may use the associative unital ring Uβ(L(fΞ±)) to define the cohomology and homology of any L-connection, flat or non-flat. Using Οfβ we get for any pair of L-connections V,W and any integer iβ₯0 isomorphisms
[TABLE]
Hence the associative ring Uβ(L(fΞ±)) may be used to calculate the βtrueβ Ext-groups of V and W. Theorem 1.1 was one of the main reasons for the introduction of the notion D-Lie algebra.
Given an almost commutative ring U with filtration Uiβ where U is generated by U1β, there is a pre-D-Lie algebra L~ and a 2-sided ideal JβUβ(L~) and an isomorphism
[TABLE]
of filtered associative rings. Hence there is an exact equivalence of categories
[TABLE]
preserving injective and projective objects, where Modβ(U) is the category of left U-modules and Modβ(L~,Id,J)is the category of L~-connections with J-curvature equal to zero.
Most rings of differential operators appearing in the litterature
are almost commutative and generated by U1β. As special cases it follows the ring Diff(L) of differential operators on a linebundle L on A
and the generalized universal enveloping algebra U(A,L,f) where L is an A/k-Lie-Rinehart algebra and f is a 2-cocycle on L with values in A,
may be realized as UJββ(L~) (see [8], Example 3.35). The rings Uβ(L~) and UJββ(L~) have canonical filtrations Uβ(L~)i,UJββ(L~)i and the associated graded rings are commutative.
Let L~β:=Uβ(L~)1/Uβ(L~)0. There is a canonical surjective map of graded A-algebras
[TABLE]
The map Ο is not always an isomorphism in the case when L~ is a projective A-module. Hence the PBW-theorem does not always hold for Uβ(L~). If (U,Uiβ) is an almost commutative ring with L:=U1β/U0β a projective A-module and the canonical map Ξ·:SymAββ(L)βGr(U) an isomorphism, it follows there is an isomorphism of associative rings Uβ
U(A,L,f) where L:=U1β/U0β and
f a 2-cocycle for L with values in A. Hence a left U-module A corresponds to a connection β:LβEndkβ(E) of curvature type f. It follows the kβth Chern class of E is determined
by the first Chern class. A similar PBW-theorem for the universal ring Uβ(L~) would put similar conditions on all Chern classes of all finite rank projective A-modules. Hence the study
of the associative ring Uβ(L~) and the PBW-theorem for Uβ(L~) will have applications in the study of the Chern classes and Chern character of A.
Note: The category Connβ(L) of connections on an A/k-Lie-Rinehart algebra L is a small abelian category, hence the Freyd-Mitchell full embedding theorem gives an equivalence Ο of Connβ(L) with
a sub category of Modβ(R) where R is an associative ring. The equivalence Ο does not preserve injective and projective objects, hence we cannot use Ο to define the cohomology and homology
of a connection. Theorem 1.1 gives a geometric construction of cohomology and homology groups of any L~- and L-connection where L~ is any D-Lie algebra and
L is any A/k-Lie-Rinehart algebra. The construction is functorial and can be done for any scheme and any sheaf of Lie-Rinehart algebras.
The aim of this paper is to classify maps of D-Lie algebras. We also classify D-Lie algebras and non-abelian extensions of D-Lie algebras with projective canonical quotient in terms of a functor FfΞ±β introduced in [8].
Let (L,Ξ±) and (Lβ²,Ξ±β²) be A/k-Lie-Rinehart algebras and let f,gβZ2(Derkβ(A),A) be two 2-cocycles. Let Fgβ(L):=L(gΞ±) and Ffβ(Lβ²):=Lβ²(fΞ±β²) be the
D-Lie algebras introduced in [8], Theorem 2.7. In Theorem 2.3 we prove the following:
Theorem 1.2**.**
There is a map of D-Lie algebras Ο:L(gΞ±)βLβ²(fΞ±β²) if and only if gΞ±β=fΞ±β in Ξ±βH2(Derkβ(A),A). In this case
there is an equality between the set of maps of D-Lie algebras Ο:L(gΞ±)βLβ²(fΞ±β²) and the set of maps of A/k-Lie-Rinehart algebras Ο2β:LβLβ².
In Theorem 3.9 we prove the following:
Theorem 1.3**.**
Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra and let (L,Ξ±) be the canonical quotient A/k-Lie-Rinehart algebra of L~. Assume L is projective as left A-module.
There is an isomorphism L~β
L(fΞ±) as D-Lie algebras where L(fΞ±):=FfΞ±β(L) and FfΞ±β is the functor from Example 3.3. Hence L~ is uniquely determined up to isomorphism
by the canonical quotient (L,Ξ±) and the 2-cocycle fβH2(Derkβ(A),A).
In Corollary 3.10 we classify maps of D-Lie algebras between two D-Lie algebras L~1β and L~2β.
In Corollary 4.20 we get a classification of non-abelian extensions of D-Lie algebras in terms of the functor FfΞ±β.
In the paper we introduce the category of D-Lie algebras, connections on D-Lie algebras using the module of principal parts. This gives an equivalent definition of a D-Lie algebra to the one introduced in [8]. We prove in Theorem 4.19 that for any L~-connection (E,Ο) there is an extension of D-Lie algebras
[TABLE]
with the property that 1.3.1 splits in the category of D-Lie algebras if and only if E has a flat L~-connection Οβ²:L~βDiff1(E). Theorem 4.19 generalize a similar result proved
in [9] for non-abelian extensions of A/k-Lie-Rinehart algebras.
Using the P-module structure on the D-Lie algebra L~ we define in Definition 4.6 the correspondence
Z(Ο,Ο1β,Ο2β) associated to a connection and degeneracy locies Ο1β and Ο2β. The correspondence Z(Ο,Ο1β,Ο2β) induce an endomorphism
[TABLE]
of the Chow-group CHβ(X). The correspondence Z(Ο,Ο1β,Ο2β) and Chow-operator I(Ο,Ο1β,Ο2β) depend in a non-trivial way on the P-module structure of L~ and P-linearity of the L~-connection Ο and cannot be defined
for an ordinary L-connection where (L,Ξ±) is an A/k-Lie-Rinehart algebra (see Example 4.3). There is no non-trivial right A-module structure on L.
We also prove in Theorem 4.15 the following result relating the category of connections on an A/k-Lie-Rinehart algebra L to the category of connections on the D-Lie algebra L(fΞ±):
Theorem 1.4**.**
Let (L,Ξ±) be an A/k-Lie-Rinehart algebra and let fβZ2(Derkβ(A),A) be a 2-cocycle. There is an equivalence of categories
[TABLE]
from the category Connβ(L,End) of (L,Ο)-connections β, to the category Connβ(L(fΞ±)) of L(fΞ±)-connections Ο.
Let u:=az+x,v:=bz+yβL(fΞ±) and let eβE. Let (E,β) be an L-connection and let Οββ:=Cfβ(β). The following holds:
[TABLE]
For any L(fΞ±)-connection (E,Ο) there is an (L,Ο)-connection (E,β) with Cfβ(E,β)=(E,Ο).
Hence there is for any 2-cocycle fβZ2(Derkβ(A),A) a functorial way to define an L(fΞ±)-connection Cfβ(E,β) from an L-connection (L,Ξ±).
As a Corollary we are able to classify L~-connections on D-Lie algebras with projective canonical quotient. In Corollary 4.16 we prove the following:
Corollary 1.5**.**
Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra and let (E,Ο) be an L~-connection. Assume the canonical quotient L of L~
is a projective A-module. It follows any L~-connection (E,Ο) is on the form Cfβ(E,β) where fβZ2(Derkβ(A),A) and (E,β) is an (L,Ο)-connection for ΟβEndAβ(E).
If Ο(D)=I it follows we may choose (E,β) to be an L-connection.
2. Classification of morphisms of D-Lie algebras
Let in the following A be a fixed commutative k-algebra where k is a fixed commutative unital ring. Let f,gβZ2(Derkβ(A),A) bw two 2-cocycles and consider the two
k-Lie algebras D1(A,f) and D1(A,g). Let P:=AβkβA/I2 be the first order module of principal parts of A/k. The k-Lie algebra D1(A,f) has the following left and right A-module structure:
[TABLE]
and
[TABLE]
for (a,x)βD1(A,f) and cβA. Let zfβ:=(1,0)βD1(A,f). It follows u:=(a,x) that uc=cu+x(c)zfβ=cu+Οfβ(u)(c)zfβ. Similarly for D1(A,g). Let d:AβAβkβA be defined by
dc:=1βcβcβ1. It follows dc(u)=ucβcu=Οfβ(u)(c)z. One checks that zc=cz hence dbdc(u)=db(Οfβ(u)(c)z=0 hence D1(A,f) is annihilated by I2βAβkβA
and it follows D1(A,f) is a left P-module.
The two projection maps Οfβ:D1(A,f)βDerkβ(A) and Οgβ:D1(A,g)βDerkβ(A) and
Οfβ,Οgβ are maps of P-modules and k-Lie algeras.
We want to classify maps
[TABLE]
of P-modules and k-Lie algebras such that
[TABLE]
Lemma 2.1**.**
Let Ο:D1(A,g)βD1(A,f) be a map of P-modules and k-Lie algebras satisfying condition 2.0.1. It follows Ο(a,x)=(a+Ο1β(x),x) where
Ο1ββC1(Derkβ(A),A), g=f+d1(Ο1β) and d1:C1(Derkβ(A),A)βC2(Derkβ(A),A) is the first differential in the Lie-Rinehart complex of Derkβ(A). Hence the map Ο exists if and only if
gβ=fββH2(Derkβ(A),A). If the map Ο exists it is an isomorphism of P-modules and k-Lie algebras with inverse Ο:D1(A,f)βD1(A,g) defined by
Ο(a,x):=(aβΟ1β(x),x).
Proof.
Let Ο:D1(A,g)βD1(A,f) be a map of P-modules and k-Lie algebras satisfying 2.0.1 It follows for any (a,x)βD1(A,g) we get
[TABLE]
with Ο1ββC1(Derkβ(A),A). One checks the map Ο is a map of k-Lie algebras if and only if g=f+d1(Ο1β). The map Ο in 2.1.1 is always a map of P-modules.
Hence the map Ο exists if and only if gβ=fββH2(Derkβ(A),A). If Ο exists, an inverse Ο is given by
Ο(a,x):=(aβΟ1β(x),x). The Lemma follows.
β
Assume (L,Ξ±) is an A/k-Lie-Rinehart algebra and let f,gβZ2(Derkβ(A),A) be two 2-cocycles and let fΞ±,gΞ±βZ2(L,A) be the pull-back cocycles. Let Fgβ(L):=L(gΞ±):=AzgββL
be the A/k-Lie-Rinehart algebra defined in [8]. It has the following k-Lie product:
[TABLE]
for (a,x),(b,y)βL(gΞ±). There is a map
[TABLE]
defined by
[TABLE]
The pair (L(gΞ±),Οgβ) is an A/k-Lie-Rinehart algebra.
Note: The map Ξ± induce a map of cohomology groups
[TABLE]
Let Im(Ξ±β):=Ξ±βH2(Derkβ(A),A) denote the image of the mape Ξ±β in H2(L,A).
Lemma 2.2**.**
There is a map Ξ±gβ:L(gΞ±)βD1(A,f) of P-modules and k-Lie algebras with ΟβΞ±gβ=Οgβ if and only if there is an element Ξ±1ββC1(L,A)
with gΞ±=fΞ±+d1(Ξ±1β). The map Ξ±gβ is defined as follows: Ξ±gβ(a,x):=(a+Ξ±1β(x),Ξ±(x))βD1(A,f). Hence there exists a 5-tuple (L(gΞ±),Ξ±gβ,Οgβ,[,],zgβ) which is a
D-Lie algebra if and only if there is an equality of cohomology classes
[TABLE]
Proof.
By definition the map Ξ±gβ must look as follows:
[TABLE]
where (a,x)βL(gΞ±) and Ξ±1ββC1(L,A). One checks the map Ξ±gβ is a map of left P-modules. The map Ξ±gβ is a map of k-Lie algebras
if and only if gΞ±=fΞ±+d1(Ξ±1β). Hence the 5-tuple (L(gΞ±),Ξ±gβ,Οgβ,[,],zgβ) is a D-Lie algebra over D1(A,f) if and only if gΞ±β=fΞ±β in
Ξ±βH2(Derkβ(A),A). The Lemma follows.
β
We may now classify maps between arbitrary D-Lie algebras. Let (L,Ξ±) and (Lβ²,Ξ±β²) be A/k-Lie-Rinehart algebras and let f,gβZ2(Derkβ(A),A) with gΞ±=fΞ±+d1(Ξ±1β)
for Ξ±1ββC1(L,A). Hence there is by Lemma 2.2 a structure as D-Lie algebra Ξ±gβ:L(gΞ±)βD1(A,f) given by the map Ξ±gβ(a,x):=(a+Ξ±1β(x),Ξ±(x)).
The 5-tuple (L(gΞ±),Ξ±gβ,Οgβ,[,],zgβ) is a D-Lie algebra over D1(A,f) with Οgβ(a,x):=Ξ±(x)βDerkβ(A).
Let (L,Ξ±) and (Lβ²,Ξ±β²) be A/k-Lie-Rinehart algebras and let f,gβZ2(Derkβ(A),A) be two 2-cocycles. Let Fgβ(L):=L(gΞ±) and Ffβ(Lβ²):=Lβ²(fΞ±β²) be the
D-Lie algebras introduced in [8], Theorem 2.7.
Theorem 2.3**.**
There is a map of D-Lie algebras Ο:L(gΞ±)βLβ²(fΞ±β²) if and only if gΞ±β=fΞ±β in Ξ±βH2(Derkβ(A),A). In this case
there is an equality between the set of maps of D-Lie algebras Ο:L(gΞ±)βLβ²(fΞ±β²) and the set of maps of A/k-Lie-Rinehart algebras Ο2β:LβLβ².
Proof.
If there is an equality gΞ±β=fΞ±β it follows there is a map
[TABLE]
defined by
[TABLE]
The map Ο:L(gΞ±)βLβ²(fΞ±β²) must look as follows:
[TABLE]
where Ο1β:LβA and Ο2β:LβLβ². Since Ξ±fβ²ββΟ=Ξ±gβ it follows Ο1β=Ξ±1β and Ξ±β²βΟ2β=Ξ±. Hence Ο2β is a map
of A/k-Lie-Rinehart algebras. One checks the map Ο is P-linear and a map of k-Lie algebras. Conversely, for an arbitrary map Ο2β:LβLβ² of
A/k-Lie-Rinehart algebras it follows the map Ο(a,x):=(a+Ξ±1β(x),Ο2β(x)) is a map of D-Lie algebras. The Theorem follows.
β
3. Classification of D-Lie algebras with projective canonical quotient
In this section we define the notion of a D-Lie algebra, the category of D-Lie algebras and connections on D-Lie algebras.
The module Diff1(A) of first order differential operators
on a commutative k-algebra A is in a canonical way a k-Lie algebra and a left P-module where P:=AβkβA/I2 is the module of first order principal parts on A/k.
By definition Diff1(A):=HomAβ(P,A) and it follows Diff1(A) is in a canonical way a left and right A-module. There is a canonical projection map Ο:Diff1(A)βDerkβ(A) which is a map of k-Lie algebras and left P-modules. There is a canonical inclusion of left and right A-modules and k-Lie algebras Diff1(A)βEndkβ(A) and
the Lie product [,] on Diff1(A) satisfies the following formula:
[TABLE]
for all u,vβDiff1(A) and cβA. There is a canonical element DβDiff1(A) with the property that
[TABLE]
The element D is central in Diff1(A) and Ο(D)=0. A D-Lie algebra L~ is a generalization of Diff1(A) and we may speak the category of D-Lie algebras, extensions and non-abelian extensions
of D-Lie algebras, cohomology of D-Lie algebras, connections etc. Equation 3.0.1 is the equation defining an A/k-Lie-Rinehart algebra and any D-Lie algebra L~ has an underlying A/k-Lie-Rinehart algebra. Hence we may view a D-Lie algebra as a refinement of the notion of a A/k-Lie-Rinehart
algebra. A D-Lie algebra is an A/k-Lie-Rinehart algebra with extra structure defined by the P-module structure, the canonical element D and Equations 3.0.1 and 3.0.2.
Note: We may consider the A/k-Lie-Rinehart algebra Derkβ(A)βDiff1(A) and Derkβ(A) has a canonical structure as k-Lie algebra and left A-module. It has no-non-trivial right
A-module structure. To get a non-trivial right A-module structure, we must consider the abelian extension Diff1(A) and this is one of the motivations for the introduction of the notion
of a D-Lie algebra. The P-module structure on Diff1(A) is canonical and is related to the notion of curvature of a connection.
Using the P-module structure on L~ we define the correspondence
Z(Ο,Ο1β,Ο2β) associated to a connection and degeneracy locies Ο1β and Ο2β. The correspondence Z(Ο,Ο1β,Ο2β) induce an endomorphism
[TABLE]
of the Chow-group CHβ(X).
The main Theorem in this section is Theorem 4.19 where we construct the non-abelian extension End(L~,E) of any L~-connection (E,Ο) in Connβ(L~,Id),
and prove the canonical projection map of D-Lie algebras
[TABLE]
splits in the category of D-Lie algebras if and only if E has a flat L~-connection in Connβ(L~,Id). In [9] a similar result was proved for A/k-Lie-Rinehart algebras.
We also define the canonical quotient A/k-Lie-Rinehart algebra (L,ΟLβ) of a D-Lie algebra L~ and classify the set of D-Lie algebras with canonical quotient L. It follows from Theorem
3.9 the D-Lie algebra L~ is uniquely determined by (L,ΟLβ) up to isomorphism when L is projective as left A-module. Hence for any cohomology class
c=fββH2(Derkβ(A),A) and any A/k-Lie-Rinehart algebra (L,Ξ±) with L a projective A-module, there is a unique D-Lie algebra Ξ±~:L~βD1(A,f) with canonical quotient (L,ΟLβ), defined by L~:=FfΞ±β(L), where FfΞ±β is the functor defined and studied in [8].
Let in the following A be a fixed commutative unital k-algebra where k is a commutative unital ring. Let Derkβ(A) be the k-Lie algebra of k-linear derivations of A.
Let fβZ2(Derkβ(A),A) be a 2-cocycle and let D1(A,f):=AβDerkβ(A) with the following k-Lie algebra structure and AβkβA-module structure:
Let u:=(a,x),v:=(b,y)βD1(A,f) and define
[TABLE]
Define for any element cβA
[TABLE]
and
[TABLE]
It follows D1(A,f) is a left AβkβA-module. Define the map Ο:D1(A,f)βDerkβ(A) by Ο(u):=Ο(a,x)=x. Define for xβDerkβ(A) xc:=cx. It follows Derkβ(A) is a left AβkβA-module.
Let P:=AβkβA/I2 where I is the kernel of the multiplication map. It follows P is the first order module of principal parts of A/k. Let d:AβP be the universal derivation.
Let p(a):=1βa and q(a):=aβ1. It follows d=pβq. Let D:=(1,0)βD1(A,f).
Lemma 3.1**.**
The k-Lie algebra D1(A,f) is in a canonical way a left P-module. The map Ο:D1(A,f)βDerkβ(A) is a map of k-Lie algebras and left P-modules.
The following holds for all u,vβD1(A,f) and cβA:
[TABLE]
The element D is a central element with Ο(D)=0.
Proof.
Since D1(A,f) is a left and right A-module with (au)b=a(ub) for all uβD1(A,f) and a,bβA it follows D1(A,f) is a left AβkβA-module. One checks that for any element
wβI2 it follows wu=0 hence D1(A,f) is a left P-module. The map Ο is left and right A-linear. It follows Ο is a map of P-modules and k-Lie algebras. One checks
Equation 3.1.1 and 3.1.2 holds and the Lemma is proved.
β
The notion of a D-Lie algebra is a generalization of the k-Lie algebra and left P-module D1(A,f) from Lemma 3.1.
Definition 3.2**.**
Let fβZ2(Derkβ(A),A) be a 2-cocycle and let D1(A,f) be the k-Lie algebra and P-module defined in Lemma 3.1.
A 5-tuple (L~,Ξ±~,Ο~,[,],D) is a D-Lie algebra if the following holds. L~ is a k-Lie algebra and left P-module. The element Ξ±~ is a map
[TABLE]
of P-modules and k-Lie algebras. The element Ο~ is a map
[TABLE]
of left P-modules and k-Lie algebras with Ο~=ΟβΞ±~. The k-Lie product satisfies
[TABLE]
for all u,vβL~ and aβA. The following holds for all aβA:
[TABLE]
The element D is a central element in L~ with Ο~(D)=0. Given two D-Lie algebras (L~,Ξ±~,Ο~,[,],D) and (L~β²,Ξ±~β²,Ο~β²,[,],Dβ²). A map of D-Lie algebras is a map
[TABLE]
of left P-modules and k-Lie algebras with Ο(D)=Dβ² and Ο~β²=Ο~βΟ. Let D-Lie denote the category of D-Lie algebras and maps. An ideal I~
in a D-Lie algebra L~ is a sub-k-Lie algebra and sub-P-module I~βL~ such that [L~,I~]βI~. A D-ideal I~ is a
sub-k-Lie algebra and sub-P-module I~βL~ with [L~,I~]βI~ and with Dβ/I~.
Note: One checks Definition 3.2 gives the same notion as Definition 2.3 in [8].
Example 3.3**.**
Non-trivial examples: D-Lie algebras and A/k-Lie-Rinehart algebras.
It follows from Lemma 3.1 that the 5-tuple (D1(A,f),id,Ο,[,],D) is a D-Lie algebra for any 2-cocycle fβZ2(Derkβ(A),A).
Given a D-Lie algebra
(L~,Ξ±~,Ο~,[,],D) it follows the left A-module L~ and the map Ο~:L~βDerkβ(A) is an A/k-Lie-Rinehart algebra.
Given any 2-cocycle fβZ2(Derkβ(A),A), let (L,Ξ±) be an A/k-Lie-Rinehart algebra and let fΞ±βZ2(L,A) be the pull back 2-cocycle. There is by Theorem 2.7 in [8] a functor
[TABLE]
from the category LR(A/k) of A/k-Lie-Rinehart algebras to the category D-Lie of D-Lie algebras, hence it is easy to give non-trivial examples of D-Lie algebras. When H2(Derkβ(A),A)ξ =0 we get for any fβZ2(Derkβ(A),A) a non-trivial functor FfΞ±β. Since FfΞ±β(L) is independent of choice of representative for the class c:=fΞ±ββH2(L,A) the following holds:
There is for any two representatives f,fβ² for the class c a canonical isomorphism Ffβ(L)β
Ffβ²β(L) of extensions of A/k-Lie-Rinehart algebras.
Hence the two functors Ffβ and Ffβ²β are equal up to isomorphism. Hence we get from Theorem 2.7 in [8] a well defined functor Fcβ for any cohomology class c:=fΞ±ββH2(L,A):
[TABLE]
When the class c is the zero class it follows the underlying A/k-Lie-Rinehart algebra of Fcβ(L):=AzβL is the trivial abelian extension of L by A. Hence the cohomology group H2(L,A) parametrize
a large class of non-trivial functors Fcβ.
Lemma 3.4**.**
Let Ο:L~1ββL~2β be a map of D-Lie algebras. It follows the kernel ker(Ο) is a D-ideal in L~1β. Let I~:=I/I2βP be the module of differentials
and let L~ be any D-Lie algebra. It follows I~L~βL~ is an ideal. Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra and let J:=\{aD:\text{ with a\in A}\}\subseteq\tilde{L}. It follows
J is an ideal. Let L:=L~/J. There is an exact sequence of P-modules and k-Lie algebras
[TABLE]
There is a canonical map of left A-modules and k-Lie algebras Ξ±Lβ:LβDerkβ(A) making (L,Ξ±Lβ) into an A/k-Lie-Rinehart algebra. The sequence
3.4.1 is an exact sequence of A/k-Lie-Rinehart algebras. The ideal J is a free left A-module on the element D.
Proof.
One checks that ker(Ο) is a P-submodule of L~1β and that [L~1β,ker(Ο)]βker(Ο). Since Dβ/ker(Ο) it follows ker(Ο) is a D-ideal.
The same holds for I~: I~ is a P-module and [L~,I~]βI~. For any element uβL~ and cβA it follows
uc=cu+Ο~(u)(c)D. It follows
[TABLE]
since D is central. It follows [L~,J]βJ. Hence [J,J]βJ and J is a k-Lie algebra. We get
[TABLE]
hence
[TABLE]
since Dc=cD. Hence J is a left P-module. It follows the Sequence 3.4.1 is an exact sequence of P-modules and k-Lie algebras. If uβL~/J is an element
with uβL~ it follows ucβcu=Ο~(u)(c)D:=0. Hence uc=cu in L~/J. Hence L~/J is trivially a left P-module.
Since Ο~(aD)=aΟ~(D)=0 it follows we get a canonical map
[TABLE]
and one checks (L~/J,ΟLβ) is an A/k-Lie-Rinehart algebra. The rest is clear and the Lemma follows.
β
Definition 3.5**.**
Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra. The quotient A/k-Lie-Rinehart algebra (L,ΟLβ) with L:=L~/J from Lemma
3.4 is the canonical quotient of L~.
Example 3.6**.**
When the canonical quotient is projective.
Assume L:=L~/J where J is the ideal defined in Lemma 3.4 and let ΟLβ:LβDerkβ(A) be the anchor map.
Assume L is projective as left A-module and let s be a left A-linear section of the canonical projection map
p:L~βL. Define for u,vβL the following map
[TABLE]
by
[TABLE]
Define the map βsβ:LβEndkβ(J) by
[TABLE]
Lemma 3.7**.**
It follows (J,βsβ) is a flat L-connection and ΟsββZ2(L,(J,βsβ)) where Cp(L,(J,βsβ)) is the Lie-Rinehart complex of the flat connection (J,βsβ).
If sβ² is another left A-linear splitting of p it follows there is an equality of connections βsβ=βsβ²β. If Οsβ²β is the 2-cocycle associated to sβ² it follows there is an element
ΟβC1(L,(J,βsβ)) with Οsβ²β=Οsβ+dβsβ1β(Ο). Hence there is an equality of cohomology classes
[TABLE]
Proof.
One checks that βsβ is a flat L-connection on J and that ΟβZ2(L,(J,βsβ)). Assume sβ²=s+Ο where ΟβHomAβ(L,J). We get
for any element uβL and xβJ the following:
[TABLE]
since J is an abelian k-Lie algebra. It moreover follows
[TABLE]
[TABLE]
[TABLE]
since J is an abelian Lie algebra and hence [Ο(u),Ο(v)]=0 for all u,vβL. The Lemma follows.
β
Define the map
[TABLE]
by
[TABLE]
Make the following definition: Jβ(Ο,Ο)L is the left A-module JβL with the following right A-module structure: Given z:=(u,x)βJβL and cβA
define
[TABLE]
We get since uc=cu the following:
[TABLE]
[TABLE]
Hence
[TABLE]
Since Ο~=ΟLββp and pβs=Id it follows Ο~(s(u))=ΟLβ(p(s(u))=ΟLβ(u). It follows zc=cz+ΟLβ(u)(c)(D,0)=cz+ΟLβ(u)(c)D~ where D~:=(D,0).
Let z:=(x,u),w:=(y,v)βJβL and define the following product:
[TABLE]
It follows the product [,] is a k-Lie product on JβL. Define the map
[TABLE]
by
[TABLE]
It follows Ο is an isomorphism of P-modules and k-Lie algebras. Define the map
[TABLE]
by
[TABLE]
Define ΟJβ:JβLβDerkβ(A) by ΟJβ(u,x):=ΟLβ(x). Associated to a left A-linear splitting s of p:L~βL we get by Lemma 3.7
a unique flat connection βsβ:LβEndkβ(J), a unique cohomology class ΟsβββH2(L,(J,βsβ)) and a 5-tuple
(JβL,Ξ±Jβ,ΟJβ,[,],D~).
Proposition 3.8**.**
Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra and let (L,Ξ±) be the canonical quotient A/k-Lie-Rinehart algebra of L~. Assume L is projective as left A-module.
The 5-tupe (JβL,Ξ±Lβ,ΟLβ,[,],D~) constructed above is a D-Lie algebra and there is an isomorphism JβLβ
L~ of D-Lie algebras.
Proof.
The proof follows from the construction and calculations above.
β
Theorem 3.9**.**
Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra and let (L,Ξ±) be the canonical quotient A/k-Lie-Rinehart algebra of L~. Assume L is projective as left A-module.
There is an isomorphism L~β
L(fΞ±) as D-Lie algebras where L(fΞ±):=FfΞ±β(L) and FfΞ±β is the functor from Example 3.3. Hence L~ is uniquely determined by the canonical quotient (L,Ξ±) and the 2-cocycle fβH2(Derkβ(A),A).
Proof.
Let s be a left A-linear splitting of the canonical projection map p:L~βL.
From Proposition 3.8 it follows the 5-tuple (JβL,Ξ±Lβ,ΟLβ,[,],D~) is a D-Lie algebra and
there is by 3.7.3 construction an isomorphism of D-Lie algebras
[TABLE]
defined by
[TABLE]
Consider the map Ξ±~:L~βD1(A,f). It looks as follows: Ξ±~(u)=Ξ±1β(u)I+Ο~(u)βAβDerkβ(A) with Ξ±1ββHomAβ(L~,A)
Let z:=(aD,u) and zβ²:=(bD,v). It follows
[TABLE]
[TABLE]
The Lie product on JβL is defined as follows:
[TABLE]
with
[TABLE]
from equation 3.6.1.
Here g(u,v)βZ2(L,A) is a 2-cocycle.
We get
[TABLE]
[TABLE]
It follows
[TABLE]
if and only if
[TABLE]
hence Ο is a map of k-Lie algebras if and only if
[TABLE]
and Ξ±1ββsβHomAβ(L,A).
Hence Ο is a map of k-Lie algebras if and only if there is an isomorphism of D-Lie algebras
[TABLE]
It follows there is an isomorphism L~β
L(fΞ±) of D-Lie algebras and the Theorem follows.
β
Corollary 3.10**.**
Let (L~iβ,Ξ±~iβ,Ο~iβ,[,],Diβ) be D-Lie algebras for i=1,2 with projective canonical quotients (Liβ,Ξ±iβ) for i=1,2.
There is an equality between the set of maps of D-Lie algebras Ο:L~1ββL~2β and the set of maps of A/k-Lie-Rinehart algebras
Οβ:L1ββL2β.
Proof.
By Theorem 3.9 there are isomorphisms L~iββ
L(fΞ±iβ) for i=1,2. It follows again from Theorem 3.9
that the set of maps of D-Lie algebras between L~1β and L~2β equal the set of maps of A/k-Lie-Rinehart algebras between L1β and L2β and the Corollary
is proved.
β
Example 3.11**.**
The D-Lie algebra associated to an A/k-Lie-Rinehart algebra.
Let (L,Ξ±) be an A/k-Lie-Rinehart algebra and let fΞ±βZ2(L,A) be the 2-cocycle associated to a 2-cocycle fβZ2(Derkβ(A),A). In [8], Theorem 2.8 we constructed a functor
[TABLE]
by
[TABLE]
where L(fΞ±):=AzβL is the abelian extension of L by the free rank one A-module on the symbol z.
Define for any D-Lie algebra (L~,Ξ±~,Ο~,[,],D) the following: G(L~,Ξ±~,Ο~,[,],D):=(L,ΟLβ) where (L,ΟLβ) is the canonical quotient of L~. Since any map of D-Lie algebras
Ο:L~βL~β² satisfies Ο(D)=Dβ² where Dβ²βL~β² is the canonical central element, we get a canonical map of A/k-Lie-Rinehart algebras
[TABLE]
where (Lβ²,ΟLβ²β) is the canonical quotient of L~β². Hence we get a functor
[TABLE]
Theorem 3.9 says that in the case when the canonical quotient L of L~ is a projective A-module, it follows G(F(L,Ξ±))β
(L,Ξ±) and F(G(L~))β
L~
are isomorphisms.
4. Classification of connections on D-Lie algebras with projective canonical quotient
In this section we classify connections on a D-Lie algebra L~ with projective canonical quotient (L,Ξ±). We prove in Theorem 4.15 there is a 2-cocycle fβZ2(Derkβ(A),A) and an equivalence of categories
[TABLE]
We use the equivalence Cfβ in 4.0.1 to classify arbitrary L~-connections in Corollary 4.16.
We also introduce the correspondence and Chow-operator of an L~-connection (E,Ο).
Lemma 4.1**.**
Let E be a left A-module and let Diff1(E) be the module of first order differential operators on E. It follows Diff1(E) is a left P-module and k-Lie algebra.
There is a map
[TABLE]
defined by
[TABLE]
where I is the identity operator. It follows Ο(β,ab)=aΟ(β,b)+Ο(β,a)b for all a,bβA and ββDiff1(E).
Proof.
A differential operator ββDiff1(E) is by definition an operator ββEndkβ(E) with [[β,aI]bI]=0 for all elements a,bβA where I is the identity operator.
The module of differential operators Diff1(E) has a left AβkβA-module structure defined by (aβb.β)(e):=aβ(be). It follows
[TABLE]
It follows da.db.β:=[[β,bI],aI]=0 hence for any element wβI2it follows wβ=0 and it follows Diff1(E) is a left P-module.
One checks the product
[TABLE]
defines a k-Lie algebra structure on Diff1(E). The rest is trivial and the Lemma follows.
β
Definition 4.2**.**
Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra and let E be a left A-module. An L~-connection Ο is a map
[TABLE]
of left P-modules. The curvature of a connection (Ο,E) is the map
[TABLE]
defined by
[TABLE]
Given two L~-connections (E,ΟEβ) and (F,ΟFβ), a map of L~-connections is a map of left A-modules
[TABLE]
with ΟFβ(u)βΟ=ΟβΟEβ(u) for all uβL~. Let Connβ(L~) denote the category of L~-connections and maps of connections. Let Connβ(L~,Id) denote the category
of L~ connections (Ο,E) with Ο(D)=IdEββEndAβ(E).
Note: It follows Connβ(L~,Id) is a full sub category of Connβ(L~).
Note: A connection Ο:L~βDiff1(E) in Connβ(L~,Id) is in particular an AβkβA-linear map with Ο(D)=IdEβ.
Example 4.3**.**
The degeneracy loci and correspondence associated to a connection.
A connection in the sense of Definition 4.2 is a map of left and right A-modules
[TABLE]
and we may associate to Ο several types of correspondences.
Definition 4.4**.**
Given a connection ΟβHomPβ(L~,Diff1(E)). Let I(Ο)βP:=AβkβA/I2 be the annihilator ideal
of the element Ο. Let ZPβ(Ο):=V(I(Ο))βSpec(P) be the correspondence of Ο
By definition I(Ο) is the set of elements in xβP with xΟ=0. The ideal I(Ο) gives rise to an ideal J(Ο)βAβkβA containing the square of the diagonal I2.
We get in a canonical way a correspondence Z(Ο):=V(J(Ο))βSpec(AβkβA):=XΓX where X:=Spec(A). Hence the connection Ο gives in a canonical way rise to a correspondence
Z(Ο) on X.
The left P-module Diff1(E) is projective as left and right A-module when A is a regular ring of finite type over a field and E a finite rank projective A-module. Hence when L~ is projective as left and right A-module it follows a connection
[TABLE]
is a map of left and right A-modules that are projective as left and right A-modules. Hence a connection is a geometric object and we may use Ο to define a correspondence on XΓX.
For a classical connection
[TABLE]
it is not immediate how to do this, since β is a map of k-vector spaces and not A-modules. We may view
an ordinary connection as an A-linear map
[TABLE]
where L is an A/k-Lie-Rinehart algebra satisfying β(x)(ae)=aβ(x)(e)+x(a)e and it is not immediate how to define a correspondence from β.
We may associate to Ο the degeneracy locies DΟlβ(Ο) and DΟrβ(Ο) where l and r refer to the degeneracy loci of Ο as a left and right A-linear map.
Here DΟlβ(Ο) and DΟrβ(Ο) are defined using local trivializations of the map Ο. Locally the map Ο is a matrix M with coefficients in a commutative ring B
and we may use minors of M of a given size to define an ideal in B associated to Ο as done in [2]. It is possible to do this in a way that is intrinsic and does not depend
of the choice of local trivialization of the map Ο. We get
for any two Ο1β,Ο2β a correspondence Z(Ο,Ο1β,Ο2β):=DΟlβ(Ο)ΓDΟrβ(Ο)βXΓX.
Hence we may associate different types of correspondences to the connection Ο. If X is smooth over k we get for each connection Ο and each correspondence Z(Ο,Ο1β,Ο2β)
an endomorphism
[TABLE]
defined by
[TABLE]
where p,q:XΓXβX are the projection maps, CHβ(X) is the Chow group of X and Z(Ο,Ο1β,Ο2β)β©Ξ±) is the intersection product. We get a similar
construction for any reasonable cohomology theory H(β) equipped with a cycle map. One would like to relate the correspondence Z(Ο,Ο1β,Ο2β) and operation I(Ο,Ο1β,Ο2β)
to the Chern classes of (E,Ο). Assume Ξ³:CHβ(XΓkβX)βHβ(XΓkβX) is a cycle map and let Ξ±βHβ(X) be a cohomology class.
we get an operator
[TABLE]
defined by
[TABLE]
Example 4.5**.**
Algebraic cycles and the Gauss-Manin connection.
If Hβ(β) is a Weil cohomology theory, there are Lefschetz operators
[TABLE]
defined for any i=0,β¦,dim(X) and the operator Ξ is conjectured to be induced by an algebraic cycle ZβXΓkβX. The cycle [Z] of the sub-scheme Z induce an operator
[TABLE]
defined by
[TABLE]
It has been conjectured that when XβPknβ is a smooth projective variety over an algebraically closed field, the Lefschetz operator Ξ is induced
by an operator on the form IHβ(Z) for some closed sub-scheme ZβXΓkβX. One wants to construct non-trivial cycle classes Ξ²βCHβ(XΓkβX) and calculate the operator IHβ(Ξ²).
The Chow group CHβ(XΓkβX) is hard to calculate and there are no general formulas for it. One also wants a construction of all Weil-cohomology theories H(β). If one could realize a Weil cohomology theory
Hβ(β) as the cohomology Hβ(L~,β) of a D-Lie algebra L~ as defined in [8], Definition 3.23, one could approach conjectures on algebraic cycles for smooth projective families of varieties.
One has to develop the formalism of the Gauss-Manin connection for the cohomology theory Hβ(L~,β) in the language of D-Lie algebras. In previous papers (see [7]) I have delevoped a formalism aimed at making explicit calculations of such connections. If one could realize the action of Ξ on a Weil cohomology Hβ(X) of the total space X of a smooth projective family Ο:XβS, as the
action of β(x) where x is a vector field on S and β the Gauss-Manin connection, this could be a first step in the direction of determining if Ξ is induced by an algebraic cycle.
The vector field x and the Gauss-Manin connection β are algebraic objects, hence we would get an algebraic construction of Ξ.
Definition 4.6**.**
Let Z(Ο,Ο1β,Ο2β) be the correspondence of Ο of type (Ο1β,Ο2β). Let I(Ο,Ο1β,Ο2β) be the
Chow-operator of Ο of type (Ο1β,Ο2β).
When using the notion of a D-Lie algebra L~, the derivation property
of the connection Ο:L~βDiff1(E) is encoded in the right A-linearity of the map Ο. Hence the correspondence Z(Ο,Ο1β,Ο2β) encodes properties of the left A-linearity of the map
Ο and the derivation property (the right A-linearity) of the map Ο. Hence the correspondence Z(Ο,Ο1β,Ο2β) and the endomorphism I(Ο,Ο1β,Ο2β)
dependes on the connection Ο is a non-trivial way. It is not clear how to make a similar definition with connections on the form 4.4.1 and 4.4.2 depending in a non-trivial way on the derivation property
of the connection. Hence in the case of an ordinary connection or a connection (E,β) on an A/k-Lie-Rinehart algebra L it is essential we work with the associated D-Lie algebra Ffβ(L):=L(fΞ±) and
L(fΞ±)-connection Cfβ(E,β) for some 2-cocycle fβZ2(Derkβ(A),A) if we want to define the correspondence and Chow-operator of (E,β).
Example 4.7**.**
Restricted Lie-Rinehart algebras, logarithmic derivaties and Picard groups.
Let B be a commutative ring over a field k characteristic p>0 and let (L,Ξ±) be a restricted B/k-Lie-Rinehart algebra. Let A:=ker(L) be the kernel of L
in B. The ring B is a purely inseparable Galois extension of A if B is a finitely generated and projective A-module and B[L]=HomAβ(B,B) in the sense of [17].
Yuan proves in [17] the existence of an exact sequence
[TABLE]
where Log(B/A) is the logarithmic derivative group of B/A, Hres2β(L,B) is the restricted Lie-Rinehart cohomology group of L with values in B,
Br(B/A) the Brauer group of B/A and Pic(B),Pic(A) the Picard groups of B and A. Hence in characteristic p>0, the restricted version of classical B/k-Lie-Rinehart cohomology calculates
Picard groups and Brauer groups. One wants to generalize the exact sequence 4.7.1 to the case of restricted D-Lie algebras.
Example 4.8**.**
D-Lie algebras, groupoid schemes and algebraic stacks.
Recall the following from [1], page 140: Let T:=(X,Y,s1β,s2β,t,p,i) be a groupoid scheme with X the scheme of arrows and Y the scheme of objects.
We say T is a schematic equivalence relation if the map (s1β,s2β):XΓXβYΓY is a monomorphism. There is the following result:
Proposition 4.9**.**
Assume Y is smooth over a field k of characteristic zero. There is a one-to-one correspondence between locally trivial sheaves of OYβ/k-Lie-Rinehart algebras
(L,Ξ±) and formal infinitesimal groupoid schemes with Y as a scheme of objects. Ynder this correspondence schematic equivalence relations corresponds
to sub-sheaves of OYβ/k-Lie-Rinehart algebras of the tangent sheaf TY/kβ.
Proof.
See [5], VIII.1.1.5.
β
Hence there is a close connection between A/k-Lie-Rinehart algebras, moduli spaces and algebraic stacks. If FβTY/kβ is a locally trivial finite rank sub OYβ-module and
sheaf of k-Lie algebras, and the corresponding equivalence relation RβYΓkβY is an etale equivalence relation, we may form the stack quotient [Y/F]. The quotient
[Y/F] is a Deligne-Mumford stack. The sheaf of universal enveloping algebras U(OYβ,F) is in a natural way a sheaf of rings on [Y/F] - the structure sheaf O[Y/F]β:=U(OYβ,F) of the stack [Y/F]. A flat F-connection (E,β) which is a quasi coherent sheaf of OYβ-modules, is in a natural way a quasi coherent sheaf of O[Y/F]β-modules.
Any Deligne-Mumford stack [Y/R] arise by Proposition 4.9 from an involutive sub-bundle FβTY/kβ. Hence if we are given the scheme of objects Y of a DM-stack we may construct the equivalence relation R on Y using a bundle on the form of F. The structure sheaf O[Y/R]β is a sheaf of filtered almost commutative rings, hence the βringed spaceβ
([Y/R],O[Y/R]β) may be viewed as a non-commutative ringed space. Hence connections arise naturally when studying algebraic stacks, quasi coherent sheaves on algebraic stacks and
non-commutative ringed spaces.
Note: Associated to any D-Lie algebra L~ which is of finite rank and projective as left A-module, we get a formal groupoid scheme over Y:=Spec(A) using the underlying A/k-Lie-Rinehart algebra of L~
and Proposition 4.9.
Example 4.10**.**
L-functions and cristalline cohomology.
Let S:=Spec(OKβ) where K is an algebraic number field and let f:XβS be a regular scheme of finite type over Sof dimension d. There is an equality of L-functions
[TABLE]
where L(X,s) is the global L-function of the scheme X and L(X(p),s) is the L-function of the fiber X(p):=fβ1(p) for a closed point pβS. The residue field ΞΊ(p) is a finite field
of characteristic q>0 and the fiber X(p) is a scheme of finite type over ΞΊ(p). Hence there is an equality
[TABLE]
where Z(X(p),t) is the Weil zeta function of X(p). By the work of Kedlaya (see [6]) it follows the Weil conjectures can be proved using a p-adic cohomology theory Hβ(β). In fact Kedlaya has proved the Weil conjectures for an arbitrary scheme X of finite type over a finite field with no condition on smoothness or projectivity on X. If the Ext-group Extβ(V,W) of two connections V,W calculate cristalline cohomology (see [5]), it may be we can use Extβ(V,W) to prove the Weil conjectures for any scheme X of finite type over a finite field. If the Weil zeta function Z(X(p),t) can be calculated using the Ext group Extβ(V,W), it may be we can use such a description in the study of the global L-function L(X,s) via the product formula in 4.10.1. The Ext-group is defined in complete generality and may be defined for connections on the family X. One has to calculate explicit examples to check if this idea leads to interesting constructions and results. If X is a regular scheme the following is conjectured in [12]:
[TABLE]
Here
[TABLE]
is the K-theoretic Euler characteristic of X as defined in [12]. The conjecture in 4.10.3 is referred to as a conjecture due to Lichtenbaum, Deligne, Bloch, Beilinson and others
in Wiles official problem description [16] for the BSD-conjecture. If it can be proved the Ext-group Extβ(V,W) calculate cristalline cohomology of the fibers X(p) for all primes pξ =(0), it might be the group Extβ(V,W) can be used in the study of Conjecture 4.10.3.
Lemma 4.11**.**
The following holds for an L~-connection Ο:L~βDiff1(E):
[TABLE]
for all elements aβA,uβL~ and eβE. It follows Ο(D)βEndAβ(E). The curvature RΟβ defines a map
[TABLE]
Assume P:L~βEndAβ(E) is an P-linear map. It follows Οβ²:=Ο+P is an L~-connection. If P(D)=0 it follows Οβ²(D)=IdEβ.
Proof.
Since Ο is P-linear it follows Ο is AβkβA-linear. It follows Ο(au)=aΟ(u) and Ο(ua)=Ο(u)a. We get
[TABLE]
Since Da=aD it follows Ο(D)βEndAβ(E). The second statement holds since the element [Ο(u),Ο(v)]:=Ο(u)Ο(v)βΟ(v)Ο(u)βDiff1(E).
Since Ο is P-linear it follows Οβ²:=Ο+P is P-linear. If P(D)=0 it follows Οβ²(D)=Ο(D)+P(D)=IdEβ.
The Lemma follows.
β
Hence the notion of an L~-connection introduced in Definition 4.2 agrees with the notion introduced in the paper [8].
Example 4.12**.**
The L(fΞ±)-connection associated to an (L,Ο)-connection.
Definition 4.13**.**
Let (L,Ξ±) an A/k-Lie-Rinehart algebra and let β:LβEndkβ(E) be an (L,Ο)-connection where ΟβEndAβ(E). This means
[TABLE]
for all aβA,eβE and xβL. Let Connβ(L,End) denote the category of (L,Ο)-connections and morphisms. The endomorphism
Ο may vary. Let Connβ(L) denote the category of ordinary L-connections and morphisms of L-connections.
Note: If β:LβEndkβ(E) is an ordinary connection and ΟβEndAβ(E) it follows ββΟ is an (L,Ο)-connection.
Recall the following construction:
Let fβZ2(Derkβ(A),A) be a 2-cocycle and let Ffβ(L):=(L(fΞ±),Ξ±fβ,Οfβ,[,],z) be the D-Lie algebra associated to L and f. Define the following map:
[TABLE]
by
[TABLE]
Let Cfβ(E,β):=(E,Οββ).
Let (F,ββ²) be an (L,Οβ²)-connection with Οβ²βEndAβ(E) and let Ο:(E,β)β(F,ββ²) be a map of connections.
Define the following map
[TABLE]
It follows the map Cfβ(Ο):(E,Οββ)β(F,Οββ²β) is a map of L(fΞ±)-connections.
Let i:LβL(fΞ±) be the canonical left A-linear map and let (E,Ο) be an L(fΞ±)-connection:
[TABLE]
Let βΟβ:=Οβi. It follows
[TABLE]
is an (L,Ο)-connection. Let Rfβ(E,Ο):=(E,βΟβ).
Lemma 4.14**.**
Let (L,Ξ±) be an A/k-Lie-Rinehart algebra and let (E,β) be an L-connection. Let (F,Ο) be an L(fΞ±)-connection
The construction above define functors
[TABLE]
by
[TABLE]
and
[TABLE]
by
[TABLE]
It follows CfββRLβ=Id and RfββCfβ=Id hence Cfβ and Rfβ are equivalences of categories for any 2-cocycle fβZ2(Derkβ(A),A)
Proof.
The proof follows immediately from the constructions above.
β
Theorem 4.15**.**
Let (L,Ξ±) be an A/k-Lie-Rinehart algebra and let fβZ2(Derkβ(A),A) be a 2-cocycle. Lemma 4.14 gives an equivalence of categories
[TABLE]
from the category Connβ(L,End) of (L,Ο)-connections, to the category Connβ(L(fΞ±)) of L(fΞ±)-connections Ο for any 2-cocycle f.
Let u:=az+x,v:=bz+yβL(fΞ±) and let eβE. Let (E,β) be an L-connection and let Οββ:=Cfβ(β). The following holds:
[TABLE]
For any L(fΞ±)-connection (E,Ο) there is an (L,Ο)-connection (E,β) with Cfβ(E,β)=(E,Ο). If Ο(z)=Id we may chose (E,β)βConnβ(L,Id).
Proof.
One checks for any (L,Ο)-connection (E,β) the corresponding map Οββ is a map
[TABLE]
of left P-modules. Moreover for any map of (L,Ο)-connections Ο:(E,β)β(F,ββ²) it follows the map
[TABLE]
is a map of L(fΞ±)-connections with Cfβ(ΟβΟ)=Cfβ(Ο)βCfβ(Ο). By definition CfββRfβ=Id and RfββCfβ=Id and the first claim follows.
The statement on the curvature follows from Lemma 2.20 in [8].
In particular given an L(fΞ±)-connection (E,Ο) it follows (E,Ο)β
Cfβ(Rfβ(E,Ο)) and Rfβ(E,Ο)βConnβ(L,End).
The Theorem follows.
β
We may classify L~-connections in terms of (L,Ο)-connections in the case when the canonical quotient L of L~ is a projective A-module.
Corollary 4.16**.**
Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra and let (E,Ο) be an L~-connection. Assume the canonical quotient L of L~
is a projective A-module. It follows any L~-connection (E,Ο) is on the form Cfβ(E,β) where fβZ2(Derkβ(A),A) and (E,β) is an (L,Ο)-connection for some
ΟβEndAβ(E). If Ο(D)=I it follows there is an L-connection (E,β) with Cfβ(E,β)=(E,Ο).
Proof.
By Theorem 3.9 there is an isomorphism L~β
L(fΞ±) for fβZ2(Derkβ(A),A). The Corollary now follows from Theorem 4.15.
β
Example 4.17**.**
Non-abelian extensions of D-Lie algebras.
Definition 4.18**.**
Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra and let (Ο,E) be an L~-connection with Ο(D)=IdEβ. Let
(End(L~,E),Ξ±Eβ,ΟEβ,[,],D~) be the D-Lie algebra constructed in Section 4 in [8].
The following Theorem generalize Theorem 2.14 from [9] from a connection on an A/k-Lie-Rinehart algebra to the case of a connection on a D-Lie algebra:
Theorem 4.19**.**
Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra and let (Ο,E) be an L~-connection with Ο(D)=IdEβ. Let End(L~,E) be the non-abelian extension
of L~ with Ο from Definition 4.18. There is a canonical flat connection
[TABLE]
defined by Ο1(Ο,u):=Ο+Ο(u). The exact sequence
[TABLE]
is split in the category of D-Lie algebras if and only if E has a flat L~-connection.
Proof.
By Proposition 4.9 in [8] we get an extensins of D-Lie algebras
[TABLE]
where End(L~,E):=EndAβ(E)βL~ with D~:=(0,D) and
pEβ:End(L~,E)βL~ defined by pEβ(Ο,u):=uβL~. Define c(Ο,u):=(cΟ,cu) and (Ο,u)c:=(Οc,uc) and define
[TABLE]
by
[TABLE]
and
[TABLE]
by
[TABLE]
Define moreover for any z:=(Ο,u),w:=(Ο,v)βEnd(L~,E)
[TABLE]
It follows from Proposition 4.9 in [8] that End(L~,E) is an extension of L~ by the A-Lie algebra EndAβ(E). Asssume s:L~βEnd(L~,E) is a section
of the map pEβ. Hence pEββs=IdL~β and s is a map of D-Lie algebras. It follows s is P-linear, a map of k-Lie algebras and s(D)=(0,D). It follows
s(u)=(P(u),u) where P:L~βEndAβ(E) is a P-linear map with P(D)=0. It follows from Lemma 4.11 the map Οβ²:=Ο+P is an L~-connection
Οβ²:L~βDiff1(E). One checks that the map s is a map of k-Lie algebras if and only if RΟβ²β=0, hence the sequence 4.19.1 is split in the category of D-Lie algebras
if and only if E has a flat L~-connection. The Theorem is proved.
β
Note: In [9], Theorem 2.4 a result similar to Theorem 4.19 is proved for A/k-Lie-Rinehart algebras. Note moreover that by Theorem 4.19 we may view any L~-connection
(Ο,E) as a representation of the k-Lie algebra End(L~,E). Since the induced connection Ο! is flat, it follows Ο! is a map of k-Lie algebras.
Corollary 4.20**.**
Let (L~,Ξ±~,Ο~,[,],D) be a D-Lie algebra and let (Ο,E) be an L~-connection with Ο(D)=IdEβ. It follows the canonical quotient of End(L~,E) is isomorphic to the
A/k-Lie-Rinehart algebra End(L,E) where (L,ΟLβ) is the canonical quotient of L~. If L is projective there is an isomorphism End(L~,E)β
End(L(fΞ±),E),
where L(fΞ±)=FfΞ±β(L) and FfΞ±β is the functor from Example 3.3.
Proof.
The proof follows from Theorem 4.19 and Theorem 3.9 since End(L~,E):=EndAβ(E)βL~ and D~:=(0,D).
β