betaFeedback

WorksMathematics

AVAV-modules on affine algebraic groups

Abstract

We study AVAV-modules on an affine algebraic group GG, that is, modules that admit compatible actions of the algebra of regular functions on GG and the Lie algebra of vector fields. A central role in this theory is played by the Lie algebra of jets of vector fields. Our main result is the Decomposition Theorem for the Lie algebra of jets of vector fields. We apply the Decomposition Theorem to describe differentiable AVAV-modules on GG.

Introduction

The theory of DD-modules on algebraic groups has been instrumental in the proof of the Kazhdan-Lusztig conjecture by Brylinski-Kashiwara and Beilinson-Bernstein [11, 3, 15]. In this paper, we discuss AVAV-modules on algebraic groups, which are natural generalizations of DD-modules. Geometric objects like tensor fields do not form a DD-module under the Lie derivative action, but belong to the category of AVAV-modules. The theory of AVAV-modules was successfully used in establishing classifications of simple Harish-Chandra modules for Lie algebras of vector fields on a torus [4], and on the affine space [13, 18].

An AVAV-module on a connected affine algebraic group GG is a module that admits an action of the algebra AA of regular functions on GG, and an action of the Lie algebra VV of vector fields, with the Leibniz rule as a compatibility condition (see Definition 1 in the next Section). The pair of algebras (A,V)(A,V) forms a Rinehart pair [16], and the category of AVAV-modules is governed by the weak enveloping algebra Uweak(A,V)U_{\mathrm{weak}}(A,V), which is the smash product of the Hopf algebra U(V)U(V) with its module AA: Uweak(A,V)=A#U(V)U_{\mathrm{weak}}(A,V)=A\#U(V). This algebra was first introduced by Jacobson in [14]. This associative algebra has a Lie subalgebra A#VA\#V, the smash product of the Lie algebra VV with a commutative algebra AA, on which VV acts by derivations. The Lie algebra A#VA\#V has a geometric interpretation as the Lie algebra of algebraic jets of vector fields, and its completion, A#V^A\#\widehat{V}, is the Lie algebra of ∞\infty-jets of vector fields.

The Lie algebra A#V^A\#\widehat{V} has an interesting structure, and is important for understanding the theory of AVAV-modules. The central result of the paper is the Decomposition Theorem (Theorem 10 in § 4), which states that on an affine algebraic

2020 Mathematics Subject Classification. Primary 17B66; Secondary 13N15. group GG, there exists a semidirect product decomposition

A#V^≅V⋉(A⊗L+),\widehat{A\# V} \cong V \ltimes(A \otimes\mathcal{L}_+),

where L+\mathcal{L}_+ is the Lie algebra of vector fields on GG, vanishing at the identity element ee. We point out that this is a non-trivial isomorphism – in the semidirect product, neither VV is mapped to the Lie algebra of vector fields 1#V⊂A#V^1\# V \subset\widehat{A\# V}, nor L+\mathcal{L}_+ is identified with a subalgebra in 1#V1\# V. Previously, such a Decomposition Theorem was known for affine space [18, 8], the torus, and only locally for an affine variety [7].

In order to prove the Decomposition Theorem, we generalize the classical Taylor formula

f(x)=∑k=0∞1k!dkfdxk(a)(x−a)kf(x) = \sum_{k=0}^{\infty} \frac{1}{k!} \frac{d^k f}{dx^k}(a)(x-a)^k

to the setting of functions on algebraic groups (Theorem 7 in § 3). We replace the family {(ddx)k}\left\{\left(\frac{d}{dx}\right)^k\right\} with a basis of the universal enveloping algebra U(gL)U(\mathfrak{g}^{L}) of the Lie algebra gL\mathfrak{g}^{L} of left-invariant vector fields, and view {(x−a)k}\left\{(x-a)^k\right\} as a basis for the algebra of GG-invariants in A⊗AA\otimes A.

The most important class of AVAV-modules is the category of differentiable AVAV-modules. An AVAV-module MM is called NN-differentiable if it is annihilated by the elements

∑k=0∞(−1)k(Nk)fN−k#fkτ\sum_{k=0}^{\infty}(-1)^k\binom{N}{k}f^{N-k}\# f^k\tau

for all f∈A,τ∈Vf\in A,\tau\in V.

DD-modules are precisely 11-differentiable AVAV-modules, and all tensor modules are 22-differentiable. Using the Decomposition Theorem, we prove that NN-differentiable AVAV-modules on GG are modules for the associative algebra A#U(g⋉(A⊗(L+/meN−1L+)))A\#U(\mathfrak{g}\ltimes(A\otimes(\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+))), where g\mathfrak{g} is the Lie algebra of GG, and me\mathfrak{m}_e is the ideal of functions on GG vanishing at ee.

Our description is more detailed for 22-differentiable AVAV-modules on GG. Such a module is given by a DD-module structure on MM, together with an action of gln\mathfrak{gl}_n, n=dim⁡Gn=\dim G, which commutes with the action of DD. In Theorem 17, we give an explicit formula for this correspondence.

In the final section of this paper, we point out that the Lax equation emerges in the context of 22-differentiable AVAV-modules, connecting them to the theory of integrable dynamical systems.

The paper is organized as follows. In § 2, we outline the basic definitions related to AVAV-modules, Rinehart pairs, and jets of vector fields. We generalize the Taylor formula to the context of affine algebraic groups in § 3. In § 4, we establish the Decomposition Theorem, the main result of the paper. In § 5, we present examples of AVAV-modules, focusing on 22-differentiable modules.

Acknowledgements. Both authors gratefully acknowledge support from the Natural Sciences and Engineering Research Council of Canada through Discovery Grants. We benefited from helpful discussions with ChatGPT.

AVA V-modules and jets of vector fields

Let AA be a commutative associative algebra over a field KK of characteristic zero. Assume that the Lie algebra VV acts on AA by derivations.

Definition 1. An AVA V-module MM is a module for AA and also a module for the Lie algebra VV, with two actions compatible via the Leibniz rule:

τ(fm)=τ(f)m+f(τm)for τ∈V, f∈A, m∈M.(1)\tau(fm)=\tau(f)m+f(\tau m) \qquad\text{for }\tau\in V,\ f\in A,\ m\in M. \tag*{(1)}

Just as the universal enveloping algebra U(L)U(L) controls the category of modules for Lie algebra LL, there is an associative algebra that controls the category of AVA V-modules. The weak enveloping algebra Uweak(A,V)=A#U(V)U_{\mathrm{weak}}(A,V)=A\mathbin{\#}U(V) is the smash product of the Hopf algebra U(V)U(V) with its module AA [14]. As a vector space, it is the tensor product A⊗KU(V)A\otimes_K U(V) with multiplication

(f⊗u)(h⊗v)=∑ifui1(h)⊗ui2v,(f\otimes u)(h\otimes v)=\sum_i f u_i^1(h)\otimes u_i^2v,

where u↦∑iui1⊗ui2u\mapsto\sum_i u_i^1\otimes u_i^2 is the coproduct written in Sweedler notations. An AVA V-module is the same as a module over the associative algebra A#U(V)A\mathbin{\#}U(V).

It turns out that the subspace A#VA\mathbin{\#}V is a Lie subalgebra of A#U(V)A\mathbin{\#}U(V).

Definition 2. Let AA be a commutative associative algebra on which the Lie algebra VV acts by derivations. The smash product A#VA\mathbin{\#}V is a Lie algebra whose underlying space is A⊗VA\otimes V and the Lie bracket is given by

[f#τ,h#η]=fτ(h)#η−hη(f)#τ+fh#[τ,η]for f,h∈A, τ,η∈V.(2)[f\mathbin{\#}\tau,h\mathbin{\#}\eta]=f\tau(h)\mathbin{\#}\eta-h\eta(f)\mathbin{\#}\tau+fh\mathbin{\#}[\tau,\eta]\qquad\text{for }f,h\in A,\ \tau,\eta\in V. \tag*{(2)}

We are particularly interested in the case where AA is the algebra of regular functions on a smooth irreducible affine algebraic variety XX and V=Der⁡(A)V=\operatorname{Der}(A). In this setting, VV is the Lie algebra of vector fields on XX. Such a pair (A,V)(A,V) is an example of a Rinehart pair [16]:

Definition 3. Let AA be a commutative associative algebra, and let VV be a Lie algebra. The pair (A,V)(A,V) is a Rinehart pair if VV acts on AA by derivations, and VV is an AA-module, satisfying:

[τ,fη]=τ(f)η+f[τ,η],[\tau,f\eta]=\tau(f)\eta+f[\tau,\eta],
(fη)(h)=f(η(h))for f,h∈A, τ,η∈V.(f\eta)(h)=f(\eta(h))\qquad\text{for }f,h\in A,\ \tau,\eta\in V.

There are two natural categories of modules for a Rinehart pair: AVAV-modules, also known as weak modules, and DD-modules, also known as strong modules, satisfying (1) as well as the following axiom:

(fτ)m=f(τm),for f∈A, τ∈V, m∈M.(3)(f\tau)m=f(\tau m),\quad\text{for }f\in A,\ \tau\in V,\ m\in M. \tag*{(3)}

The algebra AA itself is a DD-module for a Rinehart pair (A,V)(A,V), while VV is an AVAV-module, but not a DD-module in general.

The category of DD-modules is controlled by the strong enveloping algebra Ustrong(A,V)U_{\mathrm{strong}}(A,V) [16], which is the quotient

A#U(V)/⟨fh#τ−f#hτ∣f,h∈A, τ∈V⟩A\#U(V)\big/\langle fh\#\tau-f\#h\tau\mid f,h\in A,\ \tau\in V\rangle

When AA is the algebra of regular functions on a smooth irreducible affine variety XX, and V=Der⁡(A)V=\operatorname{Der}(A), the strong enveloping algebra Ustrong(A,V)U_{\mathrm{strong}}(A,V) is the algebra DD of differential operators on XX [2]. As we shall see below, in this context, the Lie algebra A#VA\#V may be interpreted as the Lie algebra of algebraic jets of vector fields on XX.

If VV acts on AA by derivations, then the pair (A,A#V)(A,A\#V) is a Rinehart pair. Moreover,

Ustrong(A,A#V)=Uweak(A,V).U_{\mathrm{strong}}(A,A\#V)=U_{\mathrm{weak}}(A,V).

Let Δ⊂A⊗A\Delta\subset A\otimes A be the kernel of the multiplication map μ:A⊗A→A\mu:A\otimes A\to A. The quotient A⊗A/ΔN+1A\otimes A/\Delta^{N+1} is the algebra of NN-jets of functions on XX, and the completion

A⊗^A=lim⁡⟵NA⊗A/ΔNA\widehat{\otimes}A=\lim_{\longleftarrow_N} A\otimes A/\Delta^N

is the algebra of ∞\infty-jets of functions (see Chapter 10 in [1] for the basics on the completions of algebras).

The space ΔN⊗AV\Delta^N\otimes_A V is an ideal in the Lie algebra A#VA\#V, and the quotient A#V/ΔN+1⊗AVA\#V/\Delta^{N+1}\otimes_A V is the Lie algebra of NN-jets of vector fields on XX. The completion

A#^V=lim⁡⟵N(A#V)/(ΔN⊗AV)A\widehat{\#}V=\lim_{\longleftarrow_N}(A\#V)/(\Delta^N\otimes_A V)

is the Lie algebra of jets of vector fields on XX [7]. Since ⋂N=1∞ΔN=(0)\bigcap_{N=1}^{\infty}\Delta^N=(0), the Lie algebra A#VA\#V injects into the Lie algebra of ∞\infty-jets of vector fields. Thus, A#VA\#V may be interpreted as the Lie algebra of algebraic jets of vector fields.

We now turn to the principal setting of this paper, the case when the underlying variety is a connected affine algebraic group, X=GX=G.

There are three natural actions of GG on itself: left translation: Lg(h)=ghL_g(h)=gh, right translation: Rg(h)=hg−1R_g(h)=hg^{-1}, and the action by conjugation: Cg(h)=ghg−1C_g(h)=ghg^{-1}, with g,h∈Gg,h\in G. Note that the left and right actions of GG commute. These induce three actions of GG on the algebra AA of regular functions on GG: Lg∗(f)=f∘Lg−1L_g^*(f)=f\circ L_{g^{-1}}, Rg∗(f)=f∘Rg−1R_g^*(f)=f\circ R_{g^{-1}}, Cg∗(f)=f∘Cg−1C_g^*(f)=f\circ C_{g^{-1}} for f∈Af\in A. Likewise, there are three actions of GG on V=Der⁡AV = \operatorname{Der} A: Lg∘τ=Lg∗∘τ∘Lg−1∗L_g^\circ\tau= L_g^* \circ\tau\circ L_{g^{-1}}^*, etc. A vector field τ∈V\tau\in V is called left-invariant if Lg∘τ=τL_g^\circ\tau= \tau for all g∈Gg \in G.

Let g\mathfrak{g} be the tangent space to GG at the identity ee, which may be identified with the space of derivations A→KA \to K at the identity: x(fh)=x(f)h(e)+f(e)x(h)x(fh) = x(f)h(e) + f(e)x(h), for x∈gx \in\mathfrak{g}, f,h∈Af,h \in A. We construct a left-invariant vector field xLx^L which restricts to xx at the identity: xL(f)(g)=x(f∘Lg)x^L(f)(g) = x(f \circ L_g). Likewise, we construct a right-invariant vector field xRx^R, which restricts to −x-x at the identity: xR(f)(g)=−x(f∘Rg−1)x^R(f)(g) = -x(f \circ R_{g^{-1}}). The negative sign is needed in order for the two constructions to induce the same Lie bracket on g\mathfrak{g}. We get two isomorphic subalgebras gL\mathfrak{g}^L and gR\mathfrak{g}^R of left- and right-invariant vector fields in VV. Alternatively, xLx^L and xRx^R may be expressed using the coproduct A→A⊗AA \to A \otimes A, f↦∑ifi1⊗fi2f \mapsto\sum_i f_i^1 \otimes f_i^2. Then xL(f)=∑ifi1x(fi2)x^L(f) = \sum_i f_i^1 x(f_i^2), xR(f)=−∑ix(fi1)fi2x^R(f) = -\sum_i x(f_i^1)f_i^2.

Note that the elements of gL\mathfrak{g}^L are the differentials of the right action of GG on AA, and gR\mathfrak{g}^R are the differentials of the left action.

There is also a homomorphism g→V\mathfrak{g} \to V induced by the conjugation action, x↦xC=xL+xRx \mapsto x^C = x^L + x^R. The kernel of this homomorphism is Z(g)Z(\mathfrak{g}). Since xCx^C evaluates to zero at the identity, it preserves the maximal ideal me\mathfrak{m}_e.

The Lie algebra of vector fields on an affine algebraic group GG is a free AA-module generated by either gL\mathfrak{g}^L or gR\mathfrak{g}^R [5]:

V=A⊗gL=A⊗gR.V = A \otimes\mathfrak{g}^L = A \otimes\mathfrak{g}^R.

In fact, these tensor products are smash products of AA with gL\mathfrak{g}^L and gR\mathfrak{g}^R. For the Lie algebra of algebraic jets of vector fields, we have the double smash product construction:

A#V=A#(A#gL).A \# V = A \# (A \# \mathfrak{g}^L).

In this setting, Lie bracket takes the following form:

[p⊗xL,q⊗yL]=(μ(p)⊗1)(xL⊗1)(q)⊗yL+p(1⊗xL)(q)⊗yL−(μ(q)⊗1)(yL⊗1)(p)⊗xL−q(1⊗yL)(p)⊗xL+pq⊗[x,y]L,for p,q∈A⊗A,x,y∈g.(4)\begin{aligned} [p \otimes x^L, q \otimes y^L] ={}& (\mu(p) \otimes1)(x^L \otimes1)(q) \otimes y^L + p(1 \otimes x^L)(q) \otimes y^L \\ &- (\mu(q) \otimes1)(y^L \otimes1)(p) \otimes x^L - q(1 \otimes y^L)(p) \otimes x^L \\ &+ pq \otimes[x,y]^L,\quad\text{for } p,q \in A \otimes A, x,y \in\mathfrak{g}. \tag*{(4)} \end{aligned}

Taylor formula for functions on algebraic groups

In this section, we will establish the Taylor formula for functions on algebraic groups.

The algebra A⊗AA \otimes A of functions on G×GG \times G is a GG-module under the diagonal left action of GG. Denote by (A⊗A)L(A \otimes A)^L the subalgebra of GG-invariants under this action. Since the differential of the left action of GG on AA yields the action of the Lie algebra of right-invariant vector fields gR\mathfrak{g}^R, we conclude that (A⊗A)L(A \otimes A)^L is the kernel of the action of gR\mathfrak{g}^R on the tensor product A⊗AA \otimes A.

Lemma 4. (A⊗A)L≅A(A \otimes A)^L \cong A as algebras.

Proof. The isomorphism maps φ:(A⊗A)L→A\varphi: (A \otimes A)^L \to A and ψ:A→(A⊗A)L\psi: A \to(A \otimes A)^L are given by

φ(∑ifi⊗hi)(g)=∑ifi(e)hi(g)andψ(f)(g1,g2)=f(g1−1g2).\varphi\left(\sum_i f_i \otimes h_i\right)(g) = \sum_i f_i(e)h_i(g) \quad\text{and} \quad\psi(f)(g_1,g_2) = f(g_1^{-1}g_2).

It is easy to see that the image of ψ\psi is consists of left-invariant functions, and that these maps are the inverses of each other:

φ(ψ(f))(g)=ψ(f)(e,g)=f(e−1g)=f(g),\varphi(\psi(f))(g) = \psi(f)(e,g) = f(e^{-1}g) = f(g),
ψ(φ(∑ifi⊗hi))(g1,g2)=φ(∑ifi⊗hi)(g1−1g2)=∑ifi(e)hi(g1−1g2)=(∑ifi⊗hi)(g1,g2),\begin{aligned} \psi\left(\varphi\left(\sum_i f_i \otimes h_i\right)\right)(g_1,g_2) &= \varphi\left(\sum_i f_i \otimes h_i\right)(g_1^{-1}g_2) \\ &= \sum_i f_i(e)h_i(g_1^{-1}g_2) = \left(\sum_i f_i \otimes h_i\right)(g_1,g_2), \end{aligned}

where in the last equality follows from left-invariance.

Since the left and right actions of GG commute, (A⊗A)L(A \otimes A)^L is a GG-module under the actions R∗⊗id⁡R^* \otimes\operatorname{id}, id⁡⊗R∗\operatorname{id} \otimes R^*, and R∗⊗R∗R^* \otimes R^*. It is straightforward to check the following claims:

Lemma 5. Under the isomorphism (A⊗A)L≅A(A \otimes A)^L \cong A, the R∗⊗id⁡R^* \otimes\operatorname{id} action corresponds to the left GG-action L∗L^* on AA, id⁡⊗R∗\operatorname{id} \otimes R^* corresponds to the right GG-action R∗R^*, and R∗⊗R∗R^* \otimes R^* corresponds to the conjugation action C∗C^* of GG on AA.

Corollary 6. Under the isomorphism φ:(A⊗A)L→A\varphi: (A \otimes A)^L \to A, (i) φ∘(xL⊗1)=xR∘φ\varphi\circ(x^L \otimes1) = x^R \circ\varphi, (ii) φ∘(1⊗xL)=xL∘φ\varphi\circ(1 \otimes x^L) = x^L \circ\varphi, (iii) φ∘(xL⊗1+1⊗xL)=xC∘φ\varphi\circ(x^L \otimes1 + 1 \otimes x^L) = x^C \circ\varphi, for x∈gx \in\mathfrak{g}.

Proof. This follows from the fact that left-invariant vector fields are the differentials of the R∗R^*-action, and right-invariant vector fields are the differentials of the L∗L^*-action of GG.

Consider the pairing U(gL)×A→KU(\mathfrak{g}^L) \times A \to K, given by applying the differential operator u∈U(gL)u \in U(\mathfrak{g}^L) to function f∈Af \in A and evaluating the result at the identity element of GG. Let U0⊂U1⊂U2⊂⋯U_0 \subset U_1 \subset U_2 \subset\cdots be the standard augmentation filtration on U(gL)U(\mathfrak{g}^L). Here U0=K⋅1U_0 = K \cdot1, U1=K⋅1+gLU_1 = K \cdot1 + \mathfrak{g}^L, etc. Consider also the decreasing filtration on AA by powers of the maximal ideal me\mathfrak{m}_e: A⊃me⊃me2⊃⋯A \supset\mathfrak{m}_e \supset\mathfrak{m}_e^2 \supset\cdots. Since Un(gL)U_n(\mathfrak{g}^L) acts on AA by differential operators of order less or equal to nn, it trivially pairs to men+1\mathfrak{m}_e^{n+1}, and we get a pairing of finite-dimensional spaces:

Un(gL)×A/men+1→K.U_n(\mathfrak{g}^L) \times A/\mathfrak{m}_e^{n+1} \to K.

By Cartier’s Theorem ([12] II.6.1), this pairing is non-degenerate, and we can choose dual bases with respect to this pairing. By increasing the value of nn, inductively, we construct dual bases {ui}\{u_i\} in U(gL)U(\mathfrak{g}^L) and {Fi}\{F_i\} in AA.

Recall the classical Taylor formula:

f(x)=∑k=0∞1k!dkfdxk(a)(x−a)k.f(x)=\sum_{k=0}^{\infty}\frac{1}{k!}\frac{d^{k}f}{dx^{k}}(a)(x-a)^{k}.

We view both sides as functions on R×R\mathbb{R}\times\mathbb{R}, where the first copy is parametrized by aa and the second copy by xx. Note that the expression x−a∈A⊗Ax-a\in A\otimes A is translation invariant (under the diagonal action of the additive group R\mathbb{R} on R×R\mathbb{R}\times\mathbb{R}). Thus, (x−a)k∈(A⊗A)L(x-a)^{k}\in(A\otimes A)^{L}. The Taylor formula yields the statement 1⊗A⊂A⊗^(A⊗A)L1\otimes A\subset A\widehat{\otimes}(A\otimes A)^{L}. The following theorem generalizes this to an arbitrary affine algebraic group.

Theorem 7 (Taylor formula). Let AA be the algebra of regular functions on an affine algebraic group GG over a field KK of characteristic zero. Then

A⊗^A≅A⊗^(A⊗A)L,A\widehat{\otimes}A\cong A\widehat{\otimes}(A\otimes A)^{L},

given by the Taylor formula

f(g1,g2)=∑i((1⊗ui)f)(g1,g1)Fi(g1−1g2).f(g_{1},g_{2})=\sum_{i}((1\otimes u_{i})f)(g_{1},g_{1})F_{i}(g_{1}^{-1}g_{2}).

The completion in the left hand side is

A⊗^A=lim⁡⟵NA⊗A/ΔN,A\widehat{\otimes}A=\lim_{\longleftarrow N} A\otimes A/\Delta^{N},

and the completion in the right hand side is

A⊗^(A⊗A)L=lim⁡⟵NA⊗(A⊗A)L/A⊗ψ(meN).A\widehat{\otimes}(A\otimes A)^{L}=\lim_{\longleftarrow N} A\otimes(A\otimes A)^{L}/A\otimes\psi(\mathfrak{m}_{e}^{N}).

Applying the isomorphism (A⊗A)L≅A(A\otimes A)^{L}\cong A, we obtain a non-trivial isomorphism

A⊗^ΔA≅A⊗^1⊗meA.(5)A\widehat{\otimes}_{\Delta}A\cong A\widehat{\otimes}_{1\otimes\mathfrak{m}_{e}}A. \tag*{(5)}

The above maps are isomorphisms of AA-modules with respect to the action of AA on the left tensor factors.

Proof. First, let us establish the formula

f(g2)=∑iui(f)(g1)Fi(g1−1g2).(6)f(g_{2})=\sum_{i}u_{i}(f)(g_{1})F_{i}(g_{1}^{-1}g_{2}). \tag*{(6)}

Cartier’s pairing yields a generalization of the Maclaurin formula:

f(g)=∑iui(f)(e)Fi(g)f(g)=\sum_{i}u_{i}(f)(e)F_{i}(g)

in the completion A^\widehat{A} of AA with respect to the me\mathfrak{m}_e-adic topology. Applying the Maclaurin formula to Lg1−1∗fL^*_{g_1^{-1}}f, we get

(Lg1−1∗f)(g)=∑iui(Lg1−1∗f)(e)Fi(g).\left(L^*_{g_1^{-1}}f\right)(g)=\sum_i u_i\left(L^*_{g_1^{-1}}f\right)(e)F_i(g).

Since uiu_i are left-invariant, this yields

f(g1g)=∑iui(f)(g1)Fi(g).f(g_1g)=\sum_i u_i(f)(g_1)F_i(g).

Setting g=g1−1g2g=g_1^{-1}g_2, we obtain (6). Next, we apply (6) to the second argument of f(g1,g2)f(g_1,g_2):

f(g1,g2)=∑i((1⊗ui)(f))(g1,g1)Fi(g1−1g2).f(g_1,g_2)=\sum_i\left((1\otimes u_i)(f)\right)(g_1,g_1)F_i(g_1^{-1}g_2).

Note that the change of variables g=g1−1g2g=g_1^{-1}g_2 transforms me\mathfrak{m}_e-adic topology into Δ\Delta-adic topology, ψ(me)=Δ∩(A⊗A)\psi(\mathfrak{m}_e)=\Delta\cap(A\otimes A). The claims of the theorem follow. □

Decomposition theorem for the jets of vector fields

In this section, we apply the Taylor formula to establish a decomposition theorem for the jets of vector fields on GG. Since V=A⊗gLV=A\otimes\mathfrak{g}^{L}, we get that

A#^V=(A⊗^ΔA)⊗gL.A\widehat{\#}V=(A\widehat{\otimes}_{\Delta}A)\otimes\mathfrak{g}^{L}.

Next, applying (5), we obtain

A#^V≅(A⊗^1⊗meA)⊗gL.A\widehat{\#}V\cong(A\widehat{\otimes}_{1\otimes\mathfrak{m}_e}A)\otimes\mathfrak{g}^{L}.

Taking into account the decomposition A=K⋅1⊕meA=K\cdot1\oplus\mathfrak{m}_e, we get

A#^V≅(A⊗1)⊗gL⊕(A⊗^1⊗meme)⊗gL.A\widehat{\#}V\cong(A\otimes1)\otimes\mathfrak{g}^{L}\oplus(A\widehat{\otimes}_{1\otimes\mathfrak{m}_e}\mathfrak{m}_e)\otimes\mathfrak{g}^{L}.

We claim that the first summand on the right-hand side forms a Lie subalgebra, while the second summand is an ideal. Let us compute the Lie brackets.

Lemma 8. The subspace A#gLA\#\mathfrak{g}^{L} in A#VA\#V forms a subalgebra, which is isomorphic to VV.

Proof. By comparing formula (2) for the Lie brackets in the smash product with the formula for the Lie bracket in VV, we see that the map A#gL→VA\#\mathfrak{g}^{L}\to V, f#x↦fxf\#x\mapsto fx, for f∈Af\in A, x∈gLx\in\mathfrak{g}^{L}, is an isomorphism of Lie algebras. □

We point out that the subalgebra A#gLA\#\mathfrak{g}^{L} is distinct from the natural subalgebra of vector fields 1#V\mathbf{1}\#V of the Lie algebra of algebraic jets of vector fields A#VA\#V.

Denote by L+\mathcal{L}_{+} the subalgebra of vector fields vanishing at ee: L+=me⊗gL\mathcal{L}_{+}=\mathfrak{m}_{e}\otimes\mathfrak{g}^{L}. Since meN⊗gL\mathfrak{m}_{e}^{N}\otimes\mathfrak{g}^{L} is an ideal in L+\mathcal{L}_{+}, we can take the completion:

L^+=lim←⁡N(me⊗gL)/(meN⊗gL).\widehat{\mathcal{L}}_{+}=\varprojlim_{N}(\mathfrak{m}_{e}\otimes\mathfrak{g}^{L})/(\mathfrak{m}_{e}^{N}\otimes\mathfrak{g}^{L}).

We will also need the completion

A⊗^L+=lim←⁡NA⊗((me⊗gL)/(meN⊗gL)).A\widehat{\otimes}\mathcal{L}_{+}=\varprojlim_{N}A\otimes\left((\mathfrak{m}_{e}\otimes\mathfrak{g}^{L})/(\mathfrak{m}_{e}^{N}\otimes\mathfrak{g}^{L})\right).

We point out that tensor products do not commute with inverse limits in general. Let L=Der⁡K[t1,…,tn]L=\operatorname{Der}K[t_{1},\ldots,t_{n}] and L^=Der⁡K[[t1,…,tn]]\widehat{L}=\operatorname{Der}K[[t_{1},\ldots,t_{n}]] with n=dim⁡Gn=\dim G. Let L+=m0LL_{+}=\mathfrak{m}_{0}L and L^+=mL^\widehat{L}_{+}=\mathfrak{m}\widehat{L}, where m0\mathfrak{m}_{0} is the ideal of functions vanishing at 00 and m\mathfrak{m} is the unique maximal ideal of K[[t1,…,tn]]K[[t_{1},\ldots,t_{n}]]. By [6] (Lemma 6), L^+≅L^+\widehat{\mathcal{L}}_{+}\cong\widehat{L}_{+}. Also, for every NN,

L^+/meNL^+≅L^+/mNL^+≅L+/m0NL+.\widehat{\mathcal{L}}_{+}/\mathfrak{m}_{e}^{N}\widehat{\mathcal{L}}_{+}\cong\widehat{L}_{+}/\mathfrak{m}^{N}\widehat{L}_{+}\cong L_{+}/\mathfrak{m}_{0}^{N}L_{+}.

Lemma 9. The subspace (A⊗1⊗meψ(me))⊗gL(A\underset{1\otimes\mathfrak{m}_{e}}{\otimes}\psi(\mathfrak{m}_{e}))\otimes\mathfrak{g}^{L} in A#V^A\#\widehat{V} is a Lie subalgebra isomorphic to A⊗^L+A\widehat{\otimes}\mathcal{L}_{+} with Lie bracket

[f1⊗τ1,f2⊗τ2]=f1f2⊗[τ1,τ2],for f1,f2∈A,τ1,τ2∈L+,[f_{1}\otimes\tau_{1},f_{2}\otimes\tau_{2}]=f_{1}f_{2}\otimes[\tau_{1},\tau_{2}],\qquad\text{for }f_{1},f_{2}\in A,\tau_{1},\tau_{2}\in\mathcal{L}_{+},

extended to the completion by continuity.

Proof. We need to compute the Lie bracket [(f1⊗1)ψ(h1)⊗xL,(f2⊗1)ψ(h2)⊗yL][(f_{1}\otimes1)\psi(h_{1})\otimes x^{L},(f_{2}\otimes1)\psi(h_{2})\otimes y^{L}] in A#VA\#V. Since ψ(h1),ψ(h2)∈Δ\psi(h_{1}),\psi(h_{2})\in\Delta, the terms with μ\mu in (4) vanish, and we get

(f1f2⊗1)ψ(h1)(1⊗xL)ψ(h2)⊗yL−(f1f2⊗1)ψ(h2)(1⊗yL)ψ(h1)⊗xL+(f1f2⊗1)ψ(h1)ψ(h2)⊗[x,y]L.\begin{aligned} &(f_{1}f_{2}\otimes1)\psi(h_{1})(1\otimes x^{L})\psi(h_{2})\otimes y^{L}-(f_{1}f_{2}\otimes1)\psi(h_{2})(1\otimes y^{L})\psi(h_{1})\otimes x^{L}\\ &\quad+(f_{1}f_{2}\otimes1)\psi(h_{1})\psi(h_{2})\otimes[x,y]^{L}. \end{aligned}

Applying Corollary 6, we get

(f1f2⊗1)ψ(h1xL(h2))⊗yL−(f1f2⊗1)ψ(h2yL(h1))⊗xL+(f1f2⊗1)ψ(h1h2)⊗[x,y]L.(f_{1}f_{2}\otimes1)\psi(h_{1}x^{L}(h_{2}))\otimes y^{L}-(f_{1}f_{2}\otimes1)\psi(h_{2}y^{L}(h_{1}))\otimes x^{L}+(f_{1}f_{2}\otimes1)\psi(h_{1}h_{2})\otimes[x,y]^{L}.

This matches the Lie bracket in A⊗L+A\otimes\mathcal{L}_{+}. □

Note that me\mathfrak{m}_{e} is a GG-module under the conjugation action C∗C^{*}. This follows from the fact that the identity element is fixed by conjugation. This GG-action induces a g\mathfrak{g}-module structure on me\mathfrak{m}_{e}, where x∈gx\in\mathfrak{g} acts as vector field xC=xL+xRx^{C}=x^{L}+x^{R}. The fact that xLx^{L} and xRx^{R} evaluate to opposite vectors at ee, gives us an alternative proof of invariance of me\mathfrak{m}_{e} under this action of g\mathfrak{g}.

Now we are ready to prove the main result of the paper:

Theorem 10 (Decomposition Theorem). Let GG be a connected affine algebraic group. Then A#^VA\widehat{\#}V decomposes into a semidirect product:

A#^V≅V⋉A⊗^L+A\widehat{\#}V \cong V \ltimes A\widehat{\otimes}\mathcal{L}_{+}

with the following action of VV on the ideal A⊗^L+A\widehat{\otimes}\mathcal{L}_{+}:

[f1⊗xL,f2⊗h⊗yL]=f1xL(f2)⊗h⊗yL+f1f2⊗xC(h)⊗yL+f1f2⊗h⊗[x,y]L,[f_1\otimes x^{L},f_2\otimes h\otimes y^{L}]=f_1x^{L}(f_2)\otimes h\otimes y^{L}+f_1f_2\otimes x^{C}(h)\otimes y^{L}+f_1f_2\otimes h\otimes[x,y]^{L},

extended to the completion by continuity. In this isomorphism, the embedding maps V→A#^VV\to A\widehat{\#}V and A⊗L+→A#^VA\otimes\mathcal{L}_{+}\to A\widehat{\#}V are given by

f⊗xL↦f#xL,f\otimes x^{L}\mapsto f\#x^{L},
f⊗h⊗yL↦(f#1)ψ(h)yL,f\otimes h\otimes y^{L}\mapsto(f\#1)\psi(h)y^{L},

extended to completions by continuity.

Remark 11. The semidirect product V⋉A⊗^L+V\ltimes A\widehat{\otimes}\mathcal{L}_{+} has a subalgebra g⋉L^+\mathfrak{g}\ltimes\widehat{\mathcal{L}}_{+} with the Lie bracket

[x,h⊗y]=xC(h)⊗y+h⊗[x,y],for h∈me,x,y∈g.[x,h\otimes y]=x^{C}(h)\otimes y+h\otimes[x,y],\qquad\text{for }h\in\mathfrak{m}_{e},x,y\in\mathfrak{g}.

The Lie algebra V⋉A⊗^L+V\ltimes A\widehat{\otimes}\mathcal{L}_{+} is a completed smash product of g⋉L^+\mathfrak{g}\ltimes\widehat{\mathcal{L}}_{+} with AA, where x∈gx\in\mathfrak{g} acts on AA via xLx^{L}, and L^+\widehat{\mathcal{L}}_{+} acts on AA trivially.

Proof. We have already seen in Lemmas 8 and 9 that the maps V→A#^VV\to A\widehat{\#}V and A⊗L+→A#^VA\otimes\mathcal{L}_{+}\to A\widehat{\#}V are embeddings of Lie algebras. We need to compute the Lie bracket of the images of f1⊗xLf_1\otimes x^{L} and f2⊗h⊗yLf_2\otimes h\otimes y^{L} in A#^VA\widehat{\#}V. For this, we use (4):

[(f1#1)xL,(f2#1)ψ(h)yL]=(f1#1)(xL(f2)#1)ψ(h)yL+(f1f2#1)(xL⊗1)ψ(h)yL+(f1f2#1)(1⊗xL)ψ(h)yL\begin{aligned} [(f_1\#1)x^{L},(f_2\#1)\psi(h)y^{L}] \\ &=(f_1\#1)(x^{L}(f_2)\#1)\psi(h)y^{L}+(f_1f_2\#1)(x^{L}\otimes1)\psi(h)y^{L}+(f_1f_2\#1)(1\otimes x^{L})\psi(h)y^{L} \end{aligned}
−(f2μ(ψ(h))#1)(yL(f1)#1)xL+(f1f2#1)ψ(h)[x,y]L.-(f_2\mu(\psi(h))\#1)(y^{L}(f_1)\#1)x^{L}+(f_1f_2\#1)\psi(h)[x,y]^{L}.

Since ψ(h)∈Δ\psi(h)\in\Delta for h∈meh\in\mathfrak{m}_{e}, we get that μ(ψ(h))=0\mu(\psi(h))=0. By Corollary 6, (xL⊗1+1⊗xL)ψ(h)=ψ(xC(h))(x^{L}\otimes1+1\otimes x^{L})\psi(h)=\psi(x^{C}(h)). The above expression simplifies to

(f1xL(f2)#1)ψ(h)yL+(f1f2#1)ψ(xC(h))yL+(f1f2#1)ψ(h)[x,y]L.(f_1x^{L}(f_2)\#1)\psi(h)y^{L}+(f_1f_2\#1)\psi(x^{C}(h))y^{L}+(f_1f_2\#1)\psi(h)[x,y]^{L}.

This matches the Lie bracket in the semidirect product V⋉A⊗^L+V\ltimes A\widehat{\otimes}\mathcal{L}_{+}. We point out that the embedding V⋉A⊗L+→A#^VV\ltimes A\otimes\mathcal{L}_{+}\to A\widehat{\#}V respects the filtrations: the image of mekL+\mathfrak{m}_{e}^{k}\mathcal{L}_{+} is in Δk+1⊗V\Delta^{k+1}\otimes V. This means that this embedding extends by continuity to a homomorphism of completed Lie algebras, which becomes an isomorphism by the Taylor formula. □

The most important class of AVAV-modules is the category of differentiable AVAV-modules. Let us recall the definition.

Definition 12. An AVAV-module MM is called NN-differentiable if it is annihilated by operators

∑k=0N(−1)k(Nk)fN−k#fkτ\sum_{k=0}^{N}(-1)^k\binom{N}{k}f^{N-k}\#f^k\tau

for all f∈Af\in A, τ∈V\tau\in V.

Equivalently, an AVAV-module is NN-differentiable if it is annihilated by ΔN⊗V\Delta^N\otimes V. DD-modules are precisely 1-differentiable AVAV-modules.

Corollary 13. NN-differentiable AVAV-modules on an affine algebraic group GG are precisely modules for the associative algebra A#U(g⋉(L+/meN−1L+))A\#U(\mathfrak{g}\ltimes(\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+)), which is the smash product of a Hopf algebra with its module AA. Here g\mathfrak{g} acts on AA by left-invariant derivations, whereas the action of L+\mathcal{L}_+ on AA is trivial.

Proof. NN-differentiable AVAV-modules are strong modules for the Rinehart pair (A,A#V/(ΔN⊗V))(A,A\#V/(\Delta^N\otimes V)). Since we are considering here a truncation of a filtration, we have an isomorphism A#V/(ΔN⊗V)≅A#V^/(Δ^N⊗V)A\#V/(\Delta^N\otimes V)\cong A\#\widehat{V}/(\widehat{\Delta}^N\otimes V). Applying the Decomposition Theorem, we get A#V^/(Δ^N⊗V)≅V⋉(A⊗L^+)/(A⊗m^eN−1L+)A\#\widehat{V}/(\widehat{\Delta}^N\otimes V)\cong V\ltimes(A\otimes\widehat{\mathcal{L}}_+)/(A\otimes\widehat{\mathfrak{m}}_e^{N-1}\mathcal{L}_+). Again, since we truncate the filtration, we have (A⊗L^+)/(A⊗m^eN−1L+)≅A⊗(L+/meN−1L+)(A\otimes\widehat{\mathcal{L}}_+)/(A\otimes\widehat{\mathfrak{m}}_e^{N-1}\mathcal{L}_+)\cong A\otimes(\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+). We conclude that A#V/(ΔN⊗V)A\#V/(\Delta^N\otimes V) is isomorphic to the smash product A#(g⋉(L+/meN−1L+))A\#(\mathfrak{g}\ltimes(\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+)). However,

Ustrong(A,A#(g⋉(L+/meN−1L+)))≅Uweak(A,g⋉(L+/meN−1L+))≅A#U(g⋉(L+/meN−1L+)),U_{\mathrm{strong}}(A,A\#(\mathfrak{g}\ltimes(\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+)))\cong U_{\mathrm{weak}}(A,\mathfrak{g}\ltimes(\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+)) \cong A\#U(\mathfrak{g}\ltimes(\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+)),

which establishes the claim.

Free AVAV-modules of finite rank and the Lax equation

In this final section, we will discuss a few examples of AVAV-modules on affine algebraic groups. We say that an AVAV-module has finite rank (resp. free) if it is finitely generated (resp. free) as an AA-module.

The first natural class of AVAV-modules has a geometric origin. These are tensor modules. The space of vector fields VV on GG is a free AVAV-module of finite rank, V≅A⊗gV\cong A\otimes\mathfrak{g}. Its dual is the space of 1-forms, Ω1=Hom⁡A(V,A)\Omega^1=\operatorname{Hom}_A(V,A), which is also a free AA-module, Ω1≅A⊗g∗\Omega^1\cong A\otimes\mathfrak{g}^*. By taking the tensor product over AA of kk copies of VV and mm copies of Ω1\Omega^1, we obtain the AVAV-module of (k,m)(k,m)-tensor fields, which will again be a free module of finite rank. All tensor modules are 2-differentiable.

Let us generalize the construction of (k,m)(k,m)-tensor modules. By Corollary 13, every 2-differentiable AVAV-module is a module for g⋉(L+/meL+)\mathfrak{g}\ltimes(\mathcal{L}_+/\mathfrak{m}_e\mathcal{L}_+). Note that L+/meL+≅L+/m0L+≅gl(g)\mathcal{L}_+/\mathfrak{m}_e\mathcal{L}_+\cong\mathcal{L}_+/\mathfrak{m}_0\mathcal{L}_+\cong\mathfrak{gl}(\mathfrak{g}).

Proposition 14. In the semidirect product g⋉(L+/meL+)≅g⋉gl(g)\mathfrak{g}\ltimes(L_{+}/\mathfrak{m}_{e}L_{+})\cong\mathfrak{g}\ltimes\mathfrak{gl}(\mathfrak{g}), the action of g\mathfrak{g} on gl(g)≅g⊗g∗\mathfrak{gl}(\mathfrak{g})\cong\mathfrak{g}\otimes\mathfrak{g}^{*} is the tensor product of the adjoint and coadjoint representations.

Proof. In the semidirect product, gL\mathfrak{g}^{L} acts on (me/me2)⊗gL(\mathfrak{m}_{e}/\mathfrak{m}_{e}^{2})\otimes\mathfrak{g}^{L} via the tensor product of the conjugation action and the adjoint action. Consider the pairing

me/me2×gL⟶A/me≅K,\mathfrak{m}_{e}/\mathfrak{m}_{e}^{2}\times\mathfrak{g}^{L}\longrightarrow A/\mathfrak{m}_{e}\cong K,

given by ⟨f,xL⟩=xL(f)(e)\langle f,x^{L}\rangle=x^{L}(f)(e). Let us check that this is an invariant pairing of g\mathfrak{g}-modules, where x∈gx\in\mathfrak{g} acts on me/me2\mathfrak{m}_{e}/\mathfrak{m}_{e}^{2} via xCx^{C} and on gL\mathfrak{g}^{L} via ad⁡(xL)\operatorname{ad}(x^{L}). Indeed for y∈gy\in\mathfrak{g},

⟨yC(f),xL⟩=xL(yC(f))(e)=xL(yL(f))(e)+xL(yR(f))(e)\langle y^{C}(f),x^{L}\rangle=x^{L}(y^{C}(f))(e)=x^{L}(y^{L}(f))(e)+x^{L}(y^{R}(f))(e)
⟨f,ad⁡(yL)xL⟩=[yL,xL](f)(e)=−xL(yL(f))(e)+yL(xL(f))(e).\langle f,\operatorname{ad}(y^{L})x^{L}\rangle=[y^{L},x^{L}](f)(e)=-x^{L}(y^{L}(f))(e)+y^{L}(x^{L}(f))(e).

Since xLx^{L} and yRy^{R} commute, we have

⟨yC(f),xL⟩+⟨f,ad⁡(yL)xL⟩=yC(xL(f))(e).\langle y^{C}(f),x^{L}\rangle+\langle f,\operatorname{ad}(y^{L})x^{L}\rangle=y^{C}(x^{L}(f))(e).

However, the vector field yCy^{C} vanishes at ee, which means that ⟨yC(f),xL⟩+⟨f,ad⁡(yL)xL⟩=0\langle y^{C}(f),x^{L}\rangle+\langle f,\operatorname{ad}(y^{L})x^{L}\rangle=0. The space me/me2\mathfrak{m}_{e}/\mathfrak{m}_{e}^{2} is identified with g∗\mathfrak{g}^{*} via the above pairing, and the invariance property means that it is isomorphic to the coadjoint g\mathfrak{g}-module. Hence, L+/meL+L_{+}/\mathfrak{m}_{e}L_{+} is isomorphic to g∗⊗g\mathfrak{g}^{*}\otimes\mathfrak{g} as a g\mathfrak{g}-module. □

Proposition 15. Consider the action of g\mathfrak{g} on gl(g)\mathfrak{gl}(\mathfrak{g}) via the tensor product of the adjoint and coadjoint actions. The corresponding semidirect product splits, g⋉gl(g)≅g⊕gl(g)\mathfrak{g}\ltimes\mathfrak{gl}(\mathfrak{g})\cong\mathfrak{g}\oplus\mathfrak{gl}(\mathfrak{g}) as Lie algebras.

Proof. The action of g\mathfrak{g} on gl(g)\mathfrak{gl}(\mathfrak{g}) gives a homomorphism ρ:g→Der⁡(gl(g))\rho:\mathfrak{g}\to\operatorname{Der}(\mathfrak{gl}(\mathfrak{g})). It is easy to see that the derived subalgebra and the center of a Lie algebra are invariant under every derivation of that Lie algebra. This means that the subspaces KIK I and sl(g)\mathfrak{sl}(\mathfrak{g}) are invariant under ρ(x)\rho(x) for every x∈gx\in\mathfrak{g}. We claim that ρ(x)I=0\rho(x)I=0. Since KIK I is an invariant subspace, it is sufficient to show that tr⁡(ρ(x)I)=0\operatorname{tr}(\rho(x)I)=0. Choose dual bases {yi}\{y_i\} of g\mathfrak{g} and {yi∗}\{y_i^*\} of g∗\mathfrak{g}^*. Then I=∑iyi⊗yi∗I=\sum_i y_i\otimes y_i^*, ρ(x)I=∑iad⁡(x)yi⊗yi∗+yi⊗ad⁡∗(x)yi∗\rho(x)I=\sum_i\operatorname{ad}(x)y_i\otimes y_i^*+y_i\otimes\operatorname{ad}^{*}(x)y_i^* and

tr⁡(ρ(x)I)=∑iyi∗([x,yi])+ad⁡∗(x)yi∗(yi)=∑iyi∗([x,yi])−yi∗([x,yi])=0.\operatorname{tr}(\rho(x)I)=\sum_i y_i^*([x,y_i])+\operatorname{ad}^{*}(x)y_i^*(y_i)=\sum_i y_i^*([x,y_i])-y_i^*([x,y_i])=0.

We conclude that ρ:g→Der⁡(sl(g))\rho:\mathfrak{g}\to\operatorname{Der}(\mathfrak{sl}(\mathfrak{g})). However, every derivation of sl(g)\mathfrak{sl}(\mathfrak{g}) is inner, Der⁡(sl(g))≅sl(g)\operatorname{Der}(\mathfrak{sl}(\mathfrak{g}))\cong\mathfrak{sl}(\mathfrak{g}). Composing these maps, we get a homomorphism ρ~:g→sl(g)\widetilde{\rho}:\mathfrak{g}\to\mathfrak{sl}(\mathfrak{g}) that realizes the action by inner derivations. It is easy to see that elements x−ρ~(x)x-\widetilde{\rho}(x) span a subalgebra of g⋉gl(g)\mathfrak{g}\ltimes\mathfrak{gl}(\mathfrak{g}) which is isomorphic to g\mathfrak{g} and commutes with gl(g)\mathfrak{gl}(\mathfrak{g}). This proves that the semidirect product splits. □

Corollary 16. A 22-differentiable AVAV-module on an affine algebraic group GG is precisely a DD-module on GG equipped with an additional action of gln\mathfrak{gl}_n, n=dim⁡Gn = \dim G, where the action of gln\mathfrak{gl}_n commutes with the action of DD.

Let us present explicit formulas for the AVAV-module action in the setting of the above corollary.

Given f∈Af \in A, x∈gx \in\mathfrak{g}, we view fad⁡(x)f\operatorname{ad}(x) as an element of A⊗gl(g)A \otimes\mathfrak{gl}(\mathfrak{g}). In the Lie algebra A⊗gl(g)A \otimes\mathfrak{gl}(\mathfrak{g}), gl(g)\mathfrak{gl}(\mathfrak{g}) acts on AA trivially.

For f∈Af \in A, the differential dfdf belongs to A⊗g∗A \otimes\mathfrak{g}^{*}, with df(y)=yL(f)df(y) = y^{L}(f). Then xdf∈A⊗g⊗g∗=A⊗gl(g)xdf \in A \otimes\mathfrak{g} \otimes\mathfrak{g}^{*} = A \otimes\mathfrak{gl}(\mathfrak{g}).

Theorem 17 (ChatGPT). Let GG be an affine algebraic group. Let MM be a DD-module on GG, and simultaneously a gl(g)\mathfrak{gl}(\mathfrak{g})-module, with commuting actions. Then MM admits a structure of an AVAV-module with AA-action defined by the DD-module structure and the following action of VV:

ρ(fxL)=fxL+fad⁡(x)−xdf,for f∈A,x∈g.\rho(fx^{L}) = fx^{L} + f\operatorname{ad}(x) - xdf,\quad\text{for } f \in A, x \in\mathfrak{g}.

The right-hand side is viewed as an operator on MM, where the first term comes from the DD-module action, and the other two terms correspond to the action of A⊗gl(g)A \otimes\mathfrak{gl}(\mathfrak{g}).

Proof. We point out that the action of gl(g)\mathfrak{gl}(\mathfrak{g}) on MM is AA-linear, since the action of gl(g)\mathfrak{gl}(\mathfrak{g}) commutes with the action of DD, and hence the action of AA. We need to show that the formula defines a representation of VV on MM. For this, we record the Lie brackets involving the operators appearing on the right-hand side.

[fxL,hyL]=fxL(h)yL−hyL(f)xL+fh[x,y]L,[fxL,had⁡(y)]=fxL(h)ad⁡(y),[fad⁡(x),had⁡(y)]=fhad⁡([x,y]),[x df,y dh]=yL(f)x dh−xL(h)y df,[fxL+fad⁡(x),y dh]=f[x,y] dh+fy d(xL(h)).\begin{aligned} [fx^{L},hy^{L}] &= fx^{L}(h)y^{L} - hy^{L}(f)x^{L} + fh[x,y]^{L},\\ [fx^{L},h\operatorname{ad}(y)] &= fx^{L}(h)\operatorname{ad}(y),\\ [f\operatorname{ad}(x),h\operatorname{ad}(y)] &= fh\operatorname{ad}([x,y]),\\ [x\,df,y\,dh] &= y^{L}(f)x\,dh - x^{L}(h)y\,df,\\ [fx^{L} + f\operatorname{ad}(x),y\,dh] &= f[x,y]\,dh + fy\,d(x^{L}(h)). \end{aligned}

These relations are straightforward to verify, and the fact that ρ\rho is a representation of VV easily follows. □

In particular, if MM is a DD-module on GG and WW is a gln\mathfrak{gl}_n-module, then M⊗WM \otimes W admits the structure of a 22-differentiable AVAV-module on GG.

Let us now discuss free AVAV-modules of finite rank. By the theorem of Bouaziz-Rocha [10], every finitely generated AVAV-module is differentiable, and hence by Corollary 13, admits the action of A#U(g⋉(L+/meN−1L+))A\#U(\mathfrak{g} \ltimes(\mathcal{L}_{+}/\mathfrak{m}_{e}^{N-1}\mathcal{L}_{+})) for some NN. This algebra contains a subalgebra A#U(g)A\#U(\mathfrak{g}), which is the algebra DD of differential operators on GG. Hence NN-differentiable AVAV-modules are DD-modules with an additional action of the Lie algebra L+/meN−1L+\mathcal{L}_{+}/\mathfrak{m}_{e}^{N-1}\mathcal{L}_{+}, compatible with the DD-module structure.

Let MM be a free DD-module of finite rank, M=A⊗WM = A \otimes W, dim⁡W<∞\dim W < \infty. The DD-module structure is determined by the action of gL\mathfrak{g}^{L} on WW: gL×W→A⊗W\mathfrak{g}^{L} \times W \to A \otimes W. Such a map is encoded by a 1-form ω\omega on GG with values in End⁡(W)\operatorname{End}(W). Then the action of VV on MM is given by

(fxL)(hw)=fxL(h)w+fhω(xL)(w),for f,h∈A, x∈g, w∈W.(fx^{L})(hw) = fx^{L}(h)w + fh\omega(x^{L})(w), \qquad\text{for } f,h \in A,\ x \in\mathfrak{g},\ w \in W.

In order for this to yield a Lie algebra action, the form ω\omega must satisfy the condition

dω+ω∧ω=0,(7)d\omega+ \omega\wedge\omega= 0, \tag*{(7)}

or equivalently,

ω([x,y]L)=xL(ω(yL))−yL(ω(xL))+[ω(xL),ω(yL)].\omega([x,y]^{L}) = x^{L}(\omega(y^{L})) - y^{L}(\omega(x^{L})) + [\omega(x^{L}),\omega(y^{L})].

Let dim⁡W=m\dim W = m. A change of basis in A⊗WA \otimes W is described by the group GLm(A)GL_m(A). Under a change of basis given by g∈GLm(A)g \in GL_m(A), the 1-form ω\omega transforms as [9]:

ω′=g−1ωg+g−1dg.\omega' = g^{-1}\omega g + g^{-1}dg.

Let us consider the simplest case, m=1m = 1. The condition on ω\omega in this case is just dω=0d\omega= 0, i.e., the form must be closed. In particular, we may take an exact form, ω=dH\omega= dH, H∈AH \in A, and the action of VV on this DD-module will be given by

(fxL)(hw)=fxL(h)w+fhxL(H)w.(fx^{L})(hw) = fx^{L}(h)w + fhx^{L}(H)w.

Note that GL1(A)=A∗GL_1(A) = A^*. If the units of AA are scalars, two such DD-modules are non-isomorphic unless H1−H2H_1 - H_2 is a constant function [5].

Now let us see how a free rank-one DD-module can be turned into an AVAV-module. For this, we need to define a compatible action of L+/meN−1L+\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+. Since End⁡(W)\operatorname{End}(W) is abelian, the derived subalgebra (L+/meN−1L+)(1)(\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+)^{(1)} must act trivially. The quotient by the derived subalgebra is 1-dimensional and is spanned by the identity operator II in gl(g)=L+/meL+\mathfrak{gl}(\mathfrak{g}) = \mathcal{L}_+/\mathfrak{m}_e\mathcal{L}_+. Assume that II acts as multiplication by λ∈A\lambda\in A. Since the action of VV must commute with the action of II, we see that xL(λ)=0x^{L}(\lambda) = 0 for all x∈gx \in\mathfrak{g}, so λ\lambda must be a scalar in KK. The action of VV on the corresponding AVAV-module is

(fxL)(hw)=fxL(h)w+fhω(xL)w+λhxL(f)w.(fx^{L})(hw) = fx^{L}(h)w + fh\omega(x^{L})w + \lambda hx^{L}(f)w.

Let us generalize the above construction of an AVAV-module to higher rank. Let B∈Mm(K[t1,…,tk])B \in M_m(K[t_1,\ldots,t_k]) satisfy the condition

[∂B∂ti,∂B∂tj]=0for all i,j.(8)\left[\frac{\partial B}{\partial t_i},\frac{\partial B}{\partial t_j}\right] = 0 \quad\text{for all } i,j. \tag*{(8)}

Fix H1,…,Hk∈AH_1,\ldots,H_k \in A. Then ω=dB(H1,…,Hk)\omega=dB(H_1,\ldots,H_k) satisfies (7) and yields a DD-module structure on A⊗KmA\otimes K^m:

(fxL)(hw)=fxL(h)w+∑j=1kfhxL(Hj)∂B∂tj(H1,…,Hk)w.(9)(fx^L)(hw)=fx^L(h)w+\sum_{j=1}^{k}fhx^L(H_j)\frac{\partial B}{\partial t_j}(H_1,\ldots,H_k)w. \tag*{(9)}

Theorem 18. Let GG be an affine algebraic group. Let B,L∈Mm(K[t1,…,tk])B,L\in M_m(K[t_1,\ldots,t_k]) satisfy (8) and

∂L∂ti=[L,∂B∂ti]for i=1,…,k.(10)\frac{\partial L}{\partial t_i}=\left[L,\frac{\partial B}{\partial t_i}\right]\quad\text{for }i=1,\ldots,k. \tag*{(10)}

Let H1,…,Hk∈AH_1,\ldots,H_k\in A. Then we have an AVAV-module structure on A⊗KmA\otimes K^m, where the action of VV is given by

(fxL)(hw)=fxL(h)w+∑j=1kfhxL(Hj)∂B∂tj(H1,…,Hk)w+xL(f)hL(H1,…,Hk)w.(11)(fx^L)(hw)=fx^L(h)w+\sum_{j=1}^{k}fhx^L(H_j)\frac{\partial B}{\partial t_j}(H_1,\ldots,H_k)w+x^L(f)hL(H_1,\ldots,H_k)w. \tag*{(11)}

Proof. To construct a differentiable AVAV-module on GG, we need to take a DD-module on GG and define on it a compatible AA-linear action of L+/meN−1L+\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+. For the construction presented in this theorem, we take a DD-module given by (9), and we define the action of L+/meN−1L+\mathcal{L}_+/\mathfrak{m}_e^{N-1}\mathcal{L}_+, where the derived subalgebra is acting trivially and the identity element I∈gl(g)I\in\mathfrak{gl}(\mathfrak{g}) acts by L(H1,…,Hk)L(H_1,\ldots,H_k). Let us verify that these actions are compatible. The compatibility condition for this setting is that the action of II must commute with the DD-module action (see the proof of Proposition 15). Indeed, using (10), we get

[xL+∑j=1kxL(Hj)∂B∂tj(H1,…,Hk),L(H1,…,Hk)]=∑j=1kxL(Hj)∂L∂tj(H1,…,Hk)+∑j=1kxL(Hj)[∂B∂tj(H1,…,Hk),L(H1,…,Hk)]=0.\begin{aligned} \left[x^L+\sum_{j=1}^{k}x^L(H_j)\frac{\partial B}{\partial t_j}(H_1,\ldots,H_k),L(H_1,\ldots,H_k)\right] \\ &=\sum_{j=1}^{k}x^L(H_j)\frac{\partial L}{\partial t_j}(H_1,\ldots,H_k)+\sum_{j=1}^{k}x^L(H_j)\left[\frac{\partial B}{\partial t_j}(H_1,\ldots,H_k),L(H_1,\ldots,H_k)\right]=0. \end{aligned}

Applying Theorem 10 to combine the two actions, we recover (11). □

We point out that when k=1k=1, condition (8) holds trivially, while (10) becomes

dLdt=[L,dBdt].\frac{dL}{dt}=\left[L,\frac{dB}{dt}\right].

This is known as the Lax equation in the theory of integrable systems. Many integrable systems may be expressed through the Lax equation. Finite-dimensional integrable systems of classical mechanics are written in Lax form in the context of algebraic groups. The Kovalevskaya top provides one such example [17]. System (10) is a generalized Lax equation with commuting flows. Thus, the construction reveals a connection between the Lax equation and the theory of AVAV-modules.

Conjecture 19. Let AA be the algebra of regular functions on a connected simplyconnected affine algebraic group. Every finitely generated projective AA-module is free.

School of Mathematics and Statistics, Carleton University, Ottawa, Canada

Email address: [email protected]

Email address: [email protected]

References

  1. [1]M.F. Atiyah, I.G. MacDonald, Introduction to Commutative Algebra, Vol. 2, Addison-Wesley, 1969.
  2. [2]V. V. Bavula, Generators and defining relations for the ring of differential operators on a smooth affine algebraic variety, Algebras and Representation Theory, 13 (2010), 159-187.arxiv.org/abs/math/0504475
  3. [3]A. Beilinson, J. Bernstein, Localisation de g-modules, C. R. Acad. Sci. Paris Sér. I Math., 292 (1981), 15-18.
  4. [4]Y. Billig, V. Futorny, Classification of irreducible representations of Lie algebra of vector fields on a torus, J. für die reine und angewandte Mathematik (Crelles Journal), 2016.720 (2016), 199-216.
  5. [5]Y. Billig, V. Futorny, Lie algebras of vector fields on smooth affine varieties, Comm. Algebra, 46 (2018), 3413-3429.
  6. [6]Y. Billig, V. Futorny, J. Nilsson, Representations of Lie algebras of vector fields on affine varieties, Israel J. Math., 233 (2019), 379-399.arxiv.org/abs/1709.08863
  7. [7]Y. Billig, C. Ingalls, A universal sheaf of algebras governing representations of vector fields on quasi-projective varieties, Proc. of the Amer. Math. Soc. 154 (2026), 1385-1399.
  8. [8]Y. Billig, C. Ingalls, A. Nasr, AV-modules of finite type on affine space, J. Algebra, 623 (2023), 481-495.DOI
  9. [9]A. Borel et al., Algebraic D-modules, Perspectives in Mathematics, 2, Academic Press, 1987.
  10. [10]E. Bouaziz, H. Rocha, Annihilators of AV-modules and differential operators, J. Algebra, 636 (2023), 869-887.DOI
  11. [11]J.-L. Brylinski, M. Kashiwara, Kazhdan-Lusztig conjecture and holonomic systems, Invent. Math. 64 (1981), 387-410.DOI
  12. [12]M. Demazure, P. Gabriel, Introduction to algebraic geometry and algebraic groups, North-Holland Mathematics Studies, 39, Elsevier, 1980.
  13. [13]D. Grantcharov, V. Serganova, Simple weight modules with finite weight multiplicities over the Lie algebra of polynomial vector fields, J. Reine Angew. Math., 792 (2022), 93-114.arxiv.org/abs/2102.09064
  14. [14]N. Jacobson, Lie Algebras, Interscience Publishers, 1962.
  15. [15]R. Hotta, K. Takeuchi, T. Tanisaki, D-modules, perverse sheaves, and representation theory. Progress in Math., 236, Springer, 2007.
  16. [16]G. S. Rinehart, Differential forms on general commutative algebras, Transactions AMS, 108 (1963), 195-222.DOI
  17. [17]A. V. Tsiganov, The Kowalewski top: A new Lax representation, J. Math. Phys., 38 (1997), 196-211.DOI
  18. [18]Y. Xue, R. Lü, Classification of simple bounded weight modules of the Lie algebra of vector fields on C^n, Israel J. Math., 253 (2023), 445-468.

Paper details

vlnc:W_4qwmjdb9
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT

View statement
Looking for related papers…
Contents