Free Jordan Algebras and Representations of 𝔰 ​ 𝔩 ^ 2 ​ ( J ) \widehat{\mathfrak{sl}}_{2}(J) Michael Lau, Olivier Mathieu Abstract Abstract: Let J be a unital Jordan algebra, and let \widehat{\mathfrak{sl}}_{2}(J) be the universal central extension of the Tits-Kantor-Koecher Lie algebra. In Part A, we study the category of (\widehat{\mathfrak{sl}}_{2}(J),{\rm SL}_{2}(\mathbb{K})) -modules. We characterize the dominant J -spaces, which are analogous to the dominant highest weights appearing in classical settings. A family of universal envelopes \mathcal{U}_{n}(J) associated to such modules is introduced and studied. We also prove some finiteness theorems. In Part C, we define the notion of smooth \widehat{\mathfrak{sl}}_{2}(J) -modules for augmented Jordan algebras J , and investigate the category of smooth modules in the spirit of Cline-Parshall-Scott highest weight categories. We show that the standard modules of this category are finite dimensional when J is finitely generated. The free unital Jordan algebra J(D) over D variables is an elusive object, but finiteness and Ext -vanishing properties suggest that the smooth \widehat{\mathfrak{sl}}_{2}(J(D)) -modules with even eigenvalues might form a generalized highest weight category. However, we prove that such an assertion would contradict recently obtained information about the growth of free Jordan algebras. See [ 24 ] and [ 13 ] for more details. It then follows that the category of smooth \widehat{\mathfrak{sl}}_{2}(J(D)) -modules with even eigenvalues is not a generalized highest weight category when D\geq 2 . Surprisingly, the proofs of most of these results make use of deep theorems of E. Zelmanov. Keywords: free Jordan algebras, Tits-Kantor-Koecher construction, Tits-Allison-Gao algebra, Weyl modules, good filtrations, generalized highest weight categories, FP_{\infty} Lie algebras, weight modules, Ext-vanishing. MSC2010: primary 17B10; secondary 17B60, 17C05, 17C50 Statements and Declarations: The authors report that they have no competing interests to declare. Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study. 1 Introduction Let \mathbb{K} be a field of characteristic zero. For any unital Jordan algebra J , we denote by \widehat{\mathfrak{sl}}_{2}(J) the universal central extension of the Tits-Kantor-Koecher Lie algebra TKK(J) . In the present paper, we investigate two categories of representations of the Lie algebra \widehat{\mathfrak{sl}}_{2}(J) . An (\widehat{\mathfrak{sl}}_{2}(J),{\mathbf{SL}}_{2}(\mathbb{K})) -module is an \widehat{\mathfrak{sl}}_{2}(J) -module that is locally finite dimensional as an \mathfrak{sl}_{2}(\mathbb{K}) -module. In Part A, we investigate finiteness properties of the category of (\widehat{\mathfrak{sl}}_{2}(J),{\mathbf{SL}}_{2}(\mathbb{K})) -modules for an arbitrary unital Jordan algebra J . Relative to the action of the usual diagonal element h of \mathfrak{sl}_{2}(\mathbb{K}) , any (\widehat{\mathfrak{sl}}_{2}(J),{\mathbf{SL}}_{2}(\mathbb{K})) -module X admits an eigenspace decomposition X=\bigoplus_{n\in\mathbb{Z}}X_{n} . In particular, the Lie algebra \mathfrak{G}:=\widehat{\mathfrak{sl}}_{2}(J) has a short grading \mathfrak{G}=\mathfrak{G}_{-2}\oplus\mathfrak{G}_{0}\oplus\mathfrak{G}_{2} . For integers n , we ask Which \mathfrak{G}_{0} -modules V are isomorphic to H^{0}(\mathfrak{G}_{2},X)\cap X_{n} for some X ? A vector space V endowed with a linear map \rho:J\to\hbox{End}_{\mathbb{K}}(V) is called a J -space when V is a \mathfrak{G}_{0} -module and \rho(a) is the action of h\otimes a\in\mathfrak{G}_{0} on V . This is equivalent to the identities [\rho(a),\rho(a^{2})]=0 and [[\rho(a),\rho(b)],\rho(c)]=4\rho(a(cb)-(ac)b) for all a,b,c\in J . A J -space is called dominant of level n if it is isomorphic to H^{0}(\mathfrak{G}_{2},X)\cap X_{n} for some (\widehat{\mathfrak{sl}}_{2}(J),{\mathbf{SL}}_{2}(\mathbb{K})) -module X . Our first result characterizes dominant J -spaces of level n . For any partition \sigma=(\sigma_{1},\sigma_{2},\ldots,\sigma_{m}) of n+1 , we write |C_{\sigma}| for the cardinality of the corresponding conjugacy class in the symmetric group S_{n+1} and \mathrm{sgn}(\sigma) for its signature. For a\in J , we write \rho_{\sigma}(a) for the expression \rho(a^{\sigma_{1}})\rho(a^{\sigma_{2}})\cdots\rho(a^{\sigma_{m}}) . Theorem 1 . Let (V,\rho) be a J -space such that \rho(1)=n . Then V is dominant of level n if and only if it satisfies the following condition: \sum_{\sigma\ \vdash\ n+1}\hbox{sgn}(\sigma)\,|C_{\sigma}|\,\rho_{\sigma}(a)=0\hbox{\ for all\ }a\in J. There is thus an associative algebra {\mathcal{U}}_{n}(J) , endowed with a linear map \rho_{n}:J\to{\mathcal{U}}_{n}(J) , which is the unital universal associative envelope of J in the category of dominant J -spaces of level n . We prove the following result: Theorem 2 . The map \rho_{n} is injective for any n\geq 4 . Next we prove two finiteness theorems: Theorem 3 . The associative algebra {\mathcal{U}}_{n}(J) is finitely generated, for all finitely generated Jordan algebras J . Theorem 4 . The Lie algebra \widehat{\mathfrak{sl}}_{2}(J) is finitely presented, for all finitely presented Jordan algebras J . The free Jordan \mathbb{K} -algebra J(D) on D generators admits a grading J(D)=\bigoplus_{n\geq 0}J_{n}(D) , where J_{0}(D)=\mathbb{K} and J_{1}(D) is the linear span of the D generators. The algebra J(D) is an elusive object. However, an explicit formula for \dim J_{n}(D) has been conjectured in [ 24 ] . The previous theorem shows that \widehat{\mathfrak{sl}}_{2}(J(D)) is FP_{2} . This theorem and other observations raise the natural question of whether the Lie algebra \widehat{\mathfrak{sl}}_{2}(J(D)) could be FP_{\infty} , that is, whether its trivial module has a projective resolution consisting of finitely generated modules. Such a finiteness condition would have profound consequences for the growth of free Jordan algebras: Theorem 5 . Assume that \widehat{\mathfrak{sl}}_{2}(J(D+1)) is of type FP_{\infty} . Then the conjectural formula of [ 24 ] for \dim J_{n}(D) holds. Part B contains the categorical background (used in Part C) on good filtrations in locally artinian categories. In particular, we introduce the notion of generalized highest weight categories ( generalized HW categories for short), which differ from highest weight categories in that they allow nontrivial self-extensions of standard objects. We close Part B with an Ext-vanishing result for generalized HW categories (Theorem 6 ). It has been conjectured in [ 24 ] that H_{k}(\widehat{\mathfrak{sl}}_{2}(J(D),\mathfrak{sl}_{2}(\mathbb{K}))=0 for k>0 . By [ 18 ] , the dual of this homology group describes the self-extensions of \mathbb{K} in the large category of (\widehat{\mathfrak{sl}}_{2}(J),{\mathbf{SL}}_{2}(\mathbb{K})) -modules. In Part C, we introduce the smaller and more manageable category of smooth modules, with a similar connection between H_{*}(\widehat{\mathfrak{sl}}_{2}(J(D),\mathfrak{sl}_{2}(\mathbb{K})) and self-extensions of the trivial module \mathbb{K} . A unital Jordan algebra J endowed with an ideal J_{+} of codimension 1 is called an augmented Jordan algebra. An \widehat{\mathfrak{sl}}_{2}(J) -module is called smooth if any cyclic submodule is a finite dimensional \widehat{\mathfrak{sl}}_{2}(J/J_{+}^{n}) -module for some n . This terminology comes from proalgebraic groups, where smoothness means that a module is locally a representation for some algebraic quotient. For p -adic groups, the definition of smoothness is similar. In Part C , we investigate the category of smooth modules in the spirit of the Cline-Parshall-Scott paper [ 9 ] . For an augmented Jordan algebra J and n\geq 0 , there is a 1 -dimensional J -space \mathbb{K}_{n} defined by \rho(1)=n and \rho(J_{+})=0 . The standard module \nabla(n) is a smooth module such that H^{0}(\mathfrak{G}_{2},\nabla(n))=\mathbb{K}_{n} , and satisfying a universal property. When J is finitely generated we prove that: Theorem 7 . The standard module \nabla(n) is finite dimensional. However, the category of smooth modules is almost never a highest weight category. Some results suggest that the category of smooth \widehat{\mathfrak{sl}}_{2}(J) -modules with even eigenvalues could be a generalized HW category when J is a finitely generated free Jordan algebra. We conclude the paper with the following result. Theorem 8 . Assume that the category of smooth \widehat{\mathfrak{sl}}_{2}(J(D)) -modules with even eigenvalues is a generalized HW category. Then the conjectural formula of [ 24 ] for \dim\,J_{n}(D) is true. About the proofs: The proof of Theorem 1 is elementary. For Theorem 2, the proof of the injectivity of \rho_{n} for even n is elementary. Strangely, it uses Zelmanov’s Classification Theorem [ 37 , 38 ] for the odd integers n\geq 5 . The proofs of Theorems 3 , 4, and 7, as well as Lemma 11.4 are based on Zelmanov’s Theorem for nil Jordan algebras. Theorems 5 and 8 are consequences of Theorem 2 and Corollary 1 of [ 24 ] . Recent developments In a very recent arXiv preprint [ 13 ] posted when the present paper was under final revisions, Dotsenko and Hentzel provide detailed computational data indicating that the conjectural graded dimension formula [ 24 ] for the free Jordan algebra on two generators fails in degree 19. When Theorems 5 and 8 are combined with Dotsenko and Hentzel’s result, we obtain the following immediate and perhaps surprising consequences for Lie algebra representation theory: Corollary. For D\geq 3 , the Lie algebra \widehat{\mathfrak{sl}}_{2}(J(D)) is not of type FP_{\infty} . Corollary. For D\geq 2 , the category of smooth \widehat{\mathfrak{sl}}_{2}(J(D)) -modules with even eigenvalues is not a generalized HW category. As illustrated by these corollaries, the results of the present paper show definite connections between the structure of free Jordan algebras, the representation theory of the Lie algebras \widehat{\mathfrak{sl}}_{2}(J(D)) , and Zelmanov’s Theorem on nil Jordan algebras of bounded index. Acknowledgements The authors would like to thank the referees for useful comments and for calling their attention to the very interesting paper of Chari, Fourier, and Khandai [ 7 ] . O.M. also thanks E. Neher and I. Shestakov for fruitful discussions. PART A: The category of (\widehat{\mathfrak{sl}}_{2}(J),{\rm SL}_{2}) -modules 2 The Lie algebra \widehat{\mathfrak{sl}}_{2}(J) Conventions and notations As before, let \mathbb{K} be a field of characteristic zero. All vector spaces, algebras, and tensor products will be taken over \mathbb{K} , unless explicitly otherwise mentioned. We start with the definitions involving the Lie algebra \mathfrak{sl}_{2}:=\mathfrak{sl}_{2}(\mathbb{K}) . We fix the usual basis \{h,e,f\} satisfying [h,e]=2e , [h,f]=-2f , and [e,f]=h . The Killing form \kappa will be normalized so that \kappa(h,h)=4 . For any \mathfrak{sl}_{2} -module M and z\in\mathbb{K} , we say that M_{z}:=\{m\in M\,|\,h.m=zm\} is an eigenspace relative to the action of h . When M_{z}\neq 0 , we say that z is an eigenvalue of the module M . All \mathfrak{sl}_{2} -modules considered in this paper will decompose as direct sums of eigenspaces of integer eigenvalues. For any nonnegative integer n , let L(n) be the simple \mathfrak{sl}_{2} -module with highest eigenvalue n . Clearly, L(0) is the trivial module \mathbb{K} , L(1)=\mathbb{K}^{2} is the natural 2 -dimensional representation, and L(n) is isomorphic to the space S^{n}L(1) of symmetric tensors for any n\geq 0 . We will write M(n) for the Verma module with highest eigenvalue n\in\mathbb{Z} . We continue with the notation involving Jordan algebras. We assume that all Jordan algebras J are unital unless explicitly stated otherwise. For every a,b\in J , let \partial_{a,b}:\ J\rightarrow J be the linear map \partial_{a,b}(x)=a(bx)-(ax)b . The Jordan identity implies that \partial_{a,b} is a derivation of J . An inner derivation is a linear combination of the derivations \partial_{a,b} . The vector space \text{\rm Inn}\,J of inner derivations is a Lie algebra. For x\in\mathfrak{sl}_{2} and a\in J , the tensor product x\otimes a\in\mathfrak{sl}_{2}\otimes J will be denoted x(a) , though x(1)=x\otimes 1 will also be denoted by x . Tits-Kantor-Koecher construction In his 1962 paper [ 35 ] , Tits defined a Lie algebra structure on the space TKK(J):=(\mathfrak{sl}_{2}\otimes J)\oplus\text{\rm Inn}\,J, with Lie bracket (T1) [x(a),y(b)]=[x,y](ab)+\kappa(x,y)\partial_{a,b} (T2) [\partial,x(a)]=x(\partial\,a) , (T3) [\partial,\partial^{\prime}]=\partial\circ\partial^{\prime}-\partial^{\prime}\circ\partial, for all x,y\in\mathfrak{sl}_{2} , \partial,\partial^{\prime}\in\text{\rm Inn}\,J , and a,b\in J . This construction was later generalized to Jordan pairs and triple systems by Kantor and Koecher, and TKK(J) is commonly known as the Tits-Kantor-Koecher (or TKK) algebra . It should be noted that the Tits construction, associating a Lie algebra TKK(J) to each Jordan algebra J , is not functorial. Tits-Allison-Gao construction The Lie algebra TKK(J) is perfect and its universal central extension, denoted by \widehat{\mathfrak{sl}_{2}}(J) , was nicely described by Allison and Gao [ 3 ] . See also [ 2 ] , [ 33 ] . As a vector space, \widehat{\mathfrak{sl}}_{2}(J)=\left({\mathfrak{sl}}_{2}\otimes J\right)\oplus\{J,J\} , where \{J,J\}=(\bigwedge^{2}J)/\mathcal{S} and \mathcal{S}=\hbox{Span}\{a\wedge a^{2}\mid\ a\in J\} . For any a,b\in J , we write \{a,b\} for the image of a\wedge b in \{J,J\} . The bracket on \widehat{\mathfrak{sl}}_{2}(J) is given by (R1) [x(a),y(b)]=[x,y](ab)+\kappa(x,y)\{a,b\} (R2) [\{a,b\},x(c)]=x(\partial_{a,b}\,c) (R3) [\{a,b\},\{c,d\}]=\{\partial_{a,b}\,c,d\}+\{c,\partial_{a,b}\,d\} . for all a,b,c,d\in J and x,y\in\mathfrak{sl}_{2} . It is a bit tricky to show that (R3) is skew-symmetric [ 3 ] . Relation (R3) is similar to Relation (T3) for Tits-Kantor-Koecher algebras, since \partial_{a,b}\circ\partial_{c,d}-\partial_{c,d}\circ\partial_{a,b}=\partial_{\partial_{a,b}c,d}+\partial_{c,\partial_{a,b}d}. Hence there is a Lie algebra epimorphism \widehat{\mathfrak{sl}}_{2}(J)\to TKK(J) which is the identity on \mathfrak{sl}_{2}\otimes J and sends the symbol \{a,b\} to \partial_{a,b} . It is clear from its definition that the Tits-Allison-Gao construction of \widehat{\mathfrak{sl}}_{2}(J) defines a functor from Jordan algebras to Lie algebras, as is explained in [ 24 ] . Therefore, for any ideal I in J , the kernel of the Lie algebra morphism \widehat{\mathfrak{sl}}_{2}(J)\to\widehat{\mathfrak{sl}_{2}}(J/I) will be denoted as \widehat{\mathfrak{sl}}_{2}(I) . When the Jordan algebra J is associative, TKK(J) is obviously \mathfrak{sl}_{2}(J):=\mathfrak{sl}_{2}\otimes J . In that case, \{J,J\} is the quotient \Omega^{1}_{J}/\mathrm{d}J of Kähler 1 -forms modulo exact forms, recovering an earlier result of Bloch [ 4 ] , who described the universal central extension of \mathfrak{sl}_{2}(A) for commutative associative unital algebras A . Specializing to the case where A=\mathbb{K}[a] is a polynomial ring in one variable, we have \widehat{\mathfrak{sl}}_{2}(\mathbb{K}[a])=\mathfrak{sl}_{2}(\mathbb{K}[a]) . In particular, any element a of a Jordan algebra J induces a Lie algebra homorphism \mathfrak{sl}_{2}(\mathbb{K}[a])\to\widehat{\mathfrak{sl}}_{2}(J) . (\widehat{\mathfrak{sl}}_{2}(J),{\rm SL}_{2}) -modules Set {\rm SL}_{2}={\mathbf{SL}}_{2}(\mathbb{K}) . An \mathfrak{sl}_{2} -module M derives from a rational {\rm SL}_{2} -module if and only if M is locally finite over \mathfrak{sl}_{2} , that is, U(\mathfrak{sl}_{2})v is finite dimensional for all v\in M . This condition is clearly equivalent to the requirement that M is a direct sum of finite dimensional simple \mathfrak{sl}_{2} -modules. By definition, an \widehat{\mathfrak{sl}}_{2}(J) -module M is an (\widehat{\mathfrak{sl}}_{2}(J),{\rm SL}_{2}) -module if its structure as a module over \mathfrak{sl}_{2} derives from a rational {\rm SL}_{2} -module. Such modules M admit an eigenspace decomposition M=\bigoplus_{k\in\mathbb{Z}}M_{k} , with M_{k}=\{m\in M\,|\,hm=km\} . The eigenspace decomposition for the adjoint module \mathfrak{G}:=\widehat{\mathfrak{sl}}_{2}(J) is a short grading [ 33 ] : \mathfrak{G}=\mathfrak{G}_{-2}\oplus\mathfrak{G}_{0}\oplus\mathfrak{G}_{2}, where \displaystyle\mathfrak{G}_{-2} \displaystyle= \displaystyle f\otimes J, \displaystyle\mathfrak{G}_{0} \displaystyle= \displaystyle(h\otimes J)\oplus\{J,J\}, \displaystyle\mathfrak{G}_{2} \displaystyle= \displaystyle e\otimes J. For the rest of the paper, we fix the notation \mathfrak{G}=\widehat{\mathfrak{sl}}_{2}(J) , so that (\widehat{\mathfrak{sl}}_{2}(J),{\rm SL}_{2}) -modules will simply be called (\mathfrak{G},{\rm SL}_{2}) -modules. J -spaces In the Lie algebra \mathfrak{G}_{0} , the subspace \{J,J\} is intricate. However, the \mathfrak{G}_{0} -modules can be conveniently described using the notion of a J -space. A vector space M endowed with a linear map \rho:\ J\rightarrow\hbox{End}(M) is called a J -space if (J1) [\rho(a),\rho(a^{2})]=0 , (J2) [[\rho(a),\rho(b)],\rho(c)]=4\rho(\partial_{a,b}\,c) , for all a,b,c\in J . It follows easily from the presentation of \mathfrak{G}_{0} that the \mathfrak{G}_{0} -module structures and the J -space structures on a vector space M are precisely equivalent. Indeed, (a) A \mathfrak{G}_{0} -module structure induces a J -space structure for which \rho(a) is the action of h(a) . (b) Conversely, a J -space structure extends to a \mathfrak{G}_{0} -module structure, where the action of h(a) is \rho(a) and the action of \{a,b\} is \frac{1}{4}[\rho(a),\rho(b)] . In the next section, we introduce the class of dominant J -spaces and give a number of examples in this context. We will also compare them with the classical notion of Jordan bimodules. A J -space is said to be of level n if \rho(1)=n\,{\rm id} . 3 Dominant J -spaces A J -space isomorphic to a J -subspace of the set H^{0}(\mathfrak{G}_{2},X) of \mathfrak{G}_{2} -invariants for some (\mathfrak{G},{\rm SL}_{2}) -module X is called dominant . In this section, we investigate the following question: Which J -spaces are dominant? In the trivial case where J=\mathbb{K} , the Lie algebra \mathfrak{G}_{0} is the Cartan subalgebra \mathbb{K}h of \widehat{\mathfrak{sl}}_{2}(J)=\mathfrak{sl}_{2} , so a J -space (V,\rho) is dominant precisely when \rho(1) is diagonalizable with nonnegative integer eigenvalues. Joseph functors and Weyl modules We now reformulate our question in terms of Weyl modules. Set \mathfrak{B}=\mathfrak{G}_{0}\oplus\mathfrak{G}_{2} , \mathfrak{b}=\mathbb{K}h\oplus\mathbb{K}e , and let {\rm B}\subset{\rm SL}_{2} be the Borel subgroup with Lie algebra \mathfrak{b} . Recall that a \mathfrak{b} -module V derives from a {\rm B} -module if h|_{V} is diagonalizable with integer eigenvalues and e|_{V} is locally nilpotent. By definition, a (\mathfrak{B},{\rm B}) -module is a \mathfrak{B} -module whose infinitesimal \mathfrak{b} -action derives from a B -module structure. Let V be a (\mathfrak{B},B) -module. The induced module M(V):=\mathrm{Ind}_{\mathfrak{B}}^{\mathfrak{G}}\,V, is called the generalized Verma module generated by V . Since \mathrm{ad}^{3}(e)=0 , the subspace Z(V):=\bigcap_{k\geq 0}\,e^{k}\,M(V) is a \mathfrak{G} -submodule. The \mathfrak{G} -module \Delta(V):=M(V)/Z(V) is called the Weyl module generated by V . It satisfies the universal property stated in the next lemma. Lemma 3.1 . For any (\mathfrak{B},B) -module V , the Weyl module \Delta(V) is a (\mathfrak{G},{\rm SL}_{2}) -module. Moreover, for any (\mathfrak{G},{\rm SL}_{2}) -module X , every morphism of \mathfrak{B} -modules \phi:V\to X uniquely extends to a morphism of \mathfrak{G} -modules \psi:\Delta(V)\to X . Proof. The first assertion follows from the fact that an \mathfrak{sl}_{2} -module X derives from an {\rm SL}_{2} -module if and only if (a) X is locally finite dimensional as a \mathfrak{b} -module, and (b) \bigcap_{k\geq 0}\,e^{k}X=0 . It follows that \Delta(V) is a (\mathfrak{G},{\rm SL}_{2}) -module. Moreover, every morphism of \mathfrak{B} -modules \phi:V\to X extends to a morphism of \mathfrak{G} -modules \psi:M(V)\to X that factors through \Delta(V) . ∎ Any J -space V of nonnegative integer level n can be seen as a (\mathfrak{B},{\rm B}) -module with a trivial action of \mathfrak{G}_{2} . Our question thus becomes: Which J -spaces V embed in the Weyl module \Delta(V) ? The answer will require two combinatorial formulas, to be stated and proved in the next two subsections. Remark: The notion of Weyl modules dates back to the 1980’s in the work of S. Donkin [ 11 ] and W. Koppinen [ 25 ] . For J=\mathbb{K}[a] , the functor V\mapsto\Delta(V) is the Joseph functor [ 22 ] for the affine Lie algebra A_{1}^{(1)} , see [ 28 , ch.III] and [ 8 ] . Garland’s formula If J=\mathbb{K}[a] is the polynomial ring in one variable, then \mathfrak{G}=\mathfrak{sl}_{2}(\mathbb{K}[a]) . Let \pi:\ U(\mathfrak{G})\rightarrow U(\mathfrak{G}_{-2}\oplus\mathfrak{G}_{0}) be the projection relative to the decomposition U(\mathfrak{G})=U(\mathfrak{G}_{-2}\oplus\mathfrak{G}_{0})\oplus U(\mathfrak{G})\mathfrak{G}_{2} . In what follows, we will compute \pi(e^{(m)}f(a)^{(n)}) , where we write x^{(m)} for the divided power x^{m}/m! . In fact, the computation will be done in a much larger algebra than U(\mathfrak{G}) . Set {\overline{U}}(\mathfrak{G})=\varprojlim\,U(\mathfrak{sl}_{2}(\mathbb{K}[a]/(a^{n})) , and let t be a formal variable. The following formula, due to Garland [ 17 ] and reformulated in [ 8 ] , involves formal series in the large algebra {\overline{U}}(\mathfrak{G})[[t]] . As neither reference included a proof, we prove the formula here. Lemma 3.2 . Let n\geq m\geq 0 be integers. Then \pi(e^{(m)}f(a)^{(n)}) is the coefficient of t^{m} in the formal series (-1)^{m}\left(\sum_{r=0}^{\infty}f(a^{r+1})t^{r}\right)^{(n-m)}\exp\left(-\sum_{s=1}^{\infty}\frac{h(a^{s})}{s}t^{s}\right). Proof. Set \Gamma:=\{G\in{\rm SL}_{2}(\mathbb{K}[[a,t]])\mid G(0,0)={\rm id}\} , and \mathfrak{P}:=\{g\in\mathfrak{sl}_{2}(\mathbb{K}[[a,t]])\mid g(0,0)=0\} . Since the exponential map \exp:\mathfrak{P}\to{\overline{U}}(\mathfrak{G})[[t]] is well defined, \Gamma is a subgroup of the group of invertible elements of {\overline{U}}(\mathfrak{G})[[t]] . An elementary matrix computation shows that \left({}^{1\,\hskip 2.84526pt-t}_{0\hskip 8.53581pt1}\right)\hskip 2.84526pt\left({}^{1\hskip 2.84526pt0}_{a\hskip 2.84526pt1}\right)=\left({}^{\hskip 11.38109pt1\qquad 0}_{a/(1-ta)\hskip 2.84526pt1}\right)\hskip 2.84526pt\left({}^{1-ta\,\qquad 0}_{\hskip 5.69054pt0\hskip 17.07164pt1/(1-ta)}\right)\hskip 2.84526pt\left({}^{1\hskip 2.84526pt\phi(a,t)}_{0\hskip 8.53581pt1}\right), (3.3) for some series \phi(a,t)=\sum_{k\geq 0}t^{k}\phi_{k}(a) in \mathbb{K}[[a,t]] with \phi(0,0)=0 . Recall that e,h, and f are the matrices e=\left({}^{0\,1}_{0\,0}\right),h=\left({}^{1\hskip 7.11317pt0}_{0\,-1}\right)\hskip 2.84526pt\hbox{and}f=\left({}^{0\,0}_{1\,0}\right), so Equation ( 3.3 ) can be written as the following identity in {\overline{U}}(\mathfrak{G})[[t]] : \exp(-te)\,\exp f(a)=\exp f(a/(1-ta))\exp\,h(\log(1-ta))\exp\big(\sum_{k\geq 0}t^{k}e(\phi_{k}(a))\big). (3.4) We extend \pi to a continous \mathbb{K}[[t]] -linear map \overline{\pi}:\overline{U}(\overline{\mathfrak{G}})[[t]]\rightarrow\overline{U}({\mathfrak{G}}_{-2}\oplus{\mathfrak{G}}_{0})[[t]], where {\overline{U}}({\mathfrak{G}}_{-2}\oplus{\mathfrak{G}}_{0}) is the closure of U(\mathfrak{G}_{-2}\oplus\mathfrak{G}_{0}) in {\overline{U}}(\mathfrak{G}) . Thus Relation ( 3.4 ) implies the identity \overline{\pi}(\exp(-te)\,\exp f(a))=\exp f(a/(1-ta))\,\exp\,h(\log(1-ta)). We decompose this expression according to the degree in t and the eigenvalue for \mathrm{ad}(h) . On the left side, the component of degree m and eigenvalue 2(m-n) is (-1)^{m}e^{(m)}f(a)^{(n)} . We now look at the right side. We have f(a/(1-ta))=\sum_{r=0}^{\infty}\,f(a^{r+1})t^{r} . Therefore the component of eigenvalue 2(m-n) on the right side is \left(\sum_{r=0}^{\infty}\,f(a^{r+1})t^{r}\right)^{(n-m)}\exp(-\sum_{s=1}^{\infty}\frac{h(a^{s})}{s}t^{s}), which proves Garland’s formula. ∎ The Girard-Newton formula We now introduce some notation. A partition \sigma=(\sigma_{1},\ldots,\sigma_{m}) is a nonincreasing sequence of positive integers, that is, \sigma_{1}\geq\cdots\geq\sigma_{m}\geq 1,\ \hbox{for some}\ m\geq 1 . We say that \sigma is a partition of n (in short, \sigma\vdash n ) if \sigma_{1}+\cdots+\sigma_{m}=n . For \sigma\vdash n , let C_{\sigma} be the conjugacy class of all permutations \tau\in S_{n} with cycle lengths \sigma_{1},\,\sigma_{2},\ldots,\sigma_{m} , let |C_{\sigma}| be its cardinality, and let \mathrm{sgn}(\sigma) be the common sign of the elements of C_{\sigma} . For any k\geq 1 , let N_{k}(x_{1},\dots,x_{n})=x_{1}^{k}+\cdots+x_{n}^{k} be the k^{th} -Newton polynomial in n indeterminates x_{1},\ldots,x_{n} . For an arbitrary partition \sigma=(\sigma_{1},\ldots,\sigma_{m})\vdash n , set N_{\sigma}(x)=N_{\sigma_{1}}(x)N_{\sigma_{2}}(x)\cdots N_{\sigma_{m}}(x) . The following identity is usually called the Girard-Newton formula . Lemma 3.5 . We have {\sum_{\sigma\ \vdash\ n}\,\hbox{sgn}(\sigma)\,|C_{\sigma}|\,N_{\sigma}(x)}=n!\,x_{1}\cdots x_{n} . Proof. We recall the standard proof of this 17 th century identity. Let V be the vector space \mathbb{K}^{n} and let h:V\to V be the diagonal matrix with diagonal entries x_{1},\dots,x_{n} . The symmetric group S_{n} acts on V^{\otimes n} by permutation of the factors. The Schur idempotent (1/n!)\sum_{\tau\in S_{n}}\hbox{sgn}(\tau)\tau is a projection of V^{\otimes n} onto the 1 -dimensional subspace \bigwedge^{n}V , so \mathrm{Tr}(h^{\otimes n}\circ\sum_{\tau\in S_{n}}\hbox{sgn}(\tau)\tau)=n!x_{1}\cdots x_{n} . Note that for \sigma\vdash n and \tau\in C_{\sigma} , we have \mathrm{Tr}(h^{\otimes n}\circ\tau)=N_{\sigma}(x) . It follows that \mathrm{Tr}(h^{\otimes n}\circ\sum_{\sigma\in S_{n}}\hbox{sgn}(\sigma)\sigma)=\sum_{\sigma\vdash n}\,\hbox{sgn}(\sigma)\,|C_{\sigma}|\,N_{\sigma}(x) , from which the lemma follows. ∎ Let A:=\mathbb{K}[Y_{1},Y_{2},\dots] be the polynomial algebra in infinitely many variables Y_{1},Y_{2},\dots . For any partition \sigma=(\sigma_{1},\dots,\sigma_{\ell}) of n , set Y_{\sigma}=Y_{\sigma_{1}}Y_{\sigma_{2}}\cdots Y_{\sigma_{\ell}} . Let t be an indeterminate. In A[[t]] , the formal expression \exp(-\sum_{k\geq 1}Y_{k}t^{k}/k) can be expanded as \sum_{k\geq 0}\,a_{k}t^{k} , where all a_{k} belongs to A . Lemma 3.6 . We have a_{n}=((-1)^{n}/n!)\,{\sum_{\sigma\ \vdash\ n}\hbox{sgn}(\sigma)|C_{\sigma}|Y_{\sigma}}. Proof. Both sides of the identity are polynomials involving only the variables Y_{1},\dots,Y_{n} . We will prove it for the elements Y_{k}=N_{k}=x_{1}^{k}+\cdots+x_{n}^{k} of the algebra A^{\prime}=\mathbb{K}[x_{1},\dots,x_{n}] . This is sufficient since N_{1},\dots,N_{n} are algebraically independent. Note that \displaystyle-\sum_{k\geq 1}N_{k}t^{k}/k \displaystyle=-\sum_{k\geq 1}\sum_{1\leq i\leq n}x_{i}^{k}t^{k}/k=-\sum_{1\leq i\leq n}\sum_{k\geq 1}x_{i}^{k}t^{k}/k=\sum_{1\leq i\leq n}\log(1-tx_{i}). Hence \exp(-\sum_{k\geq 1}N_{k}t^{k}/k)=\prod_{1\leq i\leq n}(1-tx_{i}). It follows that a_{n}=(-1)^{n}x_{1}\cdots x_{n} , and the identity now follows from Lemma 3.5 . ∎ The Dominance criterion After this combinatorial digression, we now answer the question about dominant J -spaces. Let (V,\rho) be a J -space and let a\in J . Since \{a^{n},a^{m}\}=0 for any positive integers n and m , the operators (\rho(a^{n}))_{n>0} commute with each other. For any partition \sigma=(\sigma_{1},\sigma_{2},\dots,\sigma_{m}) , we define \rho_{\sigma}(a):=\rho(a^{\sigma_{1}})\rho(a^{\sigma_{2}})\cdots\rho(a^{\sigma_{m}}) . Theorem 1 . A J -space (V,\rho) of level n is dominant if and only if \displaystyle{\sum_{\sigma\ \vdash\ n+1}\hbox{sgn}(\sigma)|C_{\sigma}|\rho_{\sigma}(a)=0} , for all a\in J . Proof. For any integer k , we denote by M_{k}(V) , Z_{k}(V) , and \Delta_{k}(V) the corresponding eigenspaces in M(V) , Z(V) , and \Delta(V) . By commutativity of \mathfrak{G}_{-2} , the enveloping algebra U(\mathfrak{G}_{-2}) is the symmetric algebra S\mathfrak{G}_{-2} , and by the Poincaré-Birkhoff-Witt Theorem, M(V)=U(\mathfrak{G}_{-2})V . It follows that M_{n-2k}(V)=S^{k}(\mathfrak{G}_{-2})V , for any integer k . Since \Delta(V) is an {\rm SL}_{2} -module and \Delta_{n+2}(V)=0 , we deduce that \Delta_{-n-2}(V)=0 . It follows that Z_{-n-2}(V)=M_{-n-2}(V) and therefore Z_{n}(V)=e^{n+1}M_{-n-2}(V)=e^{n+1}S^{n+1}(\mathfrak{G}_{-2})V . However, Z_{n}(V) is the kernel of the map V\to\Delta(V) and S^{n+1}(\mathfrak{G}_{-2}) is spanned by the elements f(a)^{n+1} with a\in J . Therefore, V is dominant if and only if e^{n+1}f(a)^{n+1}m=0 , for all a\in J and m\in V . This condition is equivalent to \pi(e^{n+1}f(a)^{n+1})m=0 . By Lemma 3.2 , \pi(e^{n+1}f(a)^{n+1}) is the coefficient of t^{n+1} of the formal series r\,\exp\left(-\sum_{s=1}^{\infty}\frac{h(a^{s})}{s}t^{s}\right) for some r\in{\mathbb{Q}}\setminus 0 , which, by Lemma 3.6 , is equal to r^{\prime}\,\,{\sum_{\sigma\ \vdash\ n+1}\hbox{sgn}(\sigma)|C_{\sigma}|\rho_{\sigma}(a)}, for some r^{\prime}\in{\mathbb{Q}}\setminus 0 . Hence V is dominant if and only if \sum_{\sigma\ \vdash\ n+1}\hbox{sgn}(\sigma)|C_{\sigma}|\rho_{\sigma}(a)=0, for all a\in J .∎ The algebra {\mathcal{U}}_{n}(J) For any partition \sigma=(\sigma_{1},\sigma_{2},\ldots,\sigma_{m}) and a\in J , let h_{\sigma}(a)\in U(\mathfrak{G}_{0}) be the element defined by: h_{\sigma}(a)=h(a^{\sigma_{1}})h(a^{\sigma_{2}})\cdots h(a^{\sigma_{m}}) . For n\geq 0 , let I_{n} be the two-sided ideal of U(\mathfrak{G}_{0}) generated by the elements h-n and {\displaystyle{\sum_{\sigma\vdash\ n+1}\mathrm{sgn}(\sigma)|C_{\sigma}|\,h_{\sigma}(a)},} (3.7) for all a\in J . Let {\mathcal{U}}_{n}(J)=U(\mathfrak{G}_{0})/I_{n} . We write \rho_{n}:J\to{\mathcal{U}}_{n}(J) for the natural map a\mapsto h(a)\mod I_{n} . Examples Theorem 1 states that the dominant J -spaces of level n are exactly the {\mathcal{U}}_{n}(J) -modules. Here are some examples. (a) It is clear that the algebra \mathcal{U}_{1}(J) is isomorphic to the unital special universal envelope S_{1}(J) , defined in [ 21 , Chap. II.1] . (b) The multiplication algebra of J is the subalgebra of \hbox{End}(J) generated by the multiplications L_{a} . It satisfies the obvious identities (b.1) L_{1}=1 (b.2) [L_{a^{2}},L_{a}]=0 (b.3) L_{b}L_{a}L_{c}+L_{c}L_{a}L_{b}+L_{(bc)a}=L_{a}L_{bc}+L_{c}L_{ab}+L_{b}L_{ac}. Following [ 21 , Chap. II.9 and II.11] , the unital universal multiplication envelope, denoted by U_{1}(J) in [ 21 ] , is the unital associative algebra generated by symbols L_{a} and defined by the relations (b.1-b.3). In fact, if M is a U_{1}(J) -module, then the split null extension J\oplus M is a Jordan algebra, which is unital by Relation (b.1). Relation (b.3) can be split into the following two identities: (b.3.1) 2(L_{a})^{3}+L_{a}^{3}=3L_{a}L_{a^{2}} (b.3.2) [[L_{a},L_{b}],L_{c}]=L_{\partial_{a,b}c} . There is thus an isomorphism from \mathcal{U}_{2}(J) to the unital universal multiplication envelope U_{1}(J) , sending h(a)\mapsto 2L_{a} . (c) For J=\mathbb{K}[a] , the proof of Theorem 1 shows that {\mathcal{U}}_{n}(J)\simeq\mathbb{K}[x_{1},\dots,x_{n}]^{S_{n}} . The isomorphism sends h(a^{k}) to N_{k} for any k . (d) Let V and V^{\prime} be dominant J -spaces of levels p and q . Since V\otimes V^{\prime} is the highest component of the (\mathfrak{G},{\rm SL}_{2}) -module \Delta(V)\otimes\Delta(V^{\prime}) , the J -space V\otimes V^{\prime} is dominant of level p+q . Hence there is an algebra morphism \Delta_{p,q}:{\mathcal{U}}_{p+q}(J)\to{\mathcal{U}}_{p}(J)\otimes{\mathcal{U}}_{q}(J) . (e) Let J be a unital commutative associative algebra. In general, \mathcal{U}_{n}(J) is not commutative, even for n=1 . The simplest example is the unital Jordan algebra J=\mathbb{K}\oplus V , where V is a 2 -dimensional ideal with V^{2}=0 . It is clear that \mathcal{U}_{1}(J) is the exterior algebra \wedge V , which is not commutative. Using the coproducts defined in Example (c), it follows that \mathcal{U}_{n}(J) is not commutative for any n\geq 1 . Assume now that J is an arbitrary unital commutative associative algebra containing an ideal \mathfrak{m} with \dim(\mathfrak{m}/\mathfrak{m}^{2})\geq 2 . Then \mathbb{K}\oplus V is a quotient of J , and therefore the algebra \mathcal{U}_{n}(J) is not commutative for any n\geq 1 . (f) There is a classical notion of Jordan modules, see [ 21 ] , and small variations depending on whether the Jordan algebras are assumed to be unital (as in the present paper) or not. For clarity, we call the \mathcal{U}_{0}(J) -modules the trivial Jordan modules , the \mathcal{U}_{1}(J) -modules the 1 -sided Jordan modules , and the \mathcal{U}_{2}(J) -modules the 2 -sided Jordan modules. In general, \mathcal{U}_{n}(J) -modules are not Jordan modules of any of the previous types for n\geq 3 . Indeed, for n\geq 3 , \mathcal{U}_{n}(\mathbb{K}[a]) is not isomorphic to any \mathcal{U}_{\ell}(J) for \ell\leq 2 . In fact, J would then be a commutative associative algebra of Gelfand-Kirillov dimension at least 2 . Example (e) would then show that \mathcal{U}_{\ell}(J) is not commutative. The terminology J -spaces and dominant J -spaces has been chosen to avoid confusion with the classical notion of Jordan modules. (g) Set \tilde{J}=J*\mathbb{K}[a] , where * denotes the free product. The Lie algebra \mathfrak{g}=\widehat{\mathfrak{sl}}_{2}(\tilde{J}) has a natural grading \mathfrak{g}=\bigoplus_{n\in\mathbb{Z}_{+}}\mathfrak{g}_{n} , with \mathfrak{g}_{0}=\widehat{\mathfrak{sl}}_{2}(J) and x(a) homogenous of degree 1 for all x\in\mathfrak{sl}_{2} . Let \mathcal{M}_{\bf T}(\widehat{\mathfrak{sl}}_{2}(J)) be the category of \widehat{\mathfrak{sl}}_{2}(J) -modules which decompose, as \mathfrak{sl}_{2} -modules, into sums of trivial and adjoint modules. By the universal property of the Joseph functor, it is clear that \Delta(\mathcal{U}_{2}(J)) is the rank 1 free module in the category \mathcal{M}_{\bf T}(\widehat{\mathfrak{sl}}_{2}(J)) . It follows easily that \Delta(\mathcal{U}_{2}(J)) is isomorphic to \mathfrak{g}_{1} , cf. [ 24 , Section 5.6] . (h) Viewing J as a J -space of level 2 , there is a natural \mathfrak{G} -equivariant homomorphism \pi:\Delta(J)\to\mathfrak{G} . The map \pi defines a (left) Leibniz bracket on \Delta(J) by [a,b]:=\pi(a).b for all a,b\in\Delta(J) , where \pi(a).b denotes the \mathfrak{G} -module action of \pi(a) on b . Indeed, \Delta(J) is the universal central cover of \mathfrak{G} in the category of Leibniz algebras. (i) Let J=\mathbb{K}[t,t^{-1}] . The Lie algebra \mathfrak{G}_{0} has a 1 -dimensional centre \mathfrak{z} . By an easy argument, \mathfrak{z} acts trivially on any Weyl module. Hence the map \mathfrak{G}_{0}\to{\mathcal{U}}_{n}(J) is not injective and \bigcap I_{n}\neq 0 . Indeed, it is easy to see that {\mathcal{U}}_{n}(J)\simeq\mathbb{K}[x_{1}^{\pm 1},\dots,x_{n}^{\pm 1}]^{S_{n}} for any n\geq 0 . (j) Let A be an associative commutative algebra. In their interesting paper [ 7 ] , V. Chari, G. Fourier, and T. Khandai investigate representations of the Lie algebra \mathfrak{g}\otimes A , where \mathfrak{g} is a finite dimensional simple Lie algebra. For \mathfrak{g}=\mathfrak{sl}_{2} , they define a certain commutative associative algebra A_{n\rho} . It can be shown that A_{n\rho} is the maximal commutative quotient of the algebra {\mathcal{U}}_{n}(A) . (k) Let X be an affine algebraic variety, and consider its ring of functions \mathbb{K}[X] as a Jordan algebra. When \dim X\geq 2 , Example (e) shows that \mathcal{U}_{n}(\mathbb{K}[X]) is not commutative for any n\neq 0 . Similarly, for singular curves X , the algebra \mathcal{U}_{n}(\mathbb{K}[X]) is not commutative. Assume now that X is a smooth curve. Then Example (i) shows that all algebras \mathcal{U}_{n}(\mathbb{K}[X]) are commutative whenever \hbox{H}^{1}_{DR}(X) is generated by logarithmic differentials, that is, when X has genus 0 . This raises the following question: For arbitrary smooth curves X , are the algebras \mathcal{U}_{n}(\mathbb{K}[X]) commutative? The algebras {\mathcal{U}}_{n}(J) for finite dimensional Jordan algebras J Let \mathfrak{g} be a finite dimensional Lie algebra, and let I\subseteq U(\mathfrak{g}) be a left ideal. Let \mathbb{K}=U_{0}(\mathfrak{g})\subseteq U_{1}(\mathfrak{g})\subseteq U_{2}(\mathfrak{g})\subseteq\cdots be the canonical filtration of the enveloping algebra. Lemma 3.8 . Let n be an integer and set {\mathcal{X}}=\{x\in\mathfrak{g}\mid x^{n}\in I+U_{n-1}(\mathfrak{g})\} . If {\mathcal{X}} generates the Lie algebra \mathfrak{g} , then I has finite codimension. Proof. Let I^{gr}\subset S\mathfrak{g} be the graded ideal associated to I , and let \sqrt{I^{gr}} be its radical. By definition, \sqrt{I^{gr}} contains \mathcal{X} . By Gabber’s Involutivity Theorem [ 15 ] , \sqrt{I^{gr}} is stable under the Poisson bracket of U(\mathfrak{g}) , hence \sqrt{I^{gr}} is the codimension 1 ideal \mathfrak{g}S\mathfrak{g} . Therefore I^{gr} and I have finite codimension. ∎ Lemma 3.9 . Assume that \dim J<\infty . (a) For any finite dimensional (\mathfrak{B},{\rm B}) -module V , the Weyl module \Delta(V) is finite dimensional. (b) The associative algebra {\mathcal{U}}_{n}(J) is finite dimensional for any n\geq 0 . Proof of (a). As an S\mathfrak{G}_{-2} -module, \Delta(V) is generated by the image of V in \Delta(V) . Hence each eigenspace \Delta_{k}(V) is finite dimensional and the eigenvalues are bounded by some integer N . By Lemma 3.1 , \Delta(V) is an {\rm SL}_{2} -module, so its eigenvalues are a finite set of integers in [-N,N] . Therefore \dim\,\Delta(V)<\infty . Proof of (b). For n\geq 0 , recall that {\mathcal{U}}_{n}(J)=U(\mathfrak{G}_{0})/I_{n} where the ideal I_{n} is defined as in Subsection 3.5 . Set {\mathcal{X}}=\{x\in\mathfrak{G}_{0}\mid x^{n+1}\in I+U_{n}(\mathfrak{G}_{0})\} . By Equation 3.7 , h(a)^{n+1} is a polynomial of degree at most n in h(a),h(a^{2}),\ldots,h(a^{n+1}) modulo I_{n} , which implies that h(a) belongs to {\mathcal{X}} . Since \mathfrak{G}_{0}=h(J)+[h(J),h(J)] , the Lie algebra \mathfrak{G}_{0} is finite dimensional and generated by {\mathcal{X}} . By Lemma 3.8 , the associative algebra {\mathcal{U}}_{n}(J) is finite dimensional.∎ Lemma 3.10 . Assume than J is simple and finite dimensional (a) The Lie algebra \widehat{\mathfrak{sl}}_{2}(J) is simple and \Delta(V) is a simple \widehat{\mathfrak{sl}}_{2}(J) -module for every simple dominant J -space V . (b) The associative algebras {\mathcal{U}}_{n}(J) are finite dimensional and semisimple. (c) The map V\mapsto\Delta(V) is a bijection from the class of simple dominant J -spaces to the class of finite dimensional simple \widehat{\mathfrak{sl}}_{2}(J) -modules L . The inverse bijection is L\mapsto H^{0}(\mathfrak{G}_{2},L) . Proof. It is well known that the Lie algebra TKK(J) is simple. Since any central extension of a finite dimensonal simple Lie algebra is trivial, we have \widehat{\mathfrak{sl}}_{2}(J)=TKK(J) , which proves Assertion (a). It is clear that the \mathfrak{G}_{0} -module \mathfrak{G}_{2} is simple and faithful. Thus the center of \mathfrak{G}_{0} is the 1-dimensional subspace \mathbb{K}h , \mathfrak{G}_{0}=\mathbb{K}h\oplus[\mathfrak{G}_{0},\mathfrak{G}_{0}] , and the Lie algebra [\mathfrak{G}_{0},\mathfrak{G}_{0}] is semisimple. By Lemma 3.9 (b), the algebra \mathcal{U}_{n}(J) is a finite dimensional quotient of U([\mathfrak{G}_{0},\mathfrak{G}_{0}]) , and therefore \mathcal{U}_{n}(J) is semisimple. Assertion (c) follows from the usual highest weight theory.∎ The algebras {\mathcal{U}}_{n}({\mathbb{A}}) for the Albert algebra {\mathbb{A}} We specialize the results of the previous subsection to the Albert algebra {\mathbb{A}} of 3\times 3 Hermitian matrices over the split octonion algebra. The Lie algebra \mathfrak{G}:=\widehat{\mathfrak{sl}}_{2}({\mathbb{A}})=TKK({\mathbb{A}}) is a split Lie algebra of type E_{7} [ 35 ] . Let \mathfrak{h}\subset\mathfrak{G}_{0} be a split Cartan subalgebra. There is a basis R of the root system such that all roots in \mathfrak{G}_{2} are positive. We can identify R with the basis \{\alpha_{1},\cdots,\alpha_{7}\} of the root system of E_{7} defined in [ 5 , ch. IV section 4.11] . Since \{\alpha_{1},\cdots,\alpha_{6}\} is a basis of the root system of [\mathfrak{G}_{0},\mathfrak{G}_{0}]\simeq E_{6} , we have \alpha_{1}(h)=\cdots=\alpha_{6}(h)=0 and \alpha_{7}(h)=2 . It follows that h=2h_{1}+3h_{2}+4h_{3}+6h_{4}+5h_{5}+4h_{6}+3h_{7}, (3.11) where h_{1},\dots,h_{7} is the corresponding basis of the dual root system. For any integral dominant weight \lambda , let L(\lambda) be the simple \mathfrak{G} -module with highest weight \lambda . Then H^{0}(\mathfrak{G}_{2},L(\lambda)) is a dominant {\mathbb{A}} -space of level \lambda(h) . Albert has proved that {\mathbb{A}} is exceptional and simple [ 1 ] , so {\mathcal{U}}_{1}({\mathbb{A}})=0 . This also follows from the fact that no coefficient in Equation 3.11 is 1 . When \lambda is the fundamental weight \omega_{2} or \omega_{7} , the dominant {\mathbb{A}} -space H^{0}(\mathfrak{G}_{2},L(\lambda)) is of level 3. We conclude: Lemma 3.12 . We have {\mathcal{U}}_{3}({\mathbb{A}})\neq 0 .∎ 4 Injectivity of the maps \rho_{n}:J\to{\mathcal{U}}_{n}(J) for n\geq 4 We have seen in Example 3.6 (h) that the map \mathfrak{G}_{0}\to{\mathcal{U}}_{n}(J) is not always injective, even for n\gg 0 . By contrast, the map \rho_{2} is obviously injective, and it is easy to deduce that \rho_{n} is injective for any even integer n\geq 2 . We now show that \rho_{n}:J\to{\mathcal{U}}_{n}(J) is injective, whenever n\geq 4 . When n is odd, our proof is based on Zelmanov’s classification theorem. A technical lemma Recall that the algebra \{0\} is called trivial . Lemma 4.1 . Assume that the algebra {\mathcal{U}}_{p}(J) is not trivial, and \rho_{q} is injective for some nonnegative integers p and q . Then \rho_{p+q} is injective Proof. Let \Delta_{p,q}:{\mathcal{U}}_{p+q}(J)\to{\mathcal{U}}_{p}(J)\otimes{\mathcal{U}}_{q}(J) , be the algebra morphism defined in Example 3.6 (f). We show that \Delta_{p,q}\circ\rho_{p+q}(a) is not zero for any a\in J\setminus 0 . We have \Delta_{p,q}\circ\rho_{p+q}(1)=(p+q)1\otimes 1 , so \Delta_{p,q}\circ\rho_{p+q}(a)\neq 0 if a is a scalar. Otherwise, we have \Delta_{p,q}\circ\rho_{p+q}(a)=\rho_{p}(a)\otimes 1+1\otimes\rho_{q}(a) , where \rho_{q}(a) and 1 are not proportional. Thus 1\otimes\rho_{q}(a) is not proportional to \rho_{p}(a)\otimes 1 . So \Delta_{p,q}\circ\rho_{p+q}(a)\neq 0 , which proves that \rho_{p+q} is injective. ∎ Zelmanov’s classification theorem The center of a simple Jordan algebra J is a field L . The Jordan algebra J is called an Albert algebra if \overline{L}\otimes J is the Albert algebra over the algebraic closure \overline{L} of L . The following theorem is proved in [ 37 ] , [ 38 ] . Zelmanov Classification Theorem 4.2 . Any simple Jordan algebra is either special or is an Albert algebra. Corollary 4.3 . For any Jordan algebra J\neq 0 , the algebra {\mathcal{U}}_{3}(J) is not trivial. Proof. By Zorn’s Lemma, J has a simple quotient. Thus it is enough to prove the assertion when J is simple. If J is special, the map \rho_{3} is injective by Lemma 4.1 . Otherwise, J is an Albert algebra by Zelmanov’s Classification Theorem 4.2 , and it follows from Lemma 3.12 that {\mathcal{U}}_{3}(J) is not trivial. ∎ The injectivity theorem Theorem 2 . The map \rho_{n}:J\to{\mathcal{U}}_{n}(J) is injective for any n\geq 4 . Proof. Since \rho_{2} is injective, an iterative use of Lemma 4.1 shows that \rho_{n} is injective for any even integer n\geq 2 . For an odd integer n\geq 5 , the algebra {\mathcal{U}}_{3}(J) is not trivial and it has been proved that \rho_{n-3} is injective. So by Lemma 4.1 , the map \rho_{n} is injective. ∎ This leaves open two questions: is there an elementary proof of Theorem 2 ? Is it always true that \rho_{3} is injective? 5 Finite generation, finite presentation, and FP_{\infty} As a motivation for considering the FP_{\infty} condition, we first state two finiteness results about Jordan algebras. Then we show that if \widehat{\mathfrak{sl}}_{2}(J(D+1)) is FP_{\infty} , then the graded dimension conjecture holds for the free Jordan algebra J(D) . Finite generation and finite presentation We now state two finiteness results. Theorem 3 . The associative algebra {\mathcal{U}}_{n}(J) is finitely generated, for all finitely generated Jordan algebras J . This generalizes the following known results for a Jordan algebra generated by a finite set x_{1},\dots,x_{n} of elements: (a) The universal special unital envelope {\mathcal{U}}_{1}(J) is generated by the the elements \rho_{1}(x_{1}) , \dots , \rho_{1}(x_{n}) . (b) The unital univeral multiplication envelope {\mathcal{U}}_{2}(J) is generated by the elements \rho_{2}(x_{i}) and \rho_{2}(x_{i}\cdot x_{j}) , for 1\leq i\leq j\leq n . (c) If J is an associative algebra A , it has been proved in [ 7 ] that the maximal commutative quotient A_{n\rho} of {\mathcal{U}}_{n}(A) is finitely generated. Another finiteness result is the following: Theorem 4 . The Lie algebra \widehat{\mathfrak{sl}}_{2}(J) is finitely presented, for all finitely presented Jordan algebras J . The proof of both theorems can be reduced to free Jordan algebras. The proofs are postponed to Subsections 11.2 and 11.4 . Koszul lemma Let \mathfrak{g}=\mathfrak{s}\ltimes\mathfrak{u} be a Lie algebra which is a semi-direct product. We write H_{*}(\mathfrak{g},\mathfrak{s}) for the relative homology , that is, the homology of the Chevalley-Eilenberg complex of coinvariants \bigwedge^{p}(\mathfrak{g}/\mathfrak{s})\big/\big(\mathfrak{s}\cdot\bigwedge^{p}(\mathfrak{g}/\mathfrak{s})\big) . See [ 26 ] or [ 27 ] for details. The following lemma is essentially due to Koszul [ 26 , Theorem 17.3] . Lemma 5.1 . Assume that \mathfrak{s} is a finite dimensional simple Lie algebra and \mathfrak{u} is locally finite as an \mathfrak{s} -module. Then H_{*}(\mathfrak{g}) is isomorphic to H_{*}(\mathfrak{s})\otimes H_{*}(\mathfrak{g},\mathfrak{s}) . In fact, Theorem 17.3 in [ 26 ] only involves finite dimensional Lie algebras, but the proof in our setting is exactly the same. Action of the Witt algebra Let J(D) be the free Jordan algebra on the generators x_{1},\dots,x_{D} . For n\geq-1 , set L_{n}:=\sum_{i=1}^{D}\,x_{i}^{n+1}{\partial\over\partial x_{i}} . We have [L_{n},L_{m}]=(m-n)L_{m+n} , so the linear span W of the L_{n} is a Lie algebra of derivations of J isomorphic to the Witt algebra \mathrm{Der}(\mathbb{K}[t]) . Lemma 5.2 . For any k\geq 4 , the space H_{k}(\widehat{\mathfrak{sl}}_{2}(J(D))) is either \{0\} or is infinite dimensional. Proof. Assume k\geq 4 . The derivation L_{0} acts on the component J_{n}(D) by multiplication by n . Since all eigenvalues of L_{0} on \Lambda^{k}(\widehat{\mathfrak{sl}}_{2}(J(D))) are positive integers, it follows that L_{0}.H_{k}(\widehat{\mathfrak{sl}}_{2}(J(D)))=H_{k}(\widehat{\mathfrak{sl}}_{2}(J(D))) . Hence H_{k}(\widehat{\mathfrak{sl}}_{2}(J(D))) is either zero, or is a nontrivial W -module. In the latter case H_{k}(\widehat{\mathfrak{sl}}_{2}(J(D))) is infinite dimensional since W is a simple infinite dimensional Lie algebra. ∎ The FP_{\infty} -property By definition, a Lie algebra \mathfrak{g} is of type FP_{n} if the trivial module \mathbb{K} admits a projective resolution 0\leftarrow\mathbb{K}\leftarrow C_{0}\leftarrow C_{1}\leftarrow C_{2}\leftarrow\cdots where the modules C_{i} are finitely generated for all i\leq n . We say that \mathfrak{g} is of type FP_{\infty} if it is of type FP_{n} for all integers n . Proposition 5.3 . The Lie algebra \widehat{\mathfrak{sl}}_{2}(J(1)) is of type FP_{\infty} . Proof. By the main result of Garland and Lepowsky’s paper [ 16 ] there is a resolution of the trivial \mathfrak{sl}_{2}(t\mathbb{K}[t]) -module \mathbb{K} 0\leftarrow\mathbb{K}\leftarrow C_{0}\leftarrow C_{1}\leftarrow C_{2}\dots where C_{n} is a free U(\mathfrak{sl}_{2}(t\mathbb{K}[t])) -module of rank 2n+1 for any n\geq 0 , so \mathfrak{sl}_{2}(t\mathbb{K}[t]) is of type FP_{\infty} . Since \widehat{\mathfrak{sl}}_{2}(J(1))=\mathfrak{sl}_{2}(J(1))=\mathfrak{sl}_{2}\ltimes\mathfrak{sl}_{2}(t\mathbb{K}[t]) the Lie algebra \widehat{\mathfrak{sl}}_{2}(J(1)) is also of type FP_{\infty} . ∎ The FP_{\infty} condition By the previous proposition, the Lie algebra \widehat{\mathfrak{sl}}_{2}(J(1)) is of type FP_{\infty} . Moreover, by Theorem 4 , the Lie algebra \widehat{\mathfrak{sl}}_{2}(J(D)) is of type FP_{2} for any D . Also Theorems 3 and 7 suggest that \widehat{\mathfrak{sl}}_{2}(J(D)) enjoys some further finiteness properties. So it is natural to suspect that \widehat{\mathfrak{sl}}_{2}(J(D)) is of type FP_{\infty} . Theorem 5 . Assume that \widehat{\mathfrak{sl}}_{2}(J(D+1)) is of type FP_{\infty} . Then the conjectural formula of [ 24 ] for \dim J_{n}(D) holds. Proof. By hypothesis, each homology space H_{k}(\widehat{\mathfrak{sl}}_{2}(J(D+1))) is finite dimensional. So, by Lemma 5.2 we have H_{k}(\widehat{\mathfrak{sl}}_{2}(J(D+1)))=0 , for any k\geq 4 . By Koszul’s Lemma 5.1 , there is an isomorphism H_{*}(\mathfrak{sl}_{2}(J(D+1)))=H^{*}(\mathfrak{sl}_{2})\otimes H_{*}(\mathfrak{sl}_{2}(J(D+1)),\mathfrak{sl}_{2}) . Since H_{3}(\mathfrak{sl}_{2}) is nonzero, we deduce that H_{k}(\widehat{\mathfrak{sl}}_{2}(J(D+1)),\mathfrak{sl}_{2})=0 , for any positive integer k . Thus by [ 24 , Theorem 2] , this implies the conjectural formula for \dim J_{n}(D) . ∎ Remark. Our notation here is different from [ 24 ] . In [ 24 ] , the free nonunital Jordan algebra is denoted J(D) , while the unital free Jordan algebra is denoted J^{u}(D) . Also it should be noted that H_{*}(\widehat{\mathfrak{sl}}_{2}(J(D)),\mathfrak{sl}_{2})=H_{*}(\widehat{\mathfrak{sl}}_{2}(J_{+}(D)))^{\mathfrak{sl}_{2}} , where J_{+}(D) is the free nonunital Jordan algebra on D generators. Recent developments. In a very recent arXiv preprint [ 13 ] posted when the present paper was under final revisions, Dotsenko and Hentzel have provided detailed computational data showing that the dimension formula conjectured in [ 24 ] does not always hold. Theorem 5 thus admits the following, perhaps surprising, corollary: Corollary 5.4 . For D\geq 3 , the Lie algebra \widehat{\mathfrak{sl}}_{2}(J(D)) is not of type FP_{\infty} .∎ PART B: Good Filtrations in LA+ categories 6 LA+ categories Definition of LA categories Let \mathcal{C} be \mathbb{K} -linear abelian category. We say that \mathcal{C} is a locally artinian category (LA category) if (LA1) \mathcal{C} admits direct limits \varinjlim\,A_{n} for any directed family A_{0}\to A_{1}\to A_{2}\to\cdots of objects indexed by \mathbb{Z}_{\geq 0} , (LA2) the functor \varinjlim is exact, and (LA3) every object is the directed union \bigcup_{n\geq 0}A_{n} of a directed family of subobjects A_{n} of finite length. By definition, a directed union \bigcup_{n\geq 0}A_{n} will be a direct limit \varinjlim\,A_{n} of a directed family (A_{n})_{n\geq 0} where all maps A_{n}\to A_{n+1} are injective. For concreteness, all directed unions will be indexed by \mathbb{Z}_{\geq 0} , so Axiom (LA3) implies that all objects have finite or countable length. In this paper, we handle inductive limits cautiously. In fact, from the time of the groundbreaking paper of Cline, Parshall, and Scott [ 9 ] , it has been known that the paper [ 34 ] about the Grothendieck conditions is not fully correct [ 10 , 32 ] . These conditions are implied by our Axiom (LA2). Definition of LA+ categories Let {\mathcal{C}} be an LA category. Recall that the socle of an object X , denoted by \mathrm{Soc}(X) , is the largest semi-simple object in X . Indeed the axioms of an LA category ensure the existence of such an object. An injective envelope of an object X is an injective object I\supseteq X such that \mathrm{Soc}\,I=\mathrm{Soc}\,X . The first two axioms of an LA+ category deal with simple objects and their socles: (AX1) There is a set \{L(n)\mid n\in\mathbb{Z}_{\geq 0}\} of objects in {\mathcal{C}} which is a complete list of nonisomorphic simple objects. (AX2) We assume that \hbox{End}_{\mathcal{C}}(L(n))=\mathbb{K} and that L(n) admits an injective envelope I(n) in \mathcal{C} , for all n\geq 0 . Some additional definitions are needed to state the third axiom. A filtration {\mathcal{G}}_{*} of an object M is a directed family ({\mathcal{G}}_{n}M)_{n\geq 0} of subobjects for which M is the directed union \bigcup_{n\geq 0}{\mathcal{G}}_{n}M . If {\mathcal{G}}_{n}M=M for n\gg 0 , the filtration is said to be finite ; otherwise, it is infinite . For any n\geq 0 , set \overline{\mathcal{G}}_{n}M={\mathcal{G}}_{n}M/{\mathcal{G}}_{n-1}M , with the convention that \overline{\mathcal{G}}_{0}M={\mathcal{G}}_{0}M . The objects \overline{\mathcal{G}}_{n}M are called the successive quotients of the filtration {\mathcal{G}}_{*} . Let {\mathcal{C}}(n) be the subcategory of objects M which are constructed as finite or infinite extensions between the modules L(0),L(1),\ldots,L(n) . That is, M admits a filtration {\mathcal{G}}_{*} whose successive quotients \overline{\mathcal{G}}_{m}M are 0 or L(k) for some k=k(m)\leq n . For M\in{\mathcal{C}} , let {\mathcal{F}}_{n}\,M be the biggest subobject in {\mathcal{C}}(n) . The filtration {\mathcal{F}}_{0}M\subseteq{\mathcal{F}}_{1}M\subseteq{\mathcal{F}}_{2}M\subseteq\cdots is called the canonical filtration of M . By definition, the standard object \nabla(n) is the inverse image in I(n) of {\mathcal{F}}_{n-1}(\,I(n)/L(n)) . Remarks: (a) Axiom (AX1) provides a linear ordering L(0)\leq L(1)\leq L(2)\leq\cdots of the simple objects L(n) , and the definitions of the canonical filtration {\mathcal{F}}_{*} and the standard objects \nabla(n) depend on the chosen order. (b) According to the context, the standard objects are sometimes called costandard objects or dual of Weyl modules. We follow Donkin’s notation [ 11 ] . These objects are denoted A(n) in [ 9 ] . Our last axiom is (AX3) For any n\geq 0 , the standard object \nabla(n) has finite length. We say that the LA category {\mathcal{C}} is a LA category plus additional axioms (for short, an LA+ category ) if it satisfies the three axioms (AX1), (AX2) and (AX3). Existence of injective envelopes in LA+ categories In our setting, we did not require that {\mathcal{C}} has enough injectives because this follows from the hypotheses, as is proved now. Lemma 6.1 . Let {\mathcal{C}} be an LA+ category. (a) Let (I_{n})_{n\in\mathbb{Z}_{+}} be a family of injective objects in {\mathcal{C}} . Then \bigoplus_{n\geq 0}\,I_{n} is an injective object. (b) Any object in {\mathcal{C}} admits an injective envelope. Proof of (a). Set I=\bigoplus_{n\geq 0}\,I_{n} . Let X be an object of {\mathcal{C}} , let Y\subseteq X be a subobject, and let g:Y\rightarrow I be a morphism. We claim that g can be extended to a morphism f:X\to I . By Axiom (LA3), X is a directed union \bigcup_{n\geq 0}X_{n} of finite length subobjects. Set Y_{n}=Y\cap X_{n} and I_{\leq n}=\oplus_{k\leq n}I_{k} . We have g(Y_{n})\subseteq I_{\leq\phi(n)} , for some nondecreasing function \phi:\mathbb{Z}_{\geq 0}\to\mathbb{Z}_{\geq 0} . If needed, we can replace I_{n} by \bigoplus_{\phi(n-1)0 , define {\mathcal{H}}_{m}M as the inverse image of {\mathcal{H}}(M/{\mathcal{H}}_{m-1}M) in M . Thanks to the previous lemma, we verify by induction that each quotient M/{\mathcal{H}}_{m-1}M is an extended \nabla(n) -module. Therefore the procedure is well defined, and by construction, the successive quotients are in {\mathcal{S}}^{0}(n) . Lemma 7.3 . The nondecreasing sequence ({\mathcal{H}}_{m}M)_{m\geq 0} is a standard filtration for the doubly extended \nabla(n) -object M . In particular, any doubly extended \nabla(n) -object is an extended \nabla(n) -object. Proof. It has just been proved that the successive quotients are direct sums of copies of \nabla(n) . To prove that the family ({\mathcal{H}}_{m}M)_{m\geq 0} is a standard filtration, it remains to prove that M=\bigcup_{m\geq 0}{\mathcal{H}}_{m}M . For m\geq 0 , the m^{th} -socle of M , denoted by \mathrm{Soc}^{m}\,M , is inductively defined by (a) \mathrm{Soc}^{0}\,M=\mathrm{Soc}\,M and (b) \mathrm{Soc}^{m}\,M is the inverse image of \mathrm{Soc}(M/\mathrm{Soc}^{m-1}M) in M . By Axiom (LA3), we have M=\bigcup_{n\geq 0}\,\mathrm{Soc}^{m}\,M . Since {\mathcal{H}}_{m}M contains \mathrm{Soc}^{m}\,M , we have M=\bigcup_{m\geq 0}{\mathcal{H}}_{m}M . Therefore the family of subobjects ({\mathcal{H}}_{m}M)_{m\geq 0} exhausts M . ∎ The filtration {\mathcal{H}}_{*} is called the canonical standard filtration of type \nabla(n) for M . Quotients of extended \nabla(n) -objects Lemma 7.4 . Let M,\,N be extended \nabla(n) -objects with N\subseteq M . Then the quotient M/N is an extended \nabla(n) -object. Proof. It follows from the definition that {\mathcal{H}}_{0}N=N\cap{\mathcal{H}}_{0}M . Thus by induction, we have {\mathcal{H}}_{m}N=N\cap{\mathcal{H}}_{m}M , for all m\geq 0 . Let {\mathcal{G}}_{*} be the filtration of M/N induced by the filtration {\mathcal{H}}_{*} of M . There is a short exact sequence 0\to\overline{\mathcal{H}}_{m}N\to\overline{\mathcal{H}}_{m}M\to\overline{\mathcal{G}}_{m}M/N\to 0 . By hypothesis, \overline{\mathcal{H}}_{m}N and \overline{\mathcal{H}}_{m}M are direct sums of the standard object \nabla(n) . By Lemma 7.1 (c), the category {\mathcal{S}}^{0}(n) is semi-simple. Thus \overline{\mathcal{G}}_{m}(M/N) is a direct sum of standard objects \nabla(n) . Hence {\mathcal{G}}_{*} is a standard filtration of M/N , which proves that M/N is an extended \nabla(n) -object. ∎ 8 Good filtrations in LA+ categories Let {\mathcal{C}} be an LA+ category. A filtration {\mathcal{G}}_{*} of an object M\in{\mathcal{C}} is called good if, for all k\geq 0 , \overline{\mathcal{G}}_{k}M is an extended \nabla(n) -object for some n=n(k) . To the best of our knowledge, good filtrations originated in the context of algebraic group representations, in the work of Humphreys and Jantzen [ 19 ] . In this setting, the canonical filtration first appeared in [ 14 ] , and our presentation closely follows [ 29 ] . We will show that the canonical filtration of M is good whenever M admits a good filtration. Then we show that a quotient M/N has a good filtration whenever both objects M and N have good filtrations. In the context of algebraic groups, the last statement is due to Donkin [ 12 ] . Our generalization is based on the approach of [ 30 ] . The functor R^{1}{\mathcal{F}}_{m} Let \mathcal{F}_{0}M\subseteq\mathcal{F}_{1}M\subseteq\mathcal{F}_{2}M\subseteq\cdots be the canonical filtration defined in Section 6.2 . For m\geq 0 , the endofunctor \mathcal{F}_{m}:\mathcal{C}\rightarrow\mathcal{C} is left exact by construction. Let R^{1}{\mathcal{F}}_{m} be its first right derived functor. Lemma 8.1 . For any M with a good filtration, we have R^{1}{\mathcal{F}}_{m}M=0 , for all m\geq 0 . Proof. First we prove that R^{1}{\mathcal{F}}_{m}\nabla(n)=0 , or equivalently, that \phi:{\mathcal{F}}_{m}I(n)\to{\mathcal{F}}_{m}(I(n)/\nabla(n)) is surjective. Let X be the inverse image in I(n) of {\mathcal{F}}_{m}(I(n)/\nabla(n)) . By definition, there is a short exact sequence 0\to\nabla(n)\to X\to{\mathcal{F}}_{m}(I(n)/\nabla(n))\to 0 . If mn-1 . That is, m=n , which proves the first claim. Since \overline{\mathcal{G}}_{k}M is an extended \nabla(n) -object, it lies in {\mathcal{F}}_{n}(M/{\mathcal{G}}_{k-1}M) . By Lemma 8.1 , we have R^{1}{\mathcal{F}}_{n}({\mathcal{G}}_{k-1}M)=0 , and therefore {\mathcal{F}}_{n}(M/{\mathcal{G}}_{k-1}M)={\mathcal{F}}_{n}M/({\mathcal{G}}_{k-1}M\cap{\mathcal{F}}_{n}M)=(\mathcal{F}_{n}M+\mathcal{G}_{k-1}M)/\mathcal{G}_{k-1}M , which proves that {\mathcal{G}}_{k}M\subseteq{\mathcal{G}}_{k-1}M+{\mathcal{F}}_{n}M . Step 2 Since the Zassenhaus condition is satisfied, Lemma 8.2 implies that \overline{\mathcal{F}}_{n}M has a filtration whose successive quotients are the objects \overline{\mathcal{G}}_{k}M for k\in A(n) . We have also proved that for k\in A(n) , the object \overline{\mathcal{G}}_{k}M is an extended \nabla(n) -object. By Lemma 7.3 , any doubly extended \nabla(n) -object is an extended \nabla(n) -object. Hence the canonical filtration of M is good. ∎ Quotients of objects with a good filtration We conclude with the main result of this section. Proposition 8.4 . Let N\subseteq M be objects in {\mathcal{C}} with a good filtration. Then (a) M/N has a good filtration, and (b) {\mathcal{F}}_{n}(M/N)={\mathcal{F}}_{n}M/{\mathcal{F}}_{n}N , for all n\geq 0 . Proof. Let n\geq 0 . Assertion (b) is a consequence of Lemma 8.1 . It follows that \overline{\mathcal{F}}_{n}(M/N) is isomorphic to \overline{\mathcal{F}}_{n}M/\overline{\mathcal{F}}_{n}N . By Lemma 8.3 , \overline{\mathcal{F}}_{n}M and \overline{\mathcal{F}}_{n}N are extended \nabla(n) -objects. Thus by Lemma 7.4 , \overline{\mathcal{F}}_{n}(M/N) is an extended \nabla(n) -object. Hence the canonical filtration of M/N is good, which proves Assertion (a). ∎ Remark: The proof of Lemma 8.4 shows that the canonical filtration of M/N is induced by the canonical filtration of M . By contrast, when M\subseteq N are extended \nabla(n) -objects, the proof of Lemma 7.4 shows that the canonical standard filtration of M induces a standard filtration of M/N , but not necesarily the canonical one. 9 Definition of highest weight categories and generalized highest weight categories Cline, Parshall, and Scott [ 9 ] introduced the notion of a highest weight category (HW category). The most important exemples of HW categories are the category \mathcal{O} , the category of modular representations of Chevalley groups, and the category of perverse sheaves with respect to a stratification by cells. These categories enjoy many cohomology vanishing results. For the applications we have in mind, we will need to work in a somewhat more general setting. We relax one axiom of a HW category to define the notion of a generalized highest weight category (generalized HW category), and prove a useful cohomology vanishing result in this setting. Definition of HW categories and generalized HW categories Let {\mathcal{C}} be a LA+ category, with the notation given in Section 6.2 . The category {\mathcal{C}} is called a HW category if it satisfies the following two axioms: (HW1) the canonical filtration of I(n) is good, and (HW2) {\mathcal{F}}_{n}I(n)=\nabla(n) , for all n\geq 0 . When only the first axiom (HW1) is satisfied, the category {\mathcal{C}} is called a generalized HW category . Assume that {\mathcal{C}} is a generalized HW category. Since \mathrm{Ext}^{1}_{{\mathcal{C}}}(\nabla(n),\nabla(n))={\rm Hom}_{\mathcal{C}}(\nabla(n),I(n)/\nabla(n)) , we conclude that the axiom (HW2) is equivalent to \mathrm{Ext}^{1}_{{\mathcal{C}}}(\nabla(n),\nabla(n))=0 , for all n\geq 0 . Thus, a HW category is a generalized HW category where the extended \nabla(n) -objects are direct sums of standard modules, for all n\geq 0 . Examples of generalized HW categories For simplicity, we have assumed that the category \mathcal{C} contains countably many nonisomorphic simple objects. There are similar definitions when \mathcal{C} contains only finitely many objects. (a) A basic example of a HW category is \mathrm{Mod}(UT_{n}(\mathbb{K})) , where UT_{n}(\mathbb{K}) is the associative algebra of n\times n upper triangular matrices [ 9 ] . (b) For a finite dimensional local \mathbb{K} -algebra A , let UT_{n}(A)=UT_{n}(\mathbb{K})\otimes A . Then \mathrm{Mod}(UT_{n}(A)) is a basic example of a generalized HW category. (c) With the usual basis \{e,h,f\} of \mathfrak{sl}_{2}(\mathbb{K}) , let {\mathcal{O}}^{\prime}_{0} be the category of \mathfrak{sl}_{2}(\mathbb{K}) -modules M such that (i) The Casimir operator h^{2}/2+ef+fe acts locally nilpotently, (ii) M=\bigoplus_{n\leq 0}M^{(n)} , where M^{(n)} is the generalized eigenspace M^{(n)}=\{m\in M\mid(h-n)^{N}.m=0 for N\gg 0\} . The category {\mathcal{O}}^{\prime} has two simple objects, the Verma module S(0):=M(-2) of highest weight -2 , and the trivial module S(1)=\mathbb{K} . Relative to this order, {\mathcal{O}}^{\prime} is a generalized HW category. We have \nabla(0)=M(-2) and \nabla(1) is the {\mathcal{O}} -dual of the Verma module M(0) . This example easily generalizes to any simple Lie algebra. About the definition of HW and generalized HW categories Our definition of HW categories is more constructive and essentially equivalent to the original one, namely Definition 3.1 of [ 9 ] . First, it is easy to see that the postulated object A(n) in [ 9 ] is necessarily the standard module \nabla(n) . Next, by Lemma 8.3 (b), Axiom (HW1) is equivalent to (HW 1^{\prime} ) I(n)/\nabla(n) has a good filtration. Therefore {\mathcal{C}} is a generalized HW category if and only if it satisfies the following axiom: (HW) For all n\geq 0 , I(n)/\nabla(n) has a filtration {\mathcal{G}}_{*} such that, for all k\geq 0 , \overline{\mathcal{G}}_{k}(I(n)/\nabla(n)) is an extended \nabla(m_{k}) -object for some m_{k}>n . However, in our setting, it would be unnatural to require Axiom 3.1(c)(iii) of [ 9 ] , namely that the multiplicities [I(n):\nabla(m)] are always finite. That’s why our definition of a good filtration is a bit different. While Cline, Parshall, and Scott [ 9 ] required that successive quotients be standard objects \nabla(n) , we require only that these quotients be extended \nabla(n) -objects. Moreover, our index set \mathbb{Z}_{\geq 0} has a minimal element. Thus Axiom 3.1(b) of [ 9 ] means that standard objects have finite length, i.e., this axiom is equivalent to our Axiom (AX3). 10 Ext-vanishing in generalized HW categories In Sections 7 and 8 , we have exhausted all elementary properties of good filtrations in LA+ categories. We now use these properties to prove an Ext-vanishing result for generalized HW categories. For clarity, we have also included the corresponding statement for HW categories. From now on, let {\mathcal{C}} be a generalized HW category. Height of objects in the category {\mathcal{C}} By definition, the height \mathrm{ht}\,X of an object X is the smallest integer n such that \mathcal{F}_{n}X\neq 0 . Equivalently, it is the smallest integer n such that X contains a copy of L(n) . Therefore, \mathrm{ht}\,X=\mathrm{ht}(\mathrm{Soc}\,X) . Lemma 10.1 . Let X be an object with a good filtration, and let I\supseteq X be an injective envelope of X . (a) The quotient I/X has a good filtration. (b) We have \mathrm{ht}\,I=\mathrm{ht}\,X and \mathrm{ht}\,I/X\geq\mathrm{ht}\,X . (c) Moreover, if {\mathcal{C}} is a HW category, then \mathrm{ht}\,I/X>\mathrm{ht}\,X . Proof. By Axiom (HW1), I has a good filtration. Therefore, I/X has a good filtration by Proposition 8.4 (a), which proves Assertion (a). Since X and I share the same socle, we have \mathrm{ht}\,I=\mathrm{ht}\,X . Set n=\mathrm{ht}\,X . By Proposition 8.4 (b), we have {\mathcal{F}}_{n-1}(I/X)={\mathcal{F}}_{n-1}I/{\mathcal{F}}_{n-1}X , hence {\mathcal{F}}_{n-1}(I/X)=0 and \mathrm{ht}(I/X)\geq n=\mathrm{ht}(X) , which proves Assertion (b). Assume now that {\mathcal{C}} is a HW category. The socle of X can be decomposed as \mathrm{Soc}\,X=L(n)^{m}\oplus Y , where \mathrm{ht}\,Y>n , and m is a positive integer or \aleph_{0} . By Proposition 8.4 (a), there is a short exact sequence 0\to{\mathcal{F}}_{n}X\to{\mathcal{F}}_{n}I\to{\mathcal{F}}_{n}(I/X)\to 0 . Since I(n)^{m} is the injective envelope of {\mathcal{F}}_{n}\mathrm{Soc}X=L(n)^{m} , we have {\mathcal{F}}_{n}I={\mathcal{F}}_{n}I(n)^{m}=\nabla(n)^{m} . By Lemma 7.1 (b), the subobject \nabla(n)^{m} lies in X , and therefore {\mathcal{F}}_{n}(I/X)=0 , i.e. \mathrm{ht}(I/X)>n=\mathrm{ht}\,X . ∎ Minimal morphisms and minimal resolutions A morphism f:X\to Y is called minimal if \mathrm{Soc}X\subseteq{\rm Ker}\,f and \mathrm{Soc}Y\subseteq\mathrm{Im}\,f . A resolution 0\to X\mathrel{\mathop{\kern 0.0pt\rightarrow}\limits^{d_{0}}}I_{0}\mathrel{\mathop{\kern 0.0pt\rightarrow}\limits^{d_{1}}}I_{1}\mathrel{\mathop{\kern 0.0pt\rightarrow}\limits^{d_{2}}}I_{2}\mathrel{\mathop{\kern 0.0pt\rightarrow}\limits^{d_{3}}}\cdots of an object X is called minimal if the morphisms d_{i} are minimal for all i>0 . Equivalently, it means that the differential of the complex \mathrm{Soc}\,I_{0}\to\mathrm{Soc}\,I_{1}\to\mathrm{Soc}\,I_{2}\to\cdots is zero. It follows from Lemma 6.1 that each object M\in{\mathcal{C}} admits a minimal injective resolution: indeed, we can take an injective envelope I_{0} of M and then build the resolution inductively, letting I_{n} be an injective envelope of \mathrm{Coker}\,d_{n-1} for all n>0 . Lemma 10.2 . Let M be an object with a good filtration and let 0\to M\mathrel{\mathop{\kern 0.0pt\rightarrow}\limits^{d_{0}}}I_{0}\mathrel{\mathop{\kern 0.0pt\rightarrow}\limits^{d_{1}}}I_{1}\mathrel{\mathop{\kern 0.0pt\rightarrow}\limits^{d_{2}}}I_{2}\mathrel{\mathop{\kern 0.0pt\rightarrow}\limits^{d_{3}}}\cdots be a minimal injective resolution of M . Then (a) {\rm Ker}\,d_{i} has a good filtration, for all i\geq 1 and (b) \mathrm{ht}\,M=\mathrm{ht}\,I_{0}\leq\mathrm{ht}\,I_{1}\leq\mathrm{ht}\,I_{2}\leq\cdots . (c) Moreover, if {\mathcal{C}} is an HW category, we have \mathrm{ht}\,M=\mathrm{ht}\,I_{0}<\mathrm{ht}\,I_{1}<\mathrm{ht}\,I_{2}<\dots . Proof. There are short exact sequences 0\to\mathrm{Im}d_{i}\to I_{i}\to\mathrm{Im}d_{i+1}\to 0 . Thus the lemma is proved inductively using Lemma 10.1 . ∎ An Ext -vanishing theorem We will now state an easy cohomology vanishing result in generalized HW categories. For clarity, we will also state the stronger version obtained by specializing to HW categories. Let M be an object of a generalized HW category \mathcal{C} , and let k\geq 0 . We inductively define a sequence m(0)\leq m(1)\leq m(2)\leq\cdots in \mathbb{Z}_{\geq 0}\cup\{\infty\} as follows. With the convention that \mathrm{Min}\,\emptyset=\infty , let m(0)=\mathrm{Min}\{n\in\mathbb{Z}_{\geq 0}\mid\mathrm{Ext}^{n}_{\mathcal{C}}(L(0),M)=0\} , and for k>0 , let m(k)=\mathrm{Min}\{n\geq m(k-1)\mid\mathrm{Ext}^{n}_{\mathcal{C}}(L(k),M)=0\} . Theorem 6 . Assume that M has a good filtration. (a) If {\mathcal{C}} is a generalized HW category, then \mathrm{Ext}_{\mathcal{C}}^{n}(L(k),M)=0 , for any n\geq m(k) . (b) If {\mathcal{C}} is an HW category, then m(0)\leq 1 and m(k)\leq 1+m(k-1) , for all k>0 . In particular, all the m(k) are finite when \mathcal{C} is an HW category. Proof. Step 1. Let 0\to M\to I_{0}\to I_{1}\to I_{2}\to\cdots be a minimal injective resolution of M . We claim that \mathrm{Ext}_{\mathcal{C}}^{n}(L(k),M)\simeq{\rm Hom}_{\mathcal{C}}(L(k),I_{n}) for all k,n\geq 0 . Since {\rm Hom}_{\mathcal{C}}(L(k),I_{n})={\rm Hom}_{\mathcal{C}}(L(k),\mathrm{Soc}\,I_{n}) , the induced complex {\rm Hom}_{\mathcal{C}}(L(k),I_{0})\to{\rm Hom}_{\mathcal{C}}(L(k),I_{1})\to{\rm Hom}_{\mathcal{C}}(L(k),I_{2})\to\cdots has zero differentials, which proves the claim. Step 2. Now we prove by induction on k that \mathrm{ht}(I_{n})>k whenever n\geq m(k) . For k=0 , the hypothesis that \mathrm{Ext}_{\mathcal{C}}^{m(0)}(L(k),M)=0 is equivalent to {\rm Hom}_{\mathcal{C}}(L(0),I_{m(0)})=0 . Hence we have \mathrm{ht}(I_{m(0)})>0 . By Lemma 10.2 (b), we deduce that \mathrm{ht}(I_{n})>0 for any n\geq m(0) . Now let k>0 be arbitrary. By the induction hypothesis, we already know that \mathrm{ht}(I_{m(k)})>k-1 . The hypothesis that \mathrm{Ext}_{\mathcal{C}}^{m(k)}(L(k),M)=0 is equivalent to {\rm Hom}_{\mathcal{C}}(L(k),I_{m(k)})=0 , hence \mathrm{ht}(I_{m(k)})>k . By Lemma 10.2 (b), we deduce that \mathrm{ht}(I_{n})>k for any n\geq m(k) . Step 3. It follows that {\rm Hom}_{\mathcal{C}}(L(k),I_{n})=0 for any n\geq m(k) , which proves the Assertion (a). Assertion (b) is an obvious consequence of Lemma 10.2 (c). ∎ We will later use the following corollary of Theorem 6 : Corollary 10.3 . Let \mathcal{C} be a generalized HW category such that \mathrm{Ext}_{\mathcal{C}}^{1}(L(0),L(0))=0 and \mathrm{Ext}_{\mathcal{C}}^{2}(L(1),L(0))=0 . Then \mathrm{Ext}_{\mathcal{C}}^{n}(L(0),L(0))=0 and \mathrm{Ext}_{\mathcal{C}}^{n+1}(L(1),L(0))=0 for all n\geq 1 . PART C: The category of smooth \widehat{\mathfrak{sl}}_{2}(J) -modules An augmented Jordan algebra is a unital Jordan algebra J endowed with an ideal J_{+} of codimension 1 , given once and for all. 11 Proof of Theorems 3 and 4 Nil Jordan algebras A nonunital Jordan algebra I is called nil of index n if x^{n}=0 for all x\in I . Recall the following result, due to Zelmanov [ 36 ] . Zelmanov Nil Theorem 11.1 . Any finitely generated nil Jordan algebra of index n is nilpotent. Let I be a nonunital Jordan algebra. For n\geq 1 , let I^{(n)} be the linear span of all a^{n} with a\in I . The space I^{(n)} is different from the ideal I^{n} , which is the span of all products of n elements in I . The following lemma is well-known Lemma 11.2 . The space I^{(n)} is an ideal. Proof. For any integers p,\,q\geq 0 , let S_{p,q} be the Jordan polynomials in two generators defined by the identity (x+ty)^{n}=\sum_{p+q=n}\,t^{q}S_{p,q}(x,y) . Since the free Jordan algebra on two generators is special, it is enough to check the following two Jordan identities in the free associative algebra on the generators x and y , where a.b denotes the symmetrized product \frac{1}{2}(ab-ba) of elements a,b in the free associative algebra. (a) y.x^{n}=S_{1,n}(y,x)-S_{1,n-1}(x.y,x) . (b) nx^{n+1}=S_{1,n-1}(x^{2},x) , These identities imply that y.I^{(n)}\subset I^{(n)}+I^{(n+1)} for any y\in I and I^{(n+1)}\subset I^{(n)} , which proves the lemma. ∎ Zelmanov Nil Corollary 11.3 . Let I be a finitely generated nonunital Jordan algebra and let n\geq 1 . Then (a) the space I/I^{(n)} is finite dimensional and (b) I^{(n)} contains I^{N} for some N\gg 0 . Proof. By Lemma 11.2 , I/I^{(n)} is a nil Jordan algebra of index n , so by Zelmanov’s Theorem 11.1 , I/I^{(n)} is a nilpotent Jordan algebra. Hence I/I^{(n)} is finite dimensional and I^{(n)} contains I^{N} for some N . ∎ Proof of Theorem 3 Proof. It is obviously enough to prove the statement for the free Jordan algebra J(D) on D generators x_{1},\dots,x_{D} . This Jordan algebra has a natural grading J(D)=\bigoplus_{k\geq 0}J_{k}(D), where J_{0}(D)=\mathbb{K} and J_{1}(D) is the linear span of x_{1},\dots,x_{D} . Set J_{+}:=\bigoplus_{k>0}J_{k}(D) and {\mathcal{A}}:={\mathcal{U}}_{n}(J(D)) . The grading on J induces an obvious grading {\mathcal{A}}=\oplus_{k\geq 0}\,{\mathcal{A}}_{k} , and we write {\mathcal{A}}_{+} for \bigoplus_{k>0}\,{\mathcal{A}}_{k} . Since \rho_{n}(1)=n , we have {\mathcal{A}}_{0}=\mathbb{K} and the nonunital algebra {\mathcal{A}}_{+} is generated by \rho_{n}(J_{+}(D)) . The defining relation \displaystyle{\sum_{\sigma\vdash\ n+1}\mathrm{sgn}(\sigma)|C_{\sigma}|\,h_{\sigma}(a)} implies that \rho_{n+1}(J_{+}^{(n+1)}) lies in {\mathcal{A}}_{+}^{2} . By Zelmanov’s Nil Corollary 11.3 , the space J_{+}/J_{+}^{(n+1)} is finite dimensional. Since the induced map \rho_{n+1}:J_{+}/J_{+}^{(n+1)}\to{\mathcal{A}}_{+}/{\mathcal{A}}_{+}^{2} is surjective, {\mathcal{A}}_{+}/{\mathcal{A}}_{+}^{2} is finite dimensional. Hence {\mathcal{A}} is finitely generated. ∎ A finiteness lemma Let J be an augmented Jordan algebra. Since \widehat{\mathfrak{sl}}_{2}(J)\simeq\mathfrak{sl}_{2}\ltimes\widehat{\mathfrak{sl}}_{2}(J_{+}) , the spaces \bigwedge^{k}\widehat{\mathfrak{sl}}_{2}(J_{+}) and H_{k}(\widehat{\mathfrak{sl}}_{2}(J_{+})) are \mathfrak{sl}_{2} -modules and their possible eigenvalues \ell are even integers in [-n,n] . Let (\bigwedge^{k}\widehat{\mathfrak{sl}}_{2}(J_{+}))_{\ell} and H_{k}(\widehat{\mathfrak{sl}}_{2}(J_{+}))_{\ell} be their respective eigenspaces of eigenvalue \ell . Lemma 11.4 . Assume that J is finitely generated. Then H_{k}(\widehat{\mathfrak{sl}}_{2}(J_{+}))_{2k} is finite dimensional. Proof. For any elements a_{1},\dots,a_{k}\in J_{+} , the chain e(a_{1})\wedge\dots\wedge e(a_{k}) is a cycle. We write [a_{1}\wedge\dots\wedge a_{k}] for the image of e(a_{1})\wedge\dots\wedge e(a_{k}) in H_{k}(\widehat{\mathfrak{sl}}_{2}(J_{+}))_{2k} . Since (\bigwedge^{k}\widehat{\mathfrak{sl}}_{2}(J_{+}))_{2k} is the linear span of the cycles e(a_{1})\wedge\dots\wedge e(a_{k}) , the space H_{k}(\widehat{\mathfrak{sl}}_{2}(J_{+}))_{2k} is the linear span of the symbols [a_{1}\wedge\dots\wedge a_{k}] . We have dh(a)\wedge e(x)\wedge\omega=2e(xa)\wedge\omega+e(x)\wedge\mathrm{ad}(h(a))(\omega) , for all \omega\in\wedge^{k-1}J_{+} . We consider J and \wedge^{k-1}J as J -spaces. The previous formula is equivalent to [\rho(a)(x)\wedge\tau+x\wedge\rho(a)(\tau)]=0 , for any x,a\in J and \tau\in\bigwedge^{k-1}J_{+} . Let \sigma=(\sigma_{1},\dots,\sigma_{m}) be a partition of 2k-1 into m parts. Since \epsilon(\sigma)=-(-1)^{m} , we have \epsilon(\sigma)[x\wedge\rho_{\sigma}(a)(\tau)]=-[\rho_{\sigma}(a)(x)\wedge\tau]. Note that \bigwedge^{k-1}J_{+} is a dominant J -space of level 2(k-1) . By Theorem 1 , we have \displaystyle{\sum_{\sigma\ \vdash\ 2k-1}\hbox{sgn}(\sigma)|C_{\sigma}|\rho_{\sigma}(a)(\tau)=0} . We deduce that \displaystyle{\sum_{\sigma\ \vdash\ 2k-1}[\hbox{sgn}(\sigma)|C_{\sigma}|\rho_{\sigma}(a)(x)\wedge\tau]=0} . For x=a , we deduce that [a^{2k}\wedge\tau]=0 . Therefore the map \bigwedge^{k}J_{+}\to H_{k}(\widehat{\mathfrak{sl}}_{2}(J_{+}))_{2k},\hskip 14.22636pta_{1}\wedge\dots\wedge a_{k}\mapsto[a_{1}\wedge\dots\wedge a_{k}] factors through \bigwedge^{k}(J_{+}/J_{+}^{(2k)}) . By Zelmanov’s Nil Corollary 11.3 , \bigwedge^{k}(J_{+}/J_{+}^{(2k)}) is finite dimensional, so H_{k}(\widehat{\mathfrak{sl}}_{2}(J_{+}))_{2k} is finite dimensional. ∎ Proof of Theorem 4 As for the proof of Theorem 3 , it is enough to prove Theorem 4 for the free Jordan algebras J(D) . As a quotient of some \widehat{\mathfrak{sl}}_{2}(J(D)) , the Lie algebra \widehat{\mathfrak{sl}}_{2}(J) coming from a finitely presented Jordan algebra J is then obviously finitely generated. Its relations are given by the relations in the finite presentation of \widehat{\mathfrak{sl}}_{2}(J(D)) and the (finitely many) additional relations coming from elements x\otimes r , where x is a basis element of \mathfrak{sl}_{2} and r is a relation in the finite presentation of the Jordan algebra J . Given a finitely presented Lie algebra \mathfrak{u} , it is clear that any Lie algebra \mathfrak{g} containing \mathfrak{u} is also finitely presented if \dim\mathfrak{g}/\mathfrak{u}<\infty . Thus the following reformulation of Theorem 4 is slightly stronger. Theorem 4 . The Lie algebra \widehat{\mathfrak{sl}}_{2}(J_{+}) is finitely presented, where J_{+} is the augmentation ideal \bigoplus_{k>0}J_{k}(D) . Proof. Since \widehat{\mathfrak{sl}}_{2}(J_{+}) is a positively graded Lie algebra, the statement follows from the finite dimensionality of H_{1}(\mathfrak{\mathfrak{sl}}_{2}(J_{+})) and H_{2}(\mathfrak{\mathfrak{sl}}_{2}(J_{+})) . Obviously, we have H_{1}(\mathfrak{\mathfrak{sl}}_{2}(J_{+}))\simeq\bigoplus_{1\leq i\leq D}\mathfrak{sl}_{2}\otimes x_{i}, so \dim H_{1}(\mathfrak{\mathfrak{sl}}_{2}(J_{+}))=3D<\infty . We now prove that H_{2}(\widehat{\mathfrak{sl}}_{2}(J_{+})) is finite dimensional. By Proposition 2 of [ 24 ] the \mathfrak{sl}_{2} -module H_{2}(\widehat{\mathfrak{sl}}_{2}(J_{+})) is isotypical of type L(4) and by Lemma 11.4 \dim H_{2}(\widehat{\mathfrak{sl}}_{2}(J_{+}))_{4} is finite dimensional. Hence \dim H_{2}(\widehat{\mathfrak{sl}}_{2}(J_{+}))=5\dim H_{2}(\widehat{\mathfrak{sl}}_{2}(J_{+}))_{4}<\infty . ∎ 12 Finite dimensionality of the Weyl modules \Delta(n) From now on, we assume that the unital Jordan algebra J is finitely generated and augmented , that is, J is endowed with a codimension one ideal J_{+} given once and for all. Smooth \widehat{\mathfrak{sl}}_{2}(J) -modules A \widehat{\mathfrak{sl}}_{2}(J) -module M is called smooth if \dim\,M\leq\aleph_{0} and for any v\in M we have (a) \widehat{\mathfrak{sl}}_{2}(J_{+}^{n}).v=0 for some n=n(v)>0 , and (b) U(\widehat{\mathfrak{sl}}_{2}(J)).v is finite dimensional. It is easy to see that \widehat{\mathfrak{sl}}_{2}(J_{+}^{n})=(\mathfrak{sl}_{2}\otimes J_{+}^{n})\oplus\{J_{+}^{n},J_{+}\}, where \{J_{+}^{n},J_{+}\} is the subspace of \{J_{+},J_{+}\} spanned by the elements \{a,b\} with a\in J_{+}^{n} and b\in J_{+} . Assertion (a) is thus equivalent to the conditions: (a1) (\mathfrak{sl}_{2}\otimes J_{+}^{n}).v=0 , and (a2) \{J_{+}^{n},J_{+}\}.v=0 , for some n=n(v)>0 . Obviously, smooth \widehat{\mathfrak{sl}}_{2}(J) -modules are always (\mathfrak{G},{\rm SL}_{2}) -modules. Finite dimensionality of the Weyl modules \Delta(n) Let \mathbb{K}_{n} be the 1 -dimensional \mathfrak{G}_{0} -module on which h(1) acts by multiplication by n and h(J_{+}) acts trivially. Set \Delta(n)=\Delta(\mathbb{K}_{n}) . Obviously, there is a surjective map \Delta(n)\to L(n) . Theorem 7 . The \widehat{\mathfrak{sl}}_{2}(J) -module \Delta(n) is smooth and finite dimensional. Proof. Let v\in\mathbb{K}_{n} be a generator of \Delta(n) . Step 1. First we prove that f(J_{+}^{(n)}).v=0 . Let a\in J_{+} . Since \Delta_{-n}(n)=\mathbb{K}.f^{n}v , we have f(a)^{n}.v=0 , and therefore \pi(\,e^{n-1}f(a)^{n}).v=0 . By Lemma 3.2 , \pi(\,e^{n-1}f(a)^{n}) is, up to a scalar multiple, the coefficient of t^{n-1} in the formal series \left(\sum_{r=0}^{\infty}f(a^{r+1})t^{r}\right)\exp\left(-\sum_{s=1}^{\infty}\frac{h(a^{s})}{s}t^{s}\right). Since a belongs to J_{+} , we have h(a^{s}).v=0 , for any s>0 . Thus \,e^{n-1}f(a)^{n}.v=rf(a^{n}).v , for some nonzero r . Hence f(J_{+}^{(n)}).v=0 . Step 2. By Zelmanov’s Nil Corollary 11.3 J_{+}^{(n)} contains J_{+}^{N} for some N . We now prove that \widehat{\mathfrak{sl}}_{2}(J_{+}^{N}).\Delta(n)=0 . Since \widehat{\mathfrak{sl}}_{2}(J_{+}^{N}) is an ideal and v generates \Delta(n) , it is enough to prove that \widehat{\mathfrak{sl}}_{2}(J_{+}^{N}).v=0 . Since h(J_{+}).v , e(J_{+}).v and f(J_{+}^{N}).v are all zero, we have (\mathfrak{sl}_{2}\otimes J_{+}^{N}).v=0 . Moreover we have [e(J_{+}),f(J_{+}^{N})]=\{J_{+}^{N},J_{+}\} modulo \mathfrak{sl}_{2}\otimes J_{+}^{N} . Thus, we also have \{J_{+}^{N},J_{+}\}.v=0 , which proves that \widehat{\mathfrak{sl}}_{2}(J_{+}^{N}).v=0 . Step 3. It remains to prove that \Delta(n) is finite dimensional. Since \Delta(n) is a (\mathfrak{sl}_{2}(J/J_{+}^{N}) -module, we can assume that J_{+}^{N}=0 . Therefore J is finite dimensional, and \Delta(n) is finite dimensional by Lemma 3.9 (a). This completes the proof. ∎ Remark: A J -space M of level k is smooth if for every v\in M , we have h(J_{+}^{n})v=0 for n=n(v)\gg 0 . It follows from Theorem 7 that a dominant smooth J -space M is finite dimensional if and only if \Delta(M) is finite dimensional. 13 Lie algebra cohomology and smooth representations As in the previous Section, let J be a finitely generated augmented Jordan algebra. In this section, we compare \hbox{Ext}^{*} in the category of smooth modules with relative Lie algebra cohomology. From now on, we consider only representations of finite or countable dimension for Lie algebras and algebraic groups. Generalities about algebraic groups Let \mathfrak{g} be a finite dimensional perfect Lie algebra. Then \mathfrak{g}\simeq\mathfrak{s}\ltimes\mathfrak{u} where \mathfrak{s} is semi-simple and \mathfrak{u} is the nilpotent radical. Let G be the unique connected and simply connected algebraic group with Lie algebra \mathfrak{g} . There is a Levi decomposition G=S\ltimes U , where S is semi-simple and U is unipotent, with Lie algebras \mathfrak{s} and \mathfrak{u} , respectively. In what follows, a rational G -module will simply be called a G -module . Let \mathrm{Mod}(G) be the category of G -modules. It is well known that \mathfrak{u} acts nilpotently on any finite dimensional \mathfrak{g} -module. It follows that any locally finite dimensional \mathfrak{g} -module is a G -module, and conversely. The group G acts on U as follows: U acts by left multiplication and S by conjugation. With respect to this action, the ring \mathbb{K}[U] of rational functions on U is a G -module. Since \mathbb{K}[U] is an injective U -module and since S acts reductively, the G -module \mathbb{K}[U] is injective. In fact, \mathbb{K}[U] is the injective envelope of the trivial module. Hence M\otimes\mathbb{K}[U] is an injective envelope of M , for any simple G -module M . Generalities about group cohomology Let \mathfrak{g} , \mathfrak{s} , \mathfrak{u} , G , S , and U be defined as in the previous subsection. Let M\in\mathrm{Mod}(G) . By definition, the relative cohomology group H^{*}(\mathfrak{g},\mathfrak{s},M) is the cohomology of the Chevalley-Eilenberg cocomplex {\rm Hom}_{\mathfrak{s}}(\bigwedge^{p}\mathfrak{g}/\mathfrak{s},M) . See [ 26 ] or [ 27 ] for details. Since \mathfrak{s} acts reductively, we have H^{*}(\mathfrak{g},\mathfrak{s},M)\simeq H^{*}(\mathfrak{u},M)^{\mathfrak{s}} . As an algebraic variety, U is isomorphic to its Lie algebra \mathfrak{u} , thus the de Rham complex \bigwedge\mathfrak{u}^{*}\otimes\mathbb{K}[U] is exact. For any U -module M , it follows easily that H^{*}(\mathfrak{u},M)\simeq H^{*}(U,M) , a result often attributed to Malcev. As \mathfrak{s} is reductive, Hochschild proved that the relative cohomology H^{*}(\mathfrak{g},\mathfrak{s},-) describes extensions in the relative category of (\mathfrak{g},\mathfrak{s}) -modules [ 18 ] . 1 1 1 A more recent account appears in [ 31 ] . Combining this result with the Malcev isomorphism, we deduce the following well known result: Lemma 13.1 . Let M and N be G -modules, with M finite dimensional. Then H^{*}(\mathfrak{g},\mathfrak{s},{\rm Hom}_{\mathbb{K}}(M,N))=\mathrm{Ext}^{*}_{\mathrm{Mod}(G)}(M,N) . The pro-algebraic group {\mathbf{SL}}_{2}(J) For any integer n\geq 1 , the Lie algebra \widehat{\mathfrak{sl}}_{2}(J/J_{+}^{n})=\mathfrak{sl}_{2}\ltimes\widehat{\mathfrak{sl}}_{2}(J_{+}/J_{+}^{n}) is perfect and finite dimensional. Let U_{n} be the unipotent group with Lie algebra \widehat{\mathfrak{sl}}_{2}(J_{+}/J_{+}^{n}) and set G_{n}={\rm SL}_{2}\ltimes U_{n} . By definition, {\mathbf{SL}}_{2}(J) is the proalgebraic group \varprojlim G_{n} , and an {\mathbf{SL}}_{2}(J) -module is a direct union \bigcup_{n\geq 1}\,M_{n} of G_{n} -modules M_{n} , where, as always, we assume that \dim M\leq\aleph_{0} . Since the Lie algebra \widehat{\mathfrak{sl}}_{2}(J/J_{+}^{n}) is perfect and finite dimensional, any smooth \widehat{\mathfrak{sl}}_{2}(J) -module is an {\mathbf{SL}}_{2}(J) -module and conversely. The category \mathrm{Mod}({\mathbf{SL}}_{2}(J))={\mathcal{C}}_{sm}(J) Let \mathrm{Mod}({\mathbf{SL}}_{2}(J)) be the category of {\mathbf{SL}}_{2}(J) -modules. By the material in Section 13.1 , \mathrm{Mod}({\mathbf{SL}}_{2}(J)) coincides with the category {\mathcal{C}}_{sm}(J) of smooth \widehat{\mathfrak{sl}}_{2}(J) -modules. Depending on the context, we will use one terminology or the other. Let n\geq 0 . In what follows, the simple {\mathfrak{sl}}_{2} -module L(n) of dimension n+1 will be viewed as a \widehat{\mathfrak{sl}}_{2}(J) -module with trivial action of \widehat{\mathfrak{sl}}_{2}(J_{+}) . It is clear that the family (L(n))_{n\in\mathbb{Z}_{\geq 0}} is a complete list of simple objects in \mathrm{Mod}({\mathbf{SL}}_{2}(J)) . Set U=\varprojlim U_{m} . By definition, its ring of rational functions is \mathbb{K}[U]=\varinjlim\mathbb{K}[U_{m}] . Viewing \mathbb{K}[U] as a {\mathbf{SL}}_{2}(J) -module, let I(n)=L(n)\otimes\mathbb{K}[U] . Lemma 13.2 . Let n\geq 0 . Then (a) I(n) is the injective envelope of L(n) and (b) \Delta(n)^{*} is the standard object \nabla(n) with socle L(n) . Proof. Clearly L(n)\otimes\mathbb{K}[U_{m}] is the injective envelope of L(n) in the category \mathrm{Mod}({\mathbf{SL}}_{2}(J/J_{+}^{m}) , which proves Assertion (a), and shows the existence of the standard object \nabla(n) . By construction, the cosocle of \Delta(n) is L(n) , so Soc(\Delta(n)^{*})=L(n) and \Delta(n)^{*}\subseteq I(n) , from which it follows that \Delta(n)^{*} is contained in the standard object \nabla(n) with socle L(n) . Conversely, the \Delta(n) surjects onto the dual \nabla(n)^{*} of the standard module, since the highest eigenspace of \nabla(n) is the J -space \mathbb{K}_{n} . Dualizing, we see that \nabla(n) is contained in \Delta(n)^{*} . Hence \Delta(n)^{*}=\nabla(n) . ∎ The definition of LA^{+} -categories is provided in the PART B, Section 6 . Since the standard module \nabla(n)=\Delta(n)^{*} is finite dimensional by Theorem 7 , we have: Corollary 13.3 . The category \mathrm{Mod}({\mathbf{SL}}_{2}(J)) is a LA+ category. Smooth cohomology We will now see how to compute \mathrm{Ext} -groups in the category \mathrm{Mod}({\mathbf{SL}}_{2}(J))=\mathcal{C}_{sm}(J) of smooth \widehat{\mathfrak{sl}}_{2}(J) -modules. Let M be a smooth \widehat{\mathfrak{sl}}_{2}(J) -module. A relative k -cochain \omega\in{\rm Hom}_{\mathfrak{sl}_{2}}(\bigwedge^{k}(\widehat{\mathfrak{sl}}_{2}(J)/\mathfrak{sl}_{2}),M) is called smooth if k=0 , or \omega(x_{1}\wedge x_{2}\wedge\cdots\wedge x_{k})=0 , whenever x_{1}\in\widehat{\mathfrak{sl}}_{2}(J_{+}^{N}) , for N\gg 0 . Let C^{k}_{sm}(M) be the space of smooth relative k -cochains with values in M . By definition, the smooth cohomology H^{*}_{sm}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},M) is the homology of the complex C_{sm}^{*}(M) . It follows that H^{*}_{sm}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},M)=\varinjlim H^{*}(\widehat{\mathfrak{sl}}_{2}(J/J_{+}^{n}),\mathfrak{sl}_{2},M) . Thus the folklore Lemma 13.1 implies the following result: Lemma 13.4 . Let M,\,N be smooth \widehat{\mathfrak{sl}}_{2}(J) -modules with M finite dimensional. Then \mathrm{Ext}^{*}_{\mathrm{Mod}({\mathbf{SL}}_{2}(J))}(M,N)=H^{*}_{sm}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},{\rm Hom}_{\mathbb{K}}(M,N)) . Comparison with relative homology. Let M be a finite dimensional {\mathbf{SL}}_{2}(J) -module. In general, it is difficult to compare the smooth cohomology H^{*}_{sm}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},M) with the Chevalley-Eilenberg relative homology H_{*}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},M^{*}) . However, it is possible to do so when M and J have compatible gradings. A positively graded Jordan algebra is a Jordan algebra endowed with a grading J=\bigoplus_{n\geq 0}\,J_{n} such that J_{0}=\mathbb{K} . It is viewed as an augmented Jordan algebra, with augmentation ideal J_{+}:=\bigoplus_{n>0}J_{n} . The Lie algebra \widehat{\mathfrak{sl}}_{2}(J) inherits the grading of J . Given a \mathbb{Z} -graded vector space E=\bigoplus_{n\in\mathbb{Z}}E_{n} , the graded dual of E is the space \bigoplus_{n\in\mathbb{Z}}E_{n}^{*} . Lemma 13.5 . Assume that J is a finitely generated positively graded Jordan algebra. Let M be a finite dimensional \mathbb{Z} -graded \widehat{\mathfrak{sl}}_{2}(J) -module. Then (a) M is smooth. (b) H^{k}_{sm}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},M) is the graded dual of H_{k}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},M^{*}) , for any integer k\geq 0 . Proof. Assertion (a) is obvious and Assertion (b) follows from the fact that C^{k}_{sm}(M^{*}) is the graded dual of the space H_{0}(\mathfrak{sl}_{2},\bigwedge^{k}(\widehat{\mathfrak{sl}}_{2}(J)/\mathfrak{sl}_{2})\otimes M) of relative chains. ∎ Assume again that J is a finitely generated positively graded Jordan algebra. The simple modules L(n) can be viewed as \mathbb{Z} -graded modules, with grading concentrated in degree 0 . The following statement is an immediate consequence of Lemmas 13.4 and 13.5 . Corollary 13.6 . For any k,n\geq 0 , the group \mathrm{Ext}^{k}_{\mathrm{Mod}({\mathbf{SL}}_{2}(J))}(L(n),L(0)) is the graded dual of H_{k}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},L(n)) . Some Ext-vanishing results. Theorem 13.7 . Let J be the free unital Jordan algebra J(D) on D>0 generators. Then (a) \mathrm{Ext}^{k}_{\mathrm{Mod}({\mathbf{SL}}_{2}(J(D))}(L(0),L(0))=0 for k=1,\,2,\,3 . (b) \mathrm{Ext}^{2}_{\mathrm{Mod}({\mathbf{SL}}_{2}(J(D))}(L(2),L(0))=0 . Proof. By Corollary 13.6 , \mathrm{Ext}^{k}_{\mathrm{Mod}({\mathbf{SL}}_{2}(J(D))}(L(n),L(0)) is the graded dual of H_{k}(\widehat{\mathfrak{sl}}_{2}(J(D)),\mathfrak{sl}_{2},L(n)) . Thus Theorem 13.7 is equivalent to the following known results about relative homology: (a) H_{1}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},\mathbb{K})=0 , which is obvious, (b) H_{2}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},\mathbb{K})=0 , due to [ 3 ] , (c) H_{3}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},\mathbb{K})=0 , proved in [ 24 , Prop 3] , (d) H_{2}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},L(2))=0 , proved in [ 24 , Lemma 13] . ∎ 14 Is \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D))) a generalized HW category? Let J be a finitely generated augmented Jordan algebra. Set {\mathbf{PSL}}_{2}(J)={\mathbf{SL}}_{2}(J)/\pm 1 . Then \mathrm{Mod}({\mathbf{PSL}}_{2}(J)) is the category of {\mathbf{SL}}_{2}(J) -modules with even eigenvalues. The simple modules in \mathrm{Mod}({\mathbf{PSL}}_{2}(J)) are ordered by their dimensions and the (n+1)^{st} simple module is L(2n) . By Corollary 13.3 , \mathrm{Mod}({\mathbf{PSL}}_{2}(J)) is an LA+ category. The free Jordan algebra J=J(D) is viewed as an augmented Jordan algebra with augmentation ideal J_{+}=\bigoplus_{n>0}J(D)_{n} . In this section, we consider the category \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D))) , relating it to the framework of Ext -vanishing and generalized highest weight categories introduced in Part B. Motivation Let J be a finitely generated augmented Jordan algebra, and let p be a prime number. The category \mathrm{Mod}({\mathbf{PSL}}_{2}({\overline{\mathbb{F}}}_{p})) of rational {\mathbf{PSL}}_{2}({\overline{\mathbb{F}}}_{p}) -modules and the category \mathrm{Mod}({\mathbf{PSL}}_{2}(J)) look very similar. In fact, Donkin [ 11 ] and Koppinen [ 25 ] showed that \mathrm{Mod}({\mathbf{PSL}}_{2}({\overline{\mathbb{F}}}_{p})) satisfies the axioms of a HW category (though this notion was introduced only some time later in [ 9 ] ). The category \mathrm{Mod}({\mathbf{PSL}}_{2}(J)) is LA+ . However, even in the most favorable case when J is the free Jordan algebra J(D) , the following proposition shows that \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D))) is not a HW category. Proposition 14.1 . For J=J(D) , \mathrm{Ext}^{1}_{\mathrm{Mod}({\mathbf{PSL}}_{2}(J))}(\nabla(2),\nabla(2))\simeq(J_{+}/J_{+}^{3})^{*}. Proof. The proof follows from [ 24 , Lemma 21] . The details are left to the reader. ∎ Nonetheless, for the free Jordan algebra J(D) the Lie algebra \widehat{\mathfrak{sl}}_{2}(J(D)) is free in a certain category [ 24 ] , which suggests it should enjoy some cohomology vanishing results. This raises the natural question: Is the category \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D))) a generalized HW category? Some evidence Several computations make it seem plausible that \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D))) might be a generalized HW category. As these calculations are very technical, we mention only one of them. If \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D))) is a generalized HW category, then {\mathcal{F}}_{n}(I(n)) is an extended \nabla(n) -object. This conclusion is obvious for n=0 or 1 , even without assuming that J is free. The first interesting case is when n=2 . For most augmented Jordan algebras, {\mathcal{F}}_{2}(I(2)) is not an extended \nabla(2) -object. However, this does hold for the free Jordan algebra J(D) . Lemma 14.2 . If J=J(D) , the \widehat{\mathfrak{sl}}_{2}(J) -module {\mathcal{F}}_{2}(I(2)) is a \nabla(2) -object. Proof. The standard module \nabla(2) is the indecomposable module with short exact sequence 0\to L(2)\to\nabla(2)\to\mathbb{K}^{D}\to 0. The module {\mathcal{F}}_{2}(I(2)) is the graded dual of the module M\simeq\Delta_{2}({\mathcal{U}}_{2}(J)) explicitely described in [ 24 , Lemma 21] . It follows easily that {\mathcal{F}}_{2}(I(2)) is a \nabla(2) -object. ∎ Consequences Theorem 8 . If \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D))) is a generalized HW category, then the conjectural formula of [ 24 ] for \dim J_{n}(D) would hold. Proof. Suppose that \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D))) is a generalized HW category. By Theorem 13.7 , we have \mathrm{Ext}^{1}_{\mathrm{Mod}({\mathbf{PSL}}_{2}(J(D)))}(L(0),L(0))=0 , and \mathrm{Ext}^{2}_{\mathrm{Mod}({\mathbf{PSL}}_{2}(J(D)))}(L(2),L(0))=0 . Note that, relative to the given order, L(0) and L(2) are the first two simple objects in \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D)) . They correspond with the objects denoted L(0) and L(1) in PART B. Thus by Corollary 10.3 , we deduce that \mathrm{Ext}^{n}_{\mathrm{Mod}({\mathbf{PSL}}_{2}(J(D)))}(L(0),L(0))=0 , and \mathrm{Ext}^{n+1}_{\mathrm{Mod}({\mathbf{PSL}}_{2}(J(D)))}(L(2),L(2))=0 , for all n>0 . By Corollary 13.6 , this is equivalent to H_{n}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},\mathbb{K})=0 and H_{n+1}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},L(2))=0 , for all n>0 . Note that H_{n}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},\mathbb{K})\simeq H_{n}(\widehat{\mathfrak{sl}}_{2}(J_{+}))^{\mathfrak{sl}_{2}} and H_{n+1}(\widehat{\mathfrak{sl}}_{2}(J),\mathfrak{sl}_{2},L(2))\otimes L(2) is the adjoint part of H_{n}(\widehat{\mathfrak{sl}}_{2}(J_{+})) . Thus by [ 24 , Corollary 1] , this implies that the conjectural formula for \dim J_{n}(D) holds. The reader should note that the notation used here conflicts with the notation in [ 24 ] . The Jordan algebras J(D) and J_{+}(D) are denoted J^{u}(D) and J(D) in [ 24 ] . Moreover, the Tits-Allison-Gao functor \widehat{\mathfrak{sl}}_{2} is denoted \mathfrak{sl}_{2} in [ 24 ] . ∎ However, as noted in the introduction and in Section 5 , very recent results posted to arXiv [ 13 ] indicate that the conjectural formula of [ 24 ] does not hold in general. As a direct consequence of Theorem 8 , we see that \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D))) and the larger category \mathcal{C}_{sm}(J(D)) of all smooth \widehat{\mathfrak{sl}}_{2}(J(D)) -modules are not generalized highest weight categories: Corollary 14.3 . For D\geq 2 , the categories \mathrm{Mod}({\mathbf{PSL}}_{2}(J(D))) and \mathcal{C}_{sm}(J(D)) are not generalized highest weight categories.∎