On the Killing form of Lie Algebras in Symmetric Ribbon Categories

As a step towards the structure theory of Lie algebras in symmetric monoidal categories we establish results involving the Killing form. The proper categorical setting for discussing these issues are symmetric ribbon categories.


Introduction
At times, a mathematical notion reveals its full nature only after viewing it in a more general context than the one in which it had originally appeared, and often this is achieved by formulating it in the appropriate categorical framework. To mention a few examples, the octonions are really an instance of an associative commutative algebra, namely once they are regarded as an object in the category of Z ×3 2 -graded vector spaces with suitably twisted associator [1,2]; chiral algebras in the sense of Beilinson and Drinfeld can be viewed as Lie algebras in some category of Dmodules [48]; and vertex algebras (which are Beilinson-Drinfeld chiral algebras on the formal disk) are singular commutative associative algebras in a certain functor category [5]. As another illustration, not only do Hopf algebras furnish a vast generalization of group algebras, but indeed a group is a Hopf algebra, namely a Hopf algebra in the category of sets.
In this note we study aspects of Lie algebras from a categorical viewpoint. This already has a long tradition, see e.g. [3,9,15,22,23]. For instance, it is well known that Lie superalgebras and, more generally, Lie color algebras, are ordinary Lie algebras in suitable categories, see Example 2.3 below. Our focus here is on the structure theory of Lie algebras in symmetric monoidal categories which, to the best of our knowledge, has not been investigated in purely categorical terms before. We restrict our attention to a few particular issues, the main goal being to identify for each of them an adequate categorical setting that both allows one to develop the respective aspect closely parallel with the classical situation and at the same time covers an as broad as possible class of cases. In particular we do not assume the underlying category to be semisimple, or Abelian, or linear (i.e., enriched over vector spaces), or that the tensor product functor is exact, albeit large classes of examples that are prominent in the literature share (some of) these properties.
On the other hand we insist on imposing relevant conditions directly on the underlying category. They are thereby not necessarily the weakest possible, and in fact they could often be considerably weakened to conditions on subclasses of objects. For instance, even an object in a non-braided category can be endowed with a Lie algebra structure, provided that it comes with This paper is a contribution to the Special Issue on New Directions in Lie Theory. The full collection is available at http://www.emis.de/journals/SIGMA/LieTheory2014.html arXiv:1502.07441v1 [math.RA] 26 Feb 2015 a self-invertible Yang-Baxter operator in the sense of [21,Definition 2.6] that takes over the role of the self-braiding. But this way one potentially loses the clear distinction between categorical and non-categorical aspects. Indeed, a benefit of the abstraction inherent in the categorical point of view is that it allows one to neatly separate features that apply only to a subclass of examples from those which are essential for the concepts and results in question and are thereby generic. We are specifically interested in the proper notion of Lie subalgebra and ideal, and in the relevance of the Killing form. Recall that non-degeneracy of the Killing form furnishes a criterion for semisimplicity of a Lie algebra over the complex numbers, but ceases to do so for e.g. Lie algebras over a field of non-zero characteristic or for Lie superalgebras.
The rest of this note is organized as follows. Section 2 is devoted to basic aspects of Lie algebras in monoidal categories. In Section 2.1 we present the definition of a Lie algebra in a symmetric additive category and illustrate it by examples. We then introduce the notions of nilpotent and solvable Lie algebras in Section 2.2. In Section 2.3 we briefly describe subalgebras and ideals of Lie algebras, paying attention to the differences that result from the specification of the relevant class of monics. In Section 3 we study aspects of the Killing form. To this end we first review in Section 3.1 some properties of the partial trace. In Section 3.2 we give the definition of the Killing form of a Lie algebra in a symmetric ribbon category and show that it is symmetric, while invariance is established in Section 3.3. Finally in Section 3.4 we explore the role of non-degeneracy of the Killing form.

Lie algebras
We freely use pertinent concepts from category theory, and in particular for algebras in monoidal categories. Some of these are recalled in the Appendix. The product of a Lie algebra is antisymmetric and satisfies the Jacobi identity. This requires that morphisms can be added and that there is a notion of exchanging the factors in a tensor product. Moreover, it turns out that one can proceed in full analogy to the classical case only if the latter is a symmetric braiding. Thus we make the Convention 2.1. In the sequel, unless stated otherwise, by a category we mean an additive symmetric monoidal category, with symmetric braiding c.
To our knowledge, Lie algebras in this setting were first considered in [22,23]. To present the definition, we introduce the short-hand notation for multiple self-braidings, as well as h 2 := h and for iterations of a morphism h ∈ Hom C (U ⊗ U, U ).

Lie algebras in additive symmetric monoidal categories
With the notations (2.1) and (2.2) we have Definition 2.2. A Lie algebra in a category C is a pair (L, ) consisting of an object L ∈ C and a morphism ∈ Hom C (L ⊗ L, L) that satisfies On the Killing form of Lie Algebras in Symmetric Ribbon Categories 3 (antisymmetry) and (Jacobi identity).
Here and below we assume monoidal categories to be strict. The additional occurrences of the associativity constraint in the non-strict case are easily restored (the explicit expression can e.g. be found in [21, Definition 2.1]). By abuse of terminology, also the object L is called a Lie algebra; the morphism is referred to as the Lie bracket of L. (i) For k a field and C = Vect k the category of k-vector spaces, with the symmetric braiding given by the flip v ⊗ v →v ⊗ v, we recover ordinary Lie algebras.
(ii) A Lie superalgebra over k [17,30] is a Lie algebra in the category SVect k of Z 2 -graded k-vector spaces with the braiding given by the superflip, acting as v⊗v on homogeneous elements (and extended by bilinearity).
(iii) A Lie color algebra (or color Lie algebra) [4,43,45] is a Lie algebra in the category Γ-Vect of Γ-graded vector spaces, where Γ is a finite Abelian group endowed with a skew bicharacter ϕ. The braiding acts on homogeneous elements analogously as for superalgebras, i.e. as a flip multiplied by a phase factor ϕ(deg(v), deg(v )).
For Γ = Z 2 with the unique cohomologically nontrivial skew bicharacter this yields Lie superalgebras, while for ϕ cohomologically trivial it yields ordinary Lie algebras for any Γ.
(iv) A so-called Hom-Lie algebra L in a symmetric monoidal category C, for which the Jacobi identity is deformed with the help of an automorphism of L (first introduced in [26] for the case C = Vect k ) is a Lie algebra in a category HC whose objects are pairs consisting of an object U of C and an automorphism of U [6].

Remark 2.4.
(i) When trying to directly generalize the notion of a Lie algebra to monoidal categories with a generic braiding, two independent Jacobi identities must be considered [53].
(ii) A more conceptual approach to the case of generic braiding has led to the notion of a quantum Lie bracket on an object L, satisfying a generalized Jacobi identity, which is related to the adjoint action for Hopf algebras in C that are compatible with an additional coalgebra structure on L [38]. In another approach one deals with a whole family of nary products on an object of Γ-Vect, with Γ a finite Abelian group endowed with a (not necessarily skew) bicharacter ψ [41]. (If ψ is skew, this reduces to Lie color algebras.) This approach has the advantage of keeping important aspects of the classical case, e.g. the primitive elements of a Hopf algebra in Γ-Vect and the derivations of an associative algebra in Γ-Vect carry such a generalized Lie algebra structure. There is also a further generalization [42] to the case that C is the category of Yetter-Drinfeld modules over a Hopf algebra with bijective antipode. Remark 2.6. If the category C is in addition Abelian, then one can define the universal enveloping algebra of a Lie algebra L ∈ C as the quotient of the tensor algebra T (L), as described in Example A.3(i), by the two-sided ideal I(L) generated by the image of the morphism − id L⊗L +c L,L (regarded as an endomorphism of T (L)). In many cases of interest, such as for Lie color algebras [23,33] or if C is the category of comodules over a coquasitriangular bialgebra in Vect [15], the ideal I(L) inherits a Hopf algebra structure from T (L), so that the quotient is naturally a Hopf algebra as well. This still applies [42,Section 6] in the case of the Lie algebras in categories of Yetter-Drinfeld modules mentioned in Remark 2.4(ii).
Remark 2.7. The class of all Lie algebras in C forms the objects of a category, with a morphism f from a Lie algebra (L, ) to a Lie algebra (L , ) being a morphism f ∈ Hom C (L, L ) satisfying The category of associative algebras in C is defined analogously; supplementing the mapping from an associative algebra to its commutator Lie algebra by the identity on the sets of algebra morphisms defines a functor from the category of associative algebras in C to the one of Lie algebras in C. Note that, as the condition (2.6) is nonlinear, these categories are not additive.

Nilpotent and solvable Lie algebras
Basic aspects of the structure theory of Lie algebras can be developed in the same way as in the classical case. We start with Definition 2.8. An Abelian Lie algebra is a Lie algebra with vanishing Lie product, = 0.
(i) Trivially, (U, = 0) is an Abelian Lie algebra for any object U ∈ C. A generic object cannot be endowed with any other Lie algebra structure.
(ii) The commutator Lie algebra (A, m ) built from a commutative associative algebra (A, m) is Abelian.
Next we introduce the concepts of nilpotent and solvable Lie algebras. Recall that according to the definition (2.2), for a Lie algebra (L, ) the (n − 1)-fold iteration of the Lie bracket is denoted by n ∈ Hom C (L ⊗n , L). We now introduce in addition morphisms n ∈ Hom C (L ⊗2 (n−1) , L) by (i) L is called nilpotent iff n = 0 for sufficiently large n.
(ii) L is called solvable iff n = 0 for sufficiently large n.
(iii) L is called derived nilpotent iff n • ⊗n = 0 for sufficiently large n.
It follows directly from the definitions that a nilpotent Lie algebra is solvable. Indeed, n = n • n−1 ⊗ n−2 ⊗ · · · ⊗ ⊗ id L ⊗ id L .

But we even have
On the Killing form of Lie Algebras in Symmetric Ribbon Categories Proposition 2.11. A derived nilpotent Lie algebra is solvable.
Proof . We have directly 3 = 2 • ⊗2 , while for n>3 one shows by induction that Thus n • ⊗n = 0 implies n+1 = 0, so that the claim follows directly from the definitions of derived nilpotency and of solvability.

Subalgebras and ideals
A subalgebra of an algebra A is an isomorphism class of monics to A that are algebra morphisms. For spelling out this notion in detail, one must specify the class of monics one is considering. We present the cases that either all monics are admitted, corresponding to working with subobjects, or only split monics are admitted, corresponding to retracts. In the latter case we give Definition 2.12.
(i) A retract subalgebra, or subalgebra retract, of an algebra where Note that part (ii) of Definition 2.12 is redundant -it is merely a special case of part (i), as we do not specify any properties of the product m in (i). For clarity we present the Lie algebra version nevertheless separately; it is worth recalling that in the Lie algebra case the Convention 2.1 is in effect. Also note that owing to p • e = e, (2.7) is equivalent to A , and analogously for (2.8), as well as for (2.9) below.
We call a Lie algebra L indecomposable iff it is non-Abelian and its only retract ideals are the zero object and L itself. Put differently, we have Definition 2.14. An indecomposable Lie algebra is a Lie algebra which is not Abelian and which does not possess any non-trivial retract ideal.
By antisymmetry of , the equality (2.9) is equivalent to . For an associative algebra, the equality analogous to (2.9) instead defines a left ideal, while exchanging the roles of the two morphisms id and e gives the notion of right ideal.
If (K, e, r) is a non-trivial retract of a Lie algebra (L, ), then together with p = r • e also p := id L −p is a non-zero idempotent, and hence if C is idempotent complete, then there is a non-trivial retract (K , e , r ) of L such that p = r • e and L ∼ = K ⊕ K as an object in C. In the present context a particularly interesting case is that both K and K are retract ideals of L.
To account for this situation we give Definition 2.15. Given a finite family of Lie algebras (K i , i ), the direct sum Lie algebra of the family is the Lie algebra (L, ) whose underlying object L is the direct sum L := i K i of the objects underlying the Lie algebras (K i , i ), and whose Lie bracket is given by (2.10) the Lie algebras (K i , i ) are retract ideals of the direct sum Lie algebra L. Conversely, one has Lemma 2.16. If a Lie algebra (L, ) is, as an object, the direct sum L = i K i , and each of the retracts K i is a retract ideal of L, then (L, ) is the direct sum of the K i also as a Lie algebra.
Proof . Because of id L = i p L K i the Lie bracket can be written as the triple sum (2.11) Now since K i and K j are ideals, one has p . As a consequence the expression (2.11) reduces to Hence there are brackets i satisfying (2.10). Moreover, invoking the uniqueness of direct sum decomposition in the additive category C, it follows (notwithstanding the non-additivity of the category of Lie algebras in C) that these brackets coincide with the assumed Lie brackets on the retract ideals K i up to isomorphism of Lie algebras.
Next, consider a retract ideal (K, e, r) of a Lie algebra L in an idempotent-complete category. We call K a primitive retract ideal iff the idempotent p = e • r cannot be written as a sum p = p 1 + p 2 of two non-zero idempotents such that at least one of the corresponding retracts (K 1 , e 1 , r 1 ) and (K 2 , e 2 , r 2 ) is a retract ideal of L. Now if the retracts associated with two idempotents in End C (L), say p and p , are retract ideals, then so is the retract associated with the idempotent p • p ∈ End C (L). If p is in addition primitive, then either p • p = p or p • p = 0. A primitive retract ideal is hence an indecomposable Lie algebra.
It is worth pointing out that in the Definitions 2.12 and 2.13 there is no need to assume the existence of an image of the Lie bracket . In contrast, if we spell out the corresponding definitions for the case of subobjects, we do have to make such an assumption: Definition 2.17. Let (L, ) ∈ C be a Lie algebra for which the image Im( ) exists in C.
Accordingly, Definition 2.14 then gets replaced by Definition 2.18. A simple Lie algebra is a Lie algebra which is not Abelian and which does not possess any non-trivial subobject ideal.
(i) Given a unital associative algebra (A, m, η) in C, the commutator m (2 is a subobject of A -meaning that the unit morphism η is monic, which is true under rather weak conditions [37], e.g. if C is k-linear with k a field [11, Section 7.7] -(1, η) is an Abelian subobject ideal of the commutator Lie algebra (A, m ).
(ii) If there exists in addition a morphism ∈ Hom C (A, 1) such that ξ := • η ∈ End C (1) is invertible, then (1, η,˜ ) with˜ := ξ −1 • η is a retract of A, and thus even an Abelian retract ideal of the Lie algebra (A, m ), and we have a direct sum decomposition of Lie algebras.
(i) If A carries the structure of a Frobenius algebra in C, then its counit ε is a natural candidate for the morphism that is assumed in Example 2.19(ii). If, moreover, the Frobenius algebra A is strongly separable in the sense of Definition A.1(ii) as well as symmetric, then ε • η = dim C (A) equals the dimension of A, i.e. (1, η,ε) being a retract ideal of the commutator Lie algebra (A, m ) is equivalent to A having invertible dimension.
(ii) As a special case, consider the tensor product A U := U ⊗ U ∨ of an object U with its right dual in a (strictly) sovereign category C. The object A U carries a canonical structure of a symmetric Frobenius algebra (A U , m U , η U , ∆ U , ε U ), for which all structural morphisms are simple combinations of the left and right evaluation and coevaluation morphisms for U , namely and similarly for the coproduct ∆ U and counit ε U . We refer to A U with this algebra and coalgebra structure as a (full) matrix algebra. Any matrix algebra is Morita equivalent to the trivial Frobenius algebra 1, and we have ε U • η U = dim C (U ), so 1 is a retract ideal of the commutator Lie algebra (A U , m U ) iff U has invertible dimension.
(iii) A simple realization of the situation described in (ii) is obtained by taking C to be the category of finite-dimensional complex representations of a classical finite-dimensional simple Lie algebra L in Vect C (or in Vect k , with k any algebraically closed field of characteristic zero) and U = D ∈ C to be the defining representation of L. The retract A D in the resulting direct sum decomposition A D = 1 ⊕ A D is then a simple Lie algebra iff L = sl n (C), in which case A D = Ad is (isomorphic to) the adjoint representation of sl n (C) and A D = 1 ⊕ Ad is just the familiar decomposition gl n (C) ∼ = C ⊕ sl n (C).
Remark 2.22. In applications, the relevance of a Lie algebra arises often through its linear representations. In the present note we do not dwell on representation theoretic issues, except for the following simple observation. For A U = U ⊗ U ∨ the associative algebra with product m U as in (2.13), the object U is naturally an A U -module, with representation morphism ρ := id U ⊗d U ∈ Hom C (A U ⊗ U, U ). The representation property is verified as follows: Being a module over the associative algebra (A U , m U ), U is also a module over the commutator Lie algebra (A U , m U ) and thus, in case U has invertible dimension, also over the Lie algebra A U appearing in the direct sum decomposition (2.12). It then follows in particular that if every object of C is a retract of a suitable tensor power of U , then each object of C carries a natural structure of an A U -module, whereby C becomes equivalent to a subcategory of the category A U -mod C of A U -modules in C. In the special case of U = D being the defining representation of sl n (C), C is in fact equivalent to A U -mod C as a monoidal category.
3 The Killing form

Partial traces
The Killing form of a Lie algebra in Vect k is defined as a trace, actually as a partial trace. Canonical traces, and thus partial traces, of morphisms exist in a category C iff C is a full monoidal subcategory of a ribbon category [29]. Accordingly from now on we restrict our attention to ribbon categories (but for now do not yet impose our Convention 2.1).
One may think of the basic data of a (strictly monoidal) ribbon category to be the right and left dualities, b, d andb,d, the braiding c, and the sovereign structure σ, rather than the twist θ. Then instead of expressing the sovereign structure with the help of the twist as in (A.2), conversely the twist is expressed as (3.1) It will prove to be convenient to exploit the graphical calculus for (strict) monoidal categories and for algebras therein that has been developed repeatedly in different guises, see e.g. [7,28,32,38,47,51] or Appendix A of [16]. For instance, in this graphical description the defining property (A.1) of the sovereign structure takes the form while the expression (3.1) for the twist becomes The (left and right) trace of an endomorphism f ∈ End C (V ) in a ribbon category is given by For defining the Killing form we also need a generalization, the notion of a partial trace: for a morphism f ∈ Hom C (U 1 ⊗ · · · ⊗U n ⊗ V, V ) we introduce the partial trace with respect to V by For n = 1 this looks graphically as follows: for f ∈ Hom C (U ⊗ V, V ). A fundamental property of the partial trace (3.2) is the following: For any morphism f ∈ Hom C (U ⊗ V, V ) in a ribbon category one has Proof . This follows directly by invoking functoriality of the twist: Alternatively one may 'drag the morphism f along the V -loop'; graphically: Here the first and third equalities hold by definition of the left and right dual morphisms ∨ f and f ∨ , respectively (and by monoidality of σ), the second equality is sovereignty, and the final equality follows by the functoriality of the braiding.
By the same sequence of manipulations one obtains analogous identities for other partial traces. In particular we have the following generalization of the cyclicity property of the ordinary trace: Lemma 3.2. Partial traces are cyclic up to braidings and twists: For any pair of morphisms f ∈ Hom C (U ⊗ X, Y ) and g ∈ Hom C (V ⊗ Y, X) in a ribbon category one has (3.5) Remark 3.3. Instead of dragging, as in (3.4), the morphism f along the V -loop in clockwise direction, we could equally well drag it in counter-clockwise direction. In the situation considered in Lemma 3.2, this amounts to the identity instead of (3.5). This is, however, equivalent to (3.5), as is seen by combining it with the result (3.3), as applied to the morphism f •(id U ⊗g), and using that Example 3.4. The category SVect of super vector spaces over a field k is k-linear rigid monoidal and is endowed with a braiding given by the super-flip map. SVect is semisimple with, up to isomorphims, two simple objects, namely the tensor unit 1 which is the 1-dimensional vector space k in degree zero, and an object S given by k in degree one. There are two structures of ribbon category on this braided rigid monoidal category. For the first, the twist obeys (besides θ 1 = id 1 , which is true in any ribbon category) θ S = − id S , which after identifying ∨ S = S ∨ results in a strictly sovereign structure and in dim(S) = 1. For the second ribbon structure instead the twist is trivial, θ S = id S , while the sovereign structure is non-trivial, namely σ S = − id S ∨ (upon still identifying ∨ S = S ∨ ), and the trace is a supertrace, in particular now dim(S) = −1. This second ribbon structure arises from the conventions used in the bulk of the literature on associative and Lie superalgebras (see e.g. [30]), while in many applications of super vector spaces in supersymmetric quantum field theory the first ribbon structure is implicit.

The Killing form of a Lie algebra in a symmetric ribbon category
We now turn to morphisms involving Lie brackets. Accordingly in the sequel our Convention 2.1 will again be in effect. In particular, as the braiding of C is now symmetric, the twist of C squares to the identity, and θ ≡ id is an allowed twist. In a symmetric ribbon category, we give Writing out the partial trace, this expression reads more explicitly In the sequel we will use the graphical notation The presence of the twist in Definition 3.5, which might not have been expected, is needed for the Killing form to play an analogous role in the structure theory as in the classical case of Lie algebras in Vect k . For comparison we will below also briefly comment on the morphism which of course coincides with κ if, as in the classical case, θ L = id L . As illustrated by Example 3.4, there are important classes of ribbon categories for which the sovereign structure is not strict. Nevertheless, for convenience in the sequel we will suppress the sovereign structure or, in other words, take it to be strict. We are allowed to do so because it is entirely straightforward to restore the sovereign structure in all relevant expressions.
Example 3.6. For the matrix Lie algebra (A U , m U ), introduced in Example 2.20(ii) as the commutator Lie algebra of the matrix algebra A U = U ⊗ U ∨ , for any U ∈ C, the Killing form evaluates to Further, recall that if the dimension dim(U ) is invertible, then we have a direct sum decomposition A U = 1 ⊕ A U . Using the graphical description of the corresponding idempotents, the Killing form on the ideal A U reads Proposition 3.7. The Killing form is symmetric, i.e. satisfies Proof . By functoriality of the twist we can rewrite the Killing form as Using the cyclicity (3.5) of the partial trace we thus get The claim now follows by noticing that, via functoriality of the twist, the right hand side of (3.9) equals tr 3 The morphism κ 0 defined in (3.6) satisfies κ 0 = κ 0 • c L,L • (θ L ⊗ id L ) instead of (3.7).
Example 3.8. Recall from Example 3.4 that there are two ribbon structures on the category SVect. For both of them the Killing form κ of a Lie algebra L in SVect, i.e. a Lie superalgebra, is supersymmetric, κ(y, x) = (−1) |x||y| κ(x, y) for homogeneous elements x, y ∈ L. For the ribbon structure that has trivial twist, this trivially holds for κ 0 as well, while for the one with nontrivial twist (and strict sovereign structure), κ 0 is instead symmetric, κ 0 (y, x) = κ 0 (x, y).

Invariance of the Killing form
We are now going to show that the Killing form is invariant. The proof relies on the use of the Jacobi identity. In the graphical description, the Jacobi identity (slightly rewritten using that the braiding is symmetric) reads (here, as well as in the pictures below, we omit the obvious L-labels).
Graphically, = . (3.12) Proof . We write the Killing form as in (3.8) and invoke the Jacobi identity in the form of (3.10) to express the left hand side of (3.12) as where the second equality holds because of the cyclicity property (3.5). On the other hand, applying the Jacobi identity to the right hand side of (3.11) yields = − . (3.14) Analogously as in the proof of symmetry, the claim then follows noticing that the expressions on the right hand sides of (3.13) and (3.14) are equal, owing to the functoriality of the twist.

Implications of non-degeneracy of the Killing form
Non-degeneracy of the Killing form plays an important role for Lie algebras in categories of vector spaces. It still does so for Lie algebras in generic additive symmetric ribbon categories. We start by recalling the appropriate notion of non-degeneracy of a pairing on an object U , i.e. for a morphism in Hom C (U ⊗ U, 1): Definition 3.10. A pairing U ∈ Hom C (U ⊗ U, 1) in a monoidal category C is called non-degenerate iff there exists a copairing − U ∈ Hom C (1, U ⊗ U ) that is side-inverse to U , i.e. iff the adjointness relations We describe the equalities (3.15) graphically as follows: If the pairing is symmetric, then each of the two equalities in (3.15) implies the other. If the monoidal category C has a right (say) duality, an equivalent definition of non-degeneracy of U is that the morphism ∧ is an isomorphism, implying in particular that U ∨ is isomorphic, albeit not necessarily equal, to U .
We have immediately the Lemma 3.11. If a Lie algebra has a non-degenerate Killing form, then its only Abelian retract ideal is 0.
Proof . Assume that (K, e, r) is an Abelian retract ideal of a Lie algebra L. Then, with p = e•r, Using that κ is non-degenerate, (3.16) We are now in a position to establish It follows in particular that or, graphically, where in the second equality it is used that M is an ideal, and in the third that κ is invariant. By the universal property of M as the kernel of p, (3.18) implies showing that (M , e , r ) is in fact a retract ideal of L.
Now take M to be a primitive retract ideal. Then either p • p = p or p • p = 0. In case that p • p = p, it follows from (3.17) that κ • (e ⊗ e) = 0, which in turn implies and therefore, when combined with Lemma 3.11, contradicts the non-degeneracy of κ. We thus conclude that p • p = 0, and hence, again by non-degeneracy, that the Lie algebra L is the direct sum of the retract ideals M and M . Furthermore, κ • (e ⊗ e) and κ • (e ⊗ e ) are just the Killing form of M and M , respectively.
The claim now follows by iteration, which terminates after a finite number of steps because End C (L) only has finitely many idempotents.
Remark 3.13. A large class of categories for which the assumptions of Proposition 3.12 are satisfied for any Lie algebra are Krull-Schmidt categories, for which every object is a finite direct sum of indecomposables.
Remark 3.14. For C the category of finite-dimensional vector spaces over a field (of arbitrary characteristic), the proof of Proposition 3.12 reduces to the one given in [13].
As is well known, for Lie algebras in Vect k , semisimplicity does not imply non-degeneracy of the Killing form unless k is a field of characteristic zero. Still, the following weaker statement holds. Then we have which implies that M is a retract ideal of L. Since L is indecomposable, this means that M either equals L or is zero. In the former case, (3.19) says that κ is zero, while the latter case amounts to κ being non-degenerate.

A.1 Categorical background
For the sake of fixing terminology and notation, we recall a few pertinent concepts from category theory. Besides in standard textbooks, details can e.g. be found on the nLab web site 1 ; for a condensed exposition of several of the relevant notions see [35].
A subobject (U , e) of an object U is a pair consisting of an object U and a monomorphism e ∈ Hom C (U , U ) or, to be precise, an equivalence class of such pairs, with equivalence defined via factorization of morphisms with the same codomain (but it is common to use the term subobject also for individual representatives.) A simple object is an object that does not have any nontrivial subobject. A retract of an object U is a subobject of U for which e has a left-inverse, i.e. is a triple (U , e, r) such that (U , e) is a subobject of U and r ∈ Hom C (U , U ) is an epimorphism satisfying r • e = id U . A retract (U , e, r) for which the idempotent p = e • r ∈ End C (U ) cannot be written as the sum of two non-zero idempotents is called primitive.
The image of a morphism f ∈ Hom C (U, V ) is a pair (V , h) consisting of an object V and a monic h ∈ Hom C (V , V ) such that there exists a morphism g ∈ Hom C (U, V ) satifying h • g = f and such that (V , h) is universal with this property, i.e. for any object W with a monic k ∈ Hom C (W, V ) and morphism j ∈ Hom C (U, W ) satifying k • j = f there exists a unique morphism l ∈ Hom C (V , W ) such that k • l = h. In short, in case it exists, the image is the smallest subobject (V , h) of V through which the morphism f factors.
A preadditive category is a category C which is enriched over the monoidal category of Abelian groups, i.e. for which the morphism set Hom C (U, V ) for any pair of objects U, V ∈ C is an Abelian group and composition of morphisms is bilinear. An additive category is a preadditive category admitting all finitary biproducts, or direct sums. A direct sum is unique up to isomorphism, and the empty biproduct is a zero object 0 satisfying Hom C (U, 0) = 0 = Hom C (0, U ) for any object U ∈ C. An object U of an additive category is called indecomposable iff it is non-zero and has only trivial direct sum decompositions, i.e. U = U 1 ⊕ U 2 implies that U 1 = 0 or U 2 = 0. Each summand U i in a direct sum U = U 1 ⊕ U 2 ⊕ · · · ⊕ U n forms a retract (U i , e i , r i ), i.e. there are monics e i ∈ Hom C (U i , U ) and epis r i ∈ Hom C (U, U i ) such that for all i, j = 1, 2, . . . , n and n i=1 e i • r i = id U .
An indecomposable object thus only has trivial retracts, the zero object and itself. A primitive retract is an indecomposable direct summand. An idempotent p is an endomorphism satisfying p 2 = p; an idempotent p ∈ End C (U ) is called split iff there exists a retract (V, e, r) of U such that e • r = p. A category C is said to be idempotent complete iff every idempotent in C is split. In an idempotent complete category, any retract is a direct sum of primitive retracts. A Krull-Schmidt category is an additive category for which every object decomposes into a finite direct sum of objects whose endomorphism sets are local rings. The objects with local endomorphism rings are then precisely the indecomposable objects. Krull-Schmidt categories are idempotent complete. The class of Krull-Schmidt categories contains in particular every Abelian category in which each object has finite length, and thus e.g. all finite tensor categories in the sense of [14], as well as [25] every triangulated category whose morphism sets are finite-dimensional vector spaces and for which the endomorphism ring of any indecomposable object is local.
The data of a monoidal category (C, ⊗, 1, a, l, r) consist of a category C, the tensor product functor ⊗ : C×C→C, the monoidal unit 1 ∈ C, the associativity constraint a = (a U,V,W ), and left and right unit constraints l = (l U ) and r = (r U ) (see e.g. [39] for a review). If C is additive, then the tensor product is required to be additive in each argument. Every monoidal category is equivalent to a strict one, i.e. one for which the associativity and unit constraints are identities. We denote by U ⊗n the tensor product of n copies of U (which can be defined unambiguously even if C isn't strict). A braided monoidal category has in addition a commutativity constraint, called the braiding and denoted by c = (c U,V ), i.e. a natural family of isomorphisms c U,V ∈ Hom C (U ⊗ V, V ⊗ U ) satisfying two coherence conditions known as hexagon equations. A symmetric monoidal category is a braided monoidal category with symmetric braiding, i.e. for which c V,U • c U,V = id U ⊗V . A rigid (or autonomous) monoidal category has right and left dualities, with natural families d = (d U ) andd = (d U ) of evaluation and b = (b U ) andb = (b U ) of coevaluation morphisms, respectively. The objects that are right and left dual to U are denoted by U ∨ and ∨ U , respectively, so that Dualities are also defined naturally on morphisms, namely for f ∈ Hom C (U, V ) by respectively, whereby they furnish (contravariant) endofunctors of C. For a monoidal category with a right (say) duality, (U, e, r) is a retract of V iff (U ∨ , r ∨ , e ∨ ) is a retract of V ∨ .
A sovereign structure on a rigid monoidal category is a (choice of) monoidal natural isomorphism between the right and left duality functors, i.e. a natural family of isomorphisms σ U ∈ Hom C (U ∨ , ∨ U ) such that for all f ∈ Hom C (U, V ). A sovereign structure is equivalent [52, Proposition 2.11] to a pivotal (or balanced) structure, i.e. to a monoidal natural isomorphism between the (left or right) double dual functor and the identity functor. If the natural isomorphism defining a sovereign structure is the identity, the category is called strictly sovereign. In a sovereign category, an endomorphism f ∈ End C (U ) has a right trace tr(f ) ∈ End C (1), given (for strictly sovereign C) by tr(f ) =d U • (f ⊗ id U ∨ ) • b U , as well as an analogous left trace, and thus in particular every object U has a right (and analogously, left) dimension dim C (U ) ≡ tr(id U ) ∈ End C (1). When the left and right traces coincide, the category is called spherical.
A twist (or balancing) for a braided monoidal category with a (right, say) duality is a natural family θ = (θ U ) of isomorphisms θ U ∈ End C (U ) satisfying θ 1 = id 1 , θ U ⊗V = (θ U ⊗θ V )•c V,U •c U,V and θ U ∨ = (θ U ) ∨ . A braided rigid monoidal category equipped with a compatible twist is called a ribbon (or tortile) category. A ribbon category is sovereign, with sovereign structure

A.2 Algebras and coalgebras
We also collect a few elementary concepts concerning algebras in monoidal categories. An algebra (or monoid, or monoid object) in a monoidal category C = (C, ⊗, 1, a, l, r) is just a pair (A, m) consisting of an object A ∈ C and a morphism m ∈ Hom C (A ⊗ A, A); m is called a product or a multiplication, and by the usual abuse of terminology also the object A is referred to as an algebra. This concept is rather empty unless an interesting property of the product and/or additional structure on A is imposed, which in turn may require also the category C to have additional structure beyond being monoidal. Monoidal categories without any further structure furnish the proper setting for associative algebras, for which the associativity property is imposed, and for unital algebras, for which there exists a morphism η ∈ Hom C (1, A) satisfying Likewise, in any monoidal category there is the notion of a coassociative and co-unital coalgebra (C, ∆, ε), with morphisms ∆ ∈ Hom C (C, C ⊗ C) and ε ∈ Hom C (C, 1) that satisfy relations obtained by reversing the arrows in (A.3) and (A. In the sequel we will simplify such equalities by taking the monoidal category C to be strict. Still working merely in a (strict) monoidal category C there is an interesting way to combine algebras and coalgebras: Definition A.1.
(i) A Frobenius algebra A in C is a quintuple (A, m, η, ∆, ε) such that (A, m, η) is a unital associative algebra, (A, ∆, ε) is co-unital coassociative coalgebra, and such that the coproduct is a morphism of A-bimodules, i.e.
(ii) A strongly separable 2 Frobenius algebra is a Frobenius algebra for which the product is a left inverse of the coproduct, i.e. m • ∆ = id A .
(i) Coalgebras, and in particular Frobenius algebras, in symmetric monoidal categories which admit the notion of an adjoint (or dagger [46]) of a morphism arise naturally in a categorical formalization of fundamental issues of quantum mechanics, like complementarity of observables or the no-cloning theorem; see e.g. [7,8,34].
(ii) To a pair of functors F : C→D and G : D→C, such that G is right adjoint to F , there is naturally associated a monad in C, i.e. an algebra in the category of endofunctors of C.
If G is also left adjoint to F , then this monad has a natural Frobenius structure [49,Proposition 1.5].
If the monoidal category C is in addition braided, one can also define (co-)commutativity as well as bi-and Hopf algebras. An algebra (A, m) in a braided (strict) monoidal category (C, ⊗, 1, c) is said to be commutative iff m • c A,A = m; dually, a coalgebra (C, ∆) in C is cocommutative iff c A,A • ∆ = ∆. A bialgebra B is both an (associative unital) algebra and a (coassociative co-unital) coalgebra, with the coproduct and counit being algebra morphisms, i.e. ∆ • m = (m ⊗ m) • (id B ⊗c B,B ⊗ id B ) • (∆ ⊗ ∆) and ε • m = ε ⊗ ε. A Hopf algebra H is a bialgebra with an antipode s ∈ End C (H) that is a two-sided inverse of id H for the convolution product.
(i) If C is braided monoidal and countable direct sums are defined in C, 3 then for any object U ∈ C the object T (U ) := ∞ n=0 U ⊗n carries a natural structure of a unital associative algebra, with the product defined via the tensor product of the retracts U ⊗n and the unit morphism given by the retract embedding of U ⊗0 = 1, see e.g. [12,44]; T (U ) is called the tensor algebra of U . In fact [44,Corollary 2.4], T (U ) has in fact a natural bi algebra structure, and by extending the endomorphism − id U of U to an anti-algebra morphism of T (U ) defines an antipode, so that T (U ) is even a Hopf algebra. T (U ) can also be defined by a universal property, it is an associative algebra together with a morphism i U ∈ Hom C (U, T (U )) such that for any algebra morphism f from U to some algebra A in C there exists a unique morphism g from T (U ) to A such that f = g • i U .
(ii) If in addition C is idempotent complete, is k-linear with k a field of characteristic zero, and has symmetric braiding, then the direct sum S(U ) := ∞ n=0 S n (U ) of all symmetric tensor powers of U carries a natural structure of a commutative unital associative algebra.
(iii) Hopf algebras appear naturally in constructions within categories of three-dimensional cobordisms. Specifically, the manifold obtained from a 2-torus by excising an open disk is a Hopf algebra 1-morphism in the bicategory of three-dimensional cobordisms with corners [10].
(iv) For a braided finite tensor category C the coend U ∈C U ∨ ⊗U exists and can be used to construct quantum invariants of three-manifolds and representations of mapping class groups [36]. It also plays a role (for C semisimple, i.e. a fusion category) in establishing the relationship between Reshetikhin-Turaev and Turaev-Viro invariants [50]. For these results it is crucial that this coend carries a natural structure of a Hopf algebra (with some additional properties) in C.
Remark A.4. A Hopf algebra H in an additive rigid braided category which has invertible antipode and which possesses a left integral Λ ∈ Hom C (1, H) and right cointegral λ ∈ Hom C (H, 1) such that λ • Λ ∈ End C (1) is invertible, naturally also has the structure of a Frobenius algebra with the same algebra structure, and with the Frobenius counit given by λ [19,31,40].
Frobenius algebras F can, just like in the classical case, also be defined with the help of a non-degenerate invariant pairing ∈ Hom C (F ⊗ F, 1); in terms of the description chosen above, = ε • m, see e.g. [20,51]. A symmetric Frobenius algebra is a Frobenius algebra F for which the pairing is symmetric.
Example A.5. In two-dimensional rational conformal field theory (RCFT), full local RCFTs that come from one and the same chiral RCFT are in bijection with Morita classes of strongly separable symmetric Frobenius algebras of non-zero dimension in the modular tensor category that underlies the chiral RCFT [18].
thank Christoph Schweigert for discussions and Scott Morrison for bringing Example 2.20(iv) to our attention. JF is supported by VR under project no. 621-2013-4207. JF thanks the Erwin-Schrödinger-Institute (ESI) for the hospitality during the programs "Modern Trends in TQFT" and "Topological Phases of Quantum Matter" while part of this work was pursued.