Arxiv:2011.12501V2 [Math.CT] 14 Feb 2021 Esmerc(Rsprrie)Mnia Uectgre.I T in Supercategories

arXiv:2011.12501v2 [math.CT] 14 Feb 2021 esmerc(rsprrie)mnia uectgre.I t In supercategories. monoidal de [ to superbraided) Kapranov is (or of paper persymmetric ideas this following of superalgebras, purpose supercommutative The of places. cat in other role among fundamental theory, a tion played have categories These category. .Introduction 1. Swsspotdb S rn DMS-1453893. grant NSF by DMS-1812462. supported grant was NSF AS by supported was SS 2020 Date b (or symmetric a is algebra commutative a of analog categorical The periodicity Bott References A. duality Appendix Schur–Sergeev eversion 9. Clifford spaces vector 8. Queer spin-species 7. Linear groups symmetric 6. Spin theory 2-category 5. Some structures 4. Monoidal 3. Supercategories 2. eray1,2021. 13, February : ahmtc ujc Classification. Subject Mathematics da fKpao.Ruhy hsi oodlctgr nwihteo the which in category monoidal a is this Roughly, are phisms Kapranov. of ideas hoyt etrudrtn oeapcso h ue supera queer the of aspects su some of monoidal understand symmetric better of supersym to 2-category theory of corresponding 2-category a certain and a o egories ca between algebra introduce equivalence we exterior an construction the defines a linear is is of first second category The The the Ganter–Kapranov). as constructions. tenso (such general half these two the from also with derived equipped be spaces can th vector examples queer examples: of fundamental category two the are There objects. homogeneous Abstract. √ nteter fQsmercfntosadShrSrevdualit Schur–Sergeev and functions Q-symmetric of theory the in 2 UESMERCMNIA CATEGORIES MONOIDAL SUPERSYMMETRIC Z edvlpteie fasprymti oodlsupercategory, monoidal supersymmetric a of idea the develop We / -rdd qipdwt isomorphisms with equipped 2-graded, TVNVSMADADE SNOWDEN ANDREW AND SAM V STEVEN 1. 80,05E10. 18M05, Introduction Contents 1 X ⊗ Y → gba uha eti factors certain as such lgebra, ruodo pnst,and spin-sets, of groupoid e rdc;ohrimportant other product; r Y ldCiodeeso.This eversion. Clifford lled ercmnia supercat- monoidal metric ectgre.W s our use We percategories. uectgr deto (due supercategory a f gr hoyadrepresenta- and theory egory ers fti introduction, this of rest he ⊗ pnseis.Teeare There species). spin Ka X eo aeoia analog categorical a velop y. , fparity of add iermonoidal linear raided) § .] ecl hs su- these call we 3.4]; jcsadmor- and bjects | X following || Y | on 59 58 54 45 41 36 30 24 11 8 1 2 STEVENVSAMANDANDREWSNOWDEN

we explain the basic idea behind the deﬁnition, some of our results and applications, and compare our contributions to the existing discussions of supercategories in the literature.

1.1. The definition. Recall that a supercommutative superalgebra is a Z/2-graded algebra x y such that xy = ( 1)| || |yx holds for all homogeneous elements x and y. Here we write x Z/2 for the degree− of a homogeneous element x. Thus we expect that in a superbraided |monoidal| ∈ category, there should be even and odd homogeneous objects, and there should be X Y an isomorphism between X Y and ( 1)| || | “times” Y X for homogeneous objects X and Y . ⊗ − ⊗ The first puzzle here is what it means to negate an object. This has been well-understood for some time. Let V be a finite dimensional super vector space. (Recall that a super vector space is simply a Z/2-graded vector space.) The super dimension of V is defined to be sdim(V ) = dim(V0) dim(V1). Let Π(V ) be the super vector space given by Π(V )i = Vi+1, i.e., Π(V ) is obtained− from V by interchanging the two graded pieces. Then sdim(Π(V )) = sdim(V ). Thus Π(V ) can be thought of as the negative of V . − The identity map V Π(V ) is an odd degree isomorphism. This map characterizes Π(V ) uniquely up to even isomorphism:→ if W is any super vector space equipped with an odd degree isomorphism V W , then there is a unique even isomorphism W Π(V ) making the triangle commute.→ We can thus speak of Π(V ) for any object V in a category→ in which the morphisms are Z/2-graded. Precisely, we define a supercategory to be a category enriched over super vector spaces, so that all Hom spaces are super vector spaces. Given an object V in a supercategory, we define Π(V ) to be an object equipped with an odd degree isomorphism V Π(V ), if such an object and morphism exist. (We assume in what follows that they always→ do.) We can thus regard Π(V ) as the negative of V . Following our initial intuition, suppose that we have a Z/2-supercategory (i.e., one in which the objects are Z/2-graded) equipped with a monoidal structure . Then we see that a superbraiding should be a natural isomorphism ⊗ X Y β : X Y Π| || |(Y X), X,Y ⊗ → ⊗ defined for all homogeneous objects X and Y . Of course, as with ordinary braidings, our superbraiding should satisfy some version of the hexagon axioms. It turns out that the usual hexagon axioms are inconsistent in this setting due to the signs that arise from composition in supercategories, but one can attempt to resolve the inconsistencies by allowing the hexagons to commute up to scalars. This ends up working if we assume the ground field has an eighth root of unity; however, the scalars that come up are not at all obvious. By working with Z/4-supercategories (or Z/q-supercategories for any q divisible by 4), one can get by with much nicer systems of scalars that only involve signs (and thus make sense over general fields). In both cases, we introduce a notion of factor system (Definition 3.5) to axiomatize the conditions on the scalars that appear in the hexagon axioms. One of the main contributions of this article is identifying these conditions and verifying them in important examples of interest. We have thus arrived at the main definition: a superbraiding is a natural isomorphism β as above such that the hexagon axioms hold up to scalars determined by a choice of factor system. See §3.5 and §3.6 for details.

1.2. The 2-categorical perspective. One can deﬁne the notion of supercommutative superalgebra in elementary terms, as we did above. There is also a more abstract approach: SUPERSYMMETRIC MONOIDAL CATEGORIES 3

0-category 1-category Vector space Supercategory Commutative algebra Symmetric monoidal supercategory Super vector space Z/q-supercategory Commutative superalgebra Symmetric monoidal Z/q-supercategory Supercommutative superalgebra Supersymmetric monoidal Z/q-supercategory

Figure 1. Some analogies relevant to this paper. Here q should be even.

one endows the monoidal category of super vector spaces with a new symmetry τ (namely x y x y ( 1)| || |y x), and realizes supercommutative superalgebras as the commutative algebra⊗ 7→ objects− in this⊗ category. The more abstract approach has several benefits: for example, any properties of commutative algebras in symmetric tensor categories can immediately be applied to supercommutative superalgebras, and it also enables one to easily consider similar structures, such as Lie superalgebras. In our discussion of superbraided supercategories above, we have taken the elementary approach. We now explain a version of the abstract approach. Let C be the 2-category of Z/q-supercategories, with q even. This 2-category has a natural monoidal structure ⊠, given by a standard product construction for supercategories. Given two Z/q-supercategories A X Y and B, we define T: A⊠B B⊠A to be the equivalence given by X ⊠Y Π| || |(Y )⊠X, where X and Y are homogeneous→ objects of A and B. One should think of7→ the equivalence T as analogous to the isomorphism τ from the previous paragraph. We show that T (together with a choice of factor system) defines a symmetry on C in §4. One can then regard supersymmetric monoidal supercategories as the symmetric monoidal objects in the symmetric monoidal 2-category (C , T). 1.3. Spin-symmetric groups. The most fundamental example of a symmetric monoidal category is the groupoid of finite sets under disjoint union. A skeletal version S of this category can be constructed explicitly, as follows. There is one object [n] for each natural number n. We put EndS([n]) = Sn and HomS([n], [m]) = ∅ for n = m. The monoidal operation is defined on objects by [n] [m] = [n + m] and on morphisms6 using the natural ∐ ∐ homomorphism Sn Sm Sn+m. Finally, the symmetry isomorphism [n] [m] [m] [n] is given by the element× τ → S defined by ∐ → ∐ n,m ∈ n+m i + m if 1 i n τ (i)= ≤ ≤ . n,m i n if n +1 i n + m ( − ≤ ≤ The hexagon axioms amount to some identities satisfied by the τn,m elements. The above picture has an analog in the setting of supersymmetric monoidal categories in which the symmetric groups are replaced by spin-symmetric groups. Recall that the nth spin-symmetric group Sn is a certain central extension of Sn by Z/2; see §5.2. We define a skeletal category S˜ as follows. There is again one object [n] for each natural number n.

We put EndS˜([n]) = k[eS˜ ]/(c + 1), where c S is the distinguished central element, and n ∈ n define Hom˜([n], [m]) = 0 if n = m. The monoidal operation on S˜ is defined on objects by S 6 ∐ [n] [m] = [n + m], and on morphisms using thee natural map S S S . Finally, ∐ n × m → n+m the supersymmetry is induced by a particular choice of liftτ ñ,m Sn+m of the element τn,m e∈ e e e 4 STEVENVSAMANDANDREWSNOWDEN

defined in the previous paragraph. In §5.4, we verify that the axioms of a supersymmetric monoidal category hold in this case; this amounts to establishing some identities satisfied by theτ ñ,m elements. The definition and study of the elementsτ ñ,m (in §5.3), while totally elementary, is one of the most important parts of this paper. For a particular choice of n and m, there are two lifts of τn,m to Sn+m, and neither is preferred. We find, however, that the liftsτ ñ,m we define have important compatibility properties as n and m vary; in other words, there is a preferred system ofe lifts (up to the action of a single Z/2). This strikes us as an important discovery. In fact, as discussed in §1.6 below, we show that essentially every supersymmetry can, in a sense, be constructed from these elements. The category S is not merely the simplest example of a symmetric monoidal category: it is also universal. Indeed, suppose that C is a symmetric monoidal category. Given an object X n of C, the symmetric group Sn naturally acts on X⊗ . This construction induces a symmetric n monoidal functor S C via [n] X⊗ . This observation shows that (S, [1]) is the universal symmetric monoidal→ category equipped7→ with an object, in the sense that giving a symmetric monoidal functor S C is equivalent to giving an object of C. → An analogous result holds for S˜. Indeed, if C is a supersymmetric monoidal Z-category and X is a degree 1 object in C then we show that the spin-symmetric group Sn naturally n acts on X⊗ . This leads to the following theorem: e Theorem 1.1. (S˜, [1]) is the universal supersymmetric monoidal Z-supercategory equipped with an object of degree 1. Another important example of a supersymmetric category comes from the representation theory of spin-symmetric groups. A linear spin-species is a sequence (Vn)n 0 where Vn is ≥ a representation of Sn (that interacts with super structures appropriately). The category of such sequences, denoted Rep(S ), admits a monoidal structure defined by inducing from ∗ Young subgroups. Thee τ ñ,m elements can be used to construct a supersymmetry on this category. The details are carriede out in §6.

1.4. The half tensor product. The second most fundamental example of a symmetric monoidal category is the category Vec of vector spaces under tensor product. The category SVec of super vector spaces is not supersymmetric: it is a “plain” symmetric monoidal supercategory since the isomorphism V W W V has even degree. (The same applies to any category where the tensor product⊗ and→ symmetry⊗ are inherited from SVec, such as the category of representations of a Lie superalgebra.) However, there is a supersymmetric analog to Vec, that we now discuss. Let V be a super vector space. A queer structure on V is an odd degree automorphism ν : V V that squares to the identity. Suppose that U and V are super vector spaces equipped→ with queer structures µ and ν. Then µ ν is an even degree automorphism of ⊗ 1 U V that squares to 1. The half tensor product of U and V , denoted 2− (U V ), is deﬁned⊗ to be the ζ -eigenspace− of µ ν. (Here ζ = √ 1.) The choice of ζ does not⊗ matter 4 ⊗ 4 − 4 much, as µ 1 (or 1 ν) interchanges the +ζ4 and ζ4 eigenspaces. This construction⊗ appears⊗ frequently in the representation− theory of superalgebras. Sup- pose that A is a C-superalgebra and M is a ﬁnite dimensional simple A-module. The super version of Schur’s lemma states that EndA(M) is either 1-dimensional (“type M”) or 2-dimensional (“type Q”); in the latter case, the algebra EndA(M) is generated over SUPERSYMMETRIC MONOIDAL CATEGORIES 5

C by a queer structure. Now suppose that B is a second C-superalgebra and N is a finite dimensional simple B-module. If either M or N has type M then M N is a simple 1 ⊗ A B-module. If both M and N have type Q then 2− (M N) is a simple module, and ⊗ 1 1 1 ⊗ M N ∼= 2− (M N) C | . (See [CW, §3.1.3] for details.) Returning⊗ to the⊗ general⊗ situation, let (U, µ) and(V, ν) be as above. A simple computation shows that the natural symmetry τU,V : U V V U carries the ζ4-eigenspace of µ ν into ⊗ → ⊗ 1 1 ⊗ the ( ζ4)-eigenspace of ν µ. In particular, τU,V does not map 2− (U V ) into 2− (V U). − ⊗ 1 1 ⊗ ⊗ However, the composition (ν 1) τU,V does map 2− (U V )to2− (V U). This is an odd degree isomorphism, which suggests⊗ ◦ a lurking supersymmetry.⊗ ⊗ In §7, we define a Z/2-supercategory Queer by taking the degree 0 objects to be super vector spaces and the degree 1 objects to be super vector spaces equipped with a queer structure. We define a monoidal operation on this category by using the half tensor product on two degree 1 objects and the ordinary tensor⊙ product on other pairs of objects. We show that this does indeed define a monoidal structure; this is not obvious, as the associator maps are non-trivial. We further show that the construction in the previous paragraph (scaled by an appropriate root of unity) defines a supersymmetry on this monoidal structure. Thus Queer is an example of a supersymmetric monoidal Z/2-supercategory. We believe this places the half tensor product construction in its proper theoretical framework.

1.5. Exterior algebras. Let A be a C-linear category. One can then deﬁne the nth sym- n metric power of A, denoted Sym (A), to be the category of Sn-equivariant objects in the nth power of A (for an appropriate notion of power). The sum of these categories, denoted Sym(A), is naturally a symmetric monoidal category, and is the universal such category to which A maps. Now suppose that A is a supercategory. Ganter–Kapranov [GK] deﬁned the nth exterior n power of A, denoted (A), to be the category of appropriately Sn-equivariant objects in the nth power of A. We show that the sum of these categories, denoted (A), is naturally a supersymmetric monoidalV supercategory, and has a universal propertye similar to the symmetric power. This construction provides a large source of supersymmetricV categories. See §6.8 for details.

1.6. Clifford eversion. Let A be a nice Z/q-supercategory, with q even; “nice” means that A is additive, Karoubian, and admits a Π structure. In §8, we define a new nice Z/q- supercategory Cl(A) called the Clifford eversion of A by considering certain kinds of Clifford modules in A. (This construction is essentially a generalization of the category Queer discussed above, see §7.5.) If A has a symmetric monoidal structure then we show that Cl(A) naturally has a supersymmetric monoidal structure; this construction again hinges on theτ ñ,m elements. Moreover, we prove: Theorem 1.2. Clifford eversion defines an equivalence between the 2-category of nice symmetric monoidal Z/q-supercategories and the 2-category of nice supersymmetric monoidal Z/q-supercategories. Let C be the 2-category of nice Z/q-supercategories. This category has a natural monoidal structure that admits an obvious symmetry Σ; it also admits the less obvious symmetry T discussed in §1.2. In fact, Clifford eversion induces an equivalence of symmetric monoidal 2-categories (C , Σ) (C , T). This recovers the above theorem, but is a more fundamental → 6 STEVENVSAMANDANDREWSNOWDEN

statement. However, we only partially prove this statement since the notion of equivalence of symmetric monoidal 2-categories is rather complicated. Let A be nice symmetric monoidal Z/q-supercategory with an exact structure, and let K+(A) be the quotient of the Grothendieck group by the relations [Π(X)] = [X]. We show (Proposition 8.21) that there is a natural isomorphism of rings A √ A √ K+( )[1/ 2] ∼= K+(Cl( ))[1/ 2]. This helps explain the appearance of certain factors of √2 in the theory of Q-symmetric functions.

1.7. Sergeev–Schur duality. (We take the base field to be C in this section.) Classical Schur–Weyl duality connects the representation theory of the general linear group and the symmetric groups. Let Reppol(GL) be the category of polynomial representations of the infinite general linear group; this is the tensor category generated by the standard repre- n sentation V. The symmetric group Sn acts on V⊗ by permuting tensor factors, and so pol n we obtain a functor Rep(Sn) Rep (GL) by M M C[Sn] V⊗ . Let Rep(S ) be → 7→ ⊗ ∗ the category of sequences (Vn)n 0 where Vn is a representation of the symmetric group Sn; such sequences are sometimes called≥ linear species. Then one formulation of Schur–Weyl duality states that the above functor extends to an equivalence Rep(S ) Reppol(GL). This equivalence respects various additional structures on the two categories.∗ → In particular, it is an equivalence of symmetric monoidal categories, where Reppol(GL) is endowed with the usual tensor product and Rep(S ) is given the induction tensor product, defined by ∗

Sn (V W )n = IndSi Sj (Vi Wj). ⊗ × ⊗ i+j=n M This equivalence is extremely useful when studying algebra within the tensor category Rep(S ), as it provides a bridge to more classical theory. Sergeev–Schur∗ duality is an analog of Schur–Weyl duality in which the general linear group is replaced by the queer Lie superalgebra. Precisely, let V = C∞|∞ and fix an odd-degree isomorphism α: V V squaring to 2. The infinite queer Lie superalgebra q is defined → n to be the subalgebra of gl(V) that preserves α. The symmetric group Sn acts on V⊗ by permuting tensor factors. However, there are additional natural endomorphisms that commute with q: namely, we can apply α to any tensor factor. To make this more precise, let Cln denote the nth Clifford algebra; this has n anti-commuting generators that each square to 2. Define the Hecke–Clifford algebra Hn to be the semi-direct product Sn ⋉ Cln. n Then Hn acts on V⊗ and commutes with q. As in the previous paragraph, we obtain an induced functor Rep(H ) Reppol(q). One formulation of Sergeev–Schur duality is that this is an equivalence (and∗ → respects the natural symmetric monoidal structures). There is a close relationship between the algebra Hn and the spin-symmetric groups: we have an isomorphism H = k[S ]/(c + 1) Cl . Thus an H -module can be viewed as a n ∼ n ⊗ n n Clifford module in the category Rep(Sn) of linear spin species. We thus see that Rep(H ) ∗ can be identified with the Clifforde eversion of the category Rep(S ) of sequences of Sn- representations. We can thus rephrasee the above discussion as follows:∗ e e Theorem 1.3. The category Reppol(q) is equivalent to the Clifford eversion of the category Rep(S ) of linear spin species. ∗ e SUPERSYMMETRIC MONOIDAL CATEGORIES 7

We show that this equivalence respects natural symmetric monoidal structures on the two categories. We view the above theorem as an elegant expression of Sergeev–Schur duality, as it provides a direct link between the queer algebra and the spin-symmetric groups. See §9 for details. The Grothendieck ring of Reppol(q) is naturally identified with the ring Γ of odd symmetric functions. This ring has a basis consisting of the Schur Q-functions Qλ. However, they do not represent classes of simple objects in general: one must adjust them by appropriate powers of √2 (which depend on λ as well as which category is being used). One of our goals was to better understand the nature of these powers of √2, and the effect of Clifford eversion on Grothendieck rings is one piece of that puzzle. Unfortunately, there is one more piece which is not explained by our general framework: some simple modules of k[Sn]/(c +1) carry an action of Cl1, while others do not, and this also features into the powers of √2 that appear. e 1.8. Relation to previous work. Kapranov proposed a sketch for the definition of a supersymmetric monoidal supercategory in [Ka, §3.4]. A large portion of this article is devoted to providing the details for this sketch. Versions of (monoidal) supercategories have been considered in [BE], [KKT], [EL]. How- ever, these sources do not treat the notion of (super)braiding or (super)symmetry for a monoidal supercategory, which as far as we know, is treated in detail for the first time here. The Clifford twist in [KKT] is similar to our notion of Clifford eversion, but is not the same (for instance, the Clifford twist is an involution even if the coefficient field does not contain √ 1). − 1.9. Open problems. We mention a few open problems: (a) In Remark 3.16, we observe a connection between even factor systems and degree 3 group cohomology. However, this connection is not complete, and it does not apply to odd factor systems (which is the relevant case for supersymmetries). It would be helpful to have a better understanding of these issues. (b) Is there a conceptual a priori explanation for why the 2-categories of symmetric and supersymmetric monoidal supercategories are equivalent? (c) Is there a “natural” construction of the groupoid of spin-sets? That is, is there a natural category where the objects are some kind of “enhanced” finite sets whose automorphism groups are spin-symmetric groups? This might clarify a number of our constructions (such as theτ ñ,m elements). Is there a non-groupoid version? (d) We have provided explanations for most of the proof of Theorem 8.16, but it remains incomplete.

1.10. Notation. We fix a field k throughout which we assume does not have characteristic 2. One could allow k to be a commutative Z[1/2]-algebra, but we stick to fields for simplicity. We fix a compatible system of primitive 2-power roots of unity ζ2k in k; by compatible, we 2 mean that ζ2k+1 = ζ2k . In this article, “superalgebra” means an algebra with a Z/2-grading and a “commutative superalgebra” refers to a commutative algebra with a Z/2-grading (which does not affect the definition of commutative). This is sometimes used in the literature to mean a skew- commutative algebra; we reserve the phrase “supercommutative superalgebra” for this, i.e., x y an algebra A with a Z/2-grading such that xy = ( 1)| || |yx for homogeneous elements x, y A. − ∈ 8 STEVENVSAMANDANDREWSNOWDEN

2. Supercategories 2.1. Super vector spaces. A super vector space over k is a Z/2-graded k-vector space. Given a super vector space V , we write Vi for the subspace of homogeneous elements of degree i. For v V , we write v = i to indicate that v is a homogeneous element of degree ∈ | | i. For n Z, we let V [n] be the super vector space with V [n]i = Vn+i. Let V ∈and W be two super vector spaces and let f : V W be a linear map. We say that f is homogeneous of parity n if f(V ) W . We write→ Hom(V, W ) for the set of all i ⊆ i+n linear maps, and Hom(V, W )i for those of parity i. Every linear map decomposes uniquely as a sum of even and odd maps, and so we have a decomposition Hom(V, W ) = Hom(V, W ) Hom(V, W ) 0 ⊕ 1 In this way, we regard Hom(V, W ) itself as a super vector space. We let SVec be the category of super vector spaces and all linear maps, and SVec◦ the category of super vector spaces and even linear maps. The tensor product of two super vector spaces is just the tensor product of the underlying vector spaces, graded in the usual manner. Given homogeneous linear maps f : V W 1 1 → 1 and f2 : V2 W2, we deﬁne f1 f2 : V1 V2 W1 W2 to be the linear map given on homogeneous→ elements by the formula⊗ ⊗ → ⊗

v1 f2 (f f )(v v )=( 1)| || |f (v ) f (v ). 1 ⊗ 2 1 ⊗ 2 − 1 1 ⊗ 2 2 We caution that is not a functor SVec SVec SVec; this will be explained in detail ⊗ × → in §3.1. However, is a functor SVec◦ SVec◦ SVec◦, and naturally endows SVec◦ with the structure⊗ of a monoidal category.× → Let V and W be super vector spaces. We deﬁne isomorphisms σ, τ : V W W V for homogeneous elements v V and w W as follows: ⊗ → ⊗ ∈ ∈ v w σ(v w)= w v, τ(v w)=( 1)| || |w v. ⊗ ⊗ ⊗ − ⊗ Both σ and τ deﬁne symmetric structures on . However, the symmetric structure σ is uninteresting in that it does not interact with⊗ the grading. In what follows, we regard SVec◦ as a symmetric monoidal category using τ, unless explicitly said otherwise.

2.2. Supercategories. A supercategory is a category A enriched in the category SVec◦ of super vector spaces. Thus A is a category and HomA(X,Y ) has the structure of a super vector space for all objects X and Y such that composition deﬁnes an even map

HomA(Y,Z) HomA(X,Y ) HomA(X,Z), f g f g. ⊗ → ⊗ 7→ ◦ Given a supercategory A, we let A◦ be the category with the same collection of objects but only the even morphisms; that is, HomA◦ (X,Y ) = HomA(X,Y )0. Let Λ be an abelian group (or, for now, any set). A homogeneous Λ-supercategory is a collection A = Ai i Λ, where each Ai is a supercategory. We refer to Ai as the category of homogeneous{ objects} ∈ of degree i. Informally, an inhomogeneous Λ-supercategory is a supercategory A equipped with subcategories A for i A such that A is a direct sum of i ∈ the Ai’s, in some sense. This concept will appear in some examples, but we do not give a rigorous definition to avoid details of what direct sum means here (there are some subtleties if, e.g., A is non-additive or if Λ is infinite). In the inhomogeneous case, any construction will be determined by what it does on homogeneous objects, so we can typically restrict SUPERSYMMETRIC MONOIDAL CATEGORIES 9 our attention to them. In what follows, “Λ-supercategory” will by default refer to the homogeneous version, and we almost always take Λ to be Z or Z/q. Suppose that A is a Λ-supercategory and we have a function f :Λ Λ′. Then we can A A A → convert into a Λ′-supercategory by defining i = j f −1(i) j for i Λ′. Here the coprod- ∈ ∈ uct is taken in the sense of linear categories: the Hom space between objects in different factors is 0. In particular, we can convert a Z-supercategory` into a Z/q-supercategory, or a Z/q-supercategory into a Z/q′-supercategory if q′ divides q.

Example 2.1. Let G be a supergroup over C containing a central Gm. Then A = Rep(G) is an (inhomogeneous) Z-supercategory, by taking Ai to be the subcategory of objects where Gm acts with weight i. 2.3. Superfunctors. Let A and B be supercategories. A superfunctor F : A B is a functor (in the usual sense) such that the maps →

F : HomA(X,Y ) HomB(F (X), F (Y )) → are even maps of super vector spaces. If A and B are Λ-supercategories then a Λ-superfunctor is a collection of superfunctors F = Fi : Ai Bi i Λ. Let F and G be two superfunctors{ from→A to} ∈B. A homogeneous natural transformation η : F G of parity p Z/2 is a rule that assigns to every object X of A a → ∈ homogeneous morphism ηX : F (X) G(X) of degree p in B such that if f : X Y is a → p f → homogeneous morphism in A then G(f) ηX =( 1) | |ηY F (f). A general natural transformation is a sum of an even and an odd◦ natural− transformation.◦ The definition is the same for Λ-superfunctors: the grading does not affect the definition at all. 2.4. Π-structures. Let A be a supercategory. A Π-structure on an object X of A is an odd degree isomorphism ξX : X Π(X), for some object Π(X). AΠ-structure on A is a choice of Π-structure on each object→ of A. Such a structure need not exist. Note that if A is Λ-graded and X is homogeneous then Π(X) is necessarily homogeneous of the same degree. Suppose we have a Π-structure on A. For a homogeneous morphism f : X Y in A, → f define Π(f) to be the unique morphism making the following diagram commute up to ( 1)| | − f X / Y .

ξX ξY Π(f) Π(X) / Π(Y )

By uniqueness, it is clear that Π is a superfunctor A A, and that ξ : idA Π is an odd natural isomorphism. (If A is a Λ-supercategory then→ Π is a Λ-superfunctor.)→ Furthermore, taking f = ξX , we find Π(ξX )= ξΠ(X). For a non-negative integer n, we− let Πn be the n-fold iterate of Π. For n < m, we define n,m n m ξX : Π (X) Π (X) to be the composition ξΠm−1(X) ξΠn+1(X) ξΠn(X); for n > m, n,m→ m,n 1 n,m ◦···◦ ◦ n,m we define ξX = (ξX )− ; and for n = m, we let ξX be the identity. The map ξX is an 0,2 2 isomorphism of parity m n. In particular, we see that ξ : idA Π is an even degree isomorphism. For q even,− we define Πn for n Z/q to be the colimit→ of the system ∈ n−q,n n+q,n n q ξ n ξ n+q / Π − / Π // Π / . ··· ··· This colimit exists since all the maps are even isomorphisms. 10 STEVENVSAMANDANDREWSNOWDEN

Suppose that (Π,ξ) and (Π′,ξ′) are two Π-structures on A. Deﬁne iX : Π(X) Π′(X) 1 → to be the morphism ξ′ ξ− . Then i is an even degree isomorphism, and deﬁnes an even X ◦ X X degree isomorphism of functors i: Π Π′ compatible with ξ and ξ′. We thus see that any two Π-structures are canonically (and,→ in fact, uniquely) isomorphic by an even degree natural transformation.

Proposition 2.2. Let F : A B be a superfunctor between supercategories with Π-structures. Then there is a canonical even→ isomorphism F Π = Π F . ◦ ◦ 1 Proof. Let X be an object of A. Then ξF (X)F (ξX)− gives an even-degree isomorphism F (Π(X)) ∼= Π(F (X)). We now check that this is a natural isomorphism. Let f : X Y be a homogeneous morphism. Then →

f f ξ f =( 1)| |Π(f)ξ , ξ F (f)=( 1)| |Π(F (f))ξ . Y − X F (Y ) − F (X) 1 f 1 We apply F to the ﬁrst equation and rearrange to get F (f)F (ξ )− =( 1)| |F (ξ )− F (Π(f)). X − Y Post-compose this equation with ξF (Y ) and use the second equation to get

1 1 ξF (Y )F (ξY )− F (Π(f)) = Π(F (f))ξF (X)F (ξX )− , which shows that we have deﬁned a natural isomorphism F Π = Π F . ◦ ∼ ◦ 2.5. Product categories. Let A and B be two supercategories. We let A ⊠ˇ B be the product category, in the sense of enriched categories. Precisely, the objects of A ⊠ˇ B are pairs (A, B) with A A and B B; we write A ⊠ B in place of (A, B). The Hom spaces are deﬁned by ∈ ∈

HomA⊠ˇ B(A ⊠ B, A′ ⊠ B′) = HomA(A, A′) HomB(B, B′). ⊗

Given homogeneous morphisms f g : A ⊠ B A′ ⊠ B′ and f ′ g′ : A′ ⊠ B′ A′′ ⊠ B′′, their composition is deﬁned by ⊗ → ⊗ →

f g′ (f ′ g′) (f g)=( 1)| || |(f ′ f) (g′ g). ⊗ ◦ ⊗ − ◦ ⊗ ◦ The sign here comes from the symmetry of the tensor product in SVec. (To deﬁne the product of enriched categories in general, the enriching category needs a symmetric monoidal structure, and here we are using the symmetry τ on SVec◦.) Let A and B be two homogeneous Λ-supercategories. We deﬁne A ⊠ B to be the homogeneous Λ-supercategory with

(A ⊠ B)i = Aj ⊠ˇ Bk. j+ak=i We note that ⊠ˇ does not distribute over direct sums of supercategories (even finite direct sums), and so does not give the expected result on inhomogeneous supercategories; this is why we have used a different symbol in the Λ-graded case. If either A or B admits a Π-structure then so does A ⊠ B. Indeed, suppose A does, and let A ⊠ B be a given object of A ⊠ B. If ξ : A Π(A) is an odd degree isomorphism then ξ id : A ⊠ B Π(A) ⊠ B is an odd degree isomorphism,→ and so the claim is verified. ⊗ B → SUPERSYMMETRIC MONOIDAL CATEGORIES 11

2.6. Grothendieck groups. Let A be a supercategory equipped with an exact structure on A◦. We can then consider the Grothendieck group K(A◦) of A◦. For ε +, , we deﬁne ∈{ −} Kε(A) to be the quotient of K(A◦) by the relations [X] = ε[Y ] whenever there is an odd degree isomorphism X Y . If A admits a Π-structure then Kε(A) is simply the quotient of A → A A A K( ◦) by the relations [Π(X)] = ε[X]. If admits a Λ-grading then K( ◦)= i Λ K( i◦), ∈ and so K(A◦) is canonically Λ-grading. This induces a Λ-grading on Kε(A). L 3. Monoidal structures 3.1. Monoidal structures. Let A be a Λ-supercategory. A monoidal structure on A consists of: a Λ-superfunctor : A ⊠ A A; • a unit object 1; ⊗ → • an even natural isomorphism αX,Y,Z : X (Y Z) (X Y ) Z satisfying the • pentagon axiom, i.e., the following diagram⊗ is commutative⊗ → for⊗ all⊗ objects A, B, C, D: (A B) (C D) ⊗❥4 ⊗ ⊗❯❯ αA,B,C⊗D ❥❥❥❥ ❯❯❯❯αA⊗B,C,D ❥❥❥❥ ❯❯❯❯ ❥❥❥❥ ❯❯❯❯ ❥❥❥ ❯❯* A (B (C D)) ((A B) C) D ⊗ ⊗ ⊗ ⊗ ⊗O ⊗ idA αB,C,D αA,B,C idD ⊗ ⊗ αA,B⊗C,D A ((B C) D) / (A (B C)) D ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ even natural isomorphisms λX : 1 X X and ρX : X 1 X satisfying the • triangle axiom, i.e., the following diagram⊗ → is commutative for⊗ all→ objects A, B:

αA,1,B A (1 B) / (A 1) B ⊗ ⊗ ◆◆ ♣ ⊗ ⊗ ◆◆◆ ♣♣♣ ◆◆◆ ♣♣♣ idA λB ◆◆ ♣♣ρA idB ⊗ ◆' w♣♣ ⊗ A B ⊗ We note that if A and B are homogeneous objects of A then A B is homogeneous and A B = A + B . A monoidal Λ-supercategory is a Λ-supercategory⊗ equipped with a |monoidal⊗ | structure.| | | | Example 3.1. The supercategory SVec is monoidal under tensor product. (Note here that we are using all of SVec, and not just the subcategory SVec◦.) Remark 3.2. Suppose that A is a monoidal supercategory and we have homogeneous morphisms

f : A A , f : A A , f ′ : A′ A′ , f ′ : A′ A′ . 1 1 → 2 2 2 → 3 1 1 → 2 2 2 → 3 Then the diagram ′ ′ f2f1 f f ⊗ 2 1 A1 A1′ / A3 A3′ ⊗ ▲▲▲ rr8 ⊗ ▲▲▲ rrr ′ ▲ r ′ f1 f ▲▲ rrf2 f ⊗ 1 ▲& rr ⊗ 2 A A′ 2 ⊗ 2 12 STEVENVSAMANDANDREWSNOWDEN

′ f1 f commutes up to ( 1)| || 2|, that is, we have the identity − ′ f1 f (f f ′ ) (f f ′ )=( 1)| || 2|(f f f ′ f ′ ). 2 ⊗ 2 ◦ 1 ⊗ 1 − 2 1 ⊗ 2 1 This follows from the definition of the composition law in the category A ⊠ A. From this, we see that does not define a functor A A A (in general). However, it does define a ⊗ × → functor A◦ A◦ A◦, and A◦ is naturally a monoidal category. × → Let A be a monoidal Λ-supercategory admitting a Π-structure. Let A and B be objects of A. Then we have odd degree isomorphisms ξA B : A B Π(A B) and ξA idB : A B Π(A) B. We thus have a canonical even degree⊗ ⊗ isomorphism→ ⊗ Π(A B)⊗ Π(A⊗) B→. Similarly,⊗ we have a canonical even degree isomorphism Π(A B) ⊗A Π(→B). We⊗ thus ⊗ → ⊗ see that we have canonical even degree isomorphisms Π(A) = Π(1) A = A Π(1). ∼ ⊗ ∼ ⊗ Definition 3.3. Let A and B be monoidal Λ-supercategories. A monoidal Λ-superfunctor F : A B is a Λ-superfunctor F equipped with even isomorphisms Φ : F (X) F (Y ) → X,Y ⊗ → F (X Y ) and ϕ: F (1A) 1B such that the following diagrams commute for all objects X,Y,Z⊗ of A →

ΦX,Y 1 ΦX⊗Y,Z (F (X) F (Y )) F (Z) ⊗ / F (X Y ) F (Z) / F ((X Y ) Z) ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ αF (X),F (Y ),F (Z) F (αX,Y,Z )

1 ΦY,Z ΦX,Y ⊗Z F (X) (F (Y ) F (Z)) ⊗ / F (X) F (Y Z) / F (X (Y Z)) ⊗ ⊗ ⊗ ⊗ ⊗ ⊗

λF (X) ρF (X) 1 F (X) / F (X) F (X) 1 / F (X) ⊗ O ⊗ O ϕ 1 F (λ ) 1 ϕ F (ρ ) ⊗ X ⊗ X Φ1,X ΦX,1 F (1) F (X) / F (1 X) F (X) F (1) / F (X 1) ⊗ ⊗ ⊗ ⊗ We will generally suppress Φ and ϕ from the notation. Definition 3.4. Given two monoidal Λ-superfunctors (F, Φ,ϕ) and (G, Γ,γ) between Λ- supercategories A and B, a monoidal natural transformation η : F G is an even natural transformation η such that the following two diagrams commute for→ all objects X,Y of A: ηX ηY γ F (X) F (Y ) ⊗ / G(X) G(Y ) , 1 / G(1) . ⊗ ⊗ ♣♣8 ϕ ♣♣ ΦX,Y ΓX,Y ♣♣♣ ♣♣♣ η1 ηX⊗Y ♣ F (X Y ) / G(X Y ) F (1) ⊗ ⊗ We thus have a 2-category of monoidal supercategories, and in particular, a notion of monoidal equivalence. 3.2. The equivalences Σ and T. Let A and B be Z/q-graded supercategories, with q even. Define an equivalence ΣA B : A ⊠ B B ⊠ A , → as follows. On homogeneous objects, we put ΣA,B(A ⊠ B) = B ⊠ A. Given homogeneous morphisms f : A A′ and g : B B′ between objects of A and B, we define ΣA,B(f g)= f g → → ⊗ ( 1)| || |g f. − ⊗ SUPERSYMMETRIC MONOIDAL CATEGORIES 13

Now suppose that A and B admit Π-structures. We deﬁne an equivalence

TA B : A ⊠ B B ⊠ A , → A B as follows. On homogeneous objects, we put TA,B(A ⊠ B) = Π| || |(B) ⊠ A. Given homogeneous morphisms f : A A′ and g : B B′ of homogeneous objects, we put f g A B → → TA,B(f g)=( 1)| || |Π| || |(g) f. We regard⊗ Σ and− T as analogous⊗ to the two symmetries σ and τ we deﬁned on SVec in §2.1. They will give rise to the notion of symmetric and supersymmetric monoidal supercategories, just as σ and τ give rise to the notion of commutative and supercommutative superalgebras.

3.3. Factor systems. We next want to define braidings and symmetries, and their super analogs. For this, we will require certain systems of scalars that we now introduce. Definition 3.5. Let q be an even integer and let p Z/2. A B-factor system of parity p ∈ is a pair ω =(ω1,ω2) of functions 3 ω ,ω : (Z/q) k× 1 2 → satisfying the following conditions for all a,b,c,d Z/q (in what follows, the semicolon in, ∈ e.g., ω1(a; b, c), indicates that the b and c parameters play similar roles, while the a parameter is somewhat different):

(3.5a) ω1(a; b, c)ω1(a; b + c,d)= ω1(a; b, c + d)ω1(a; c,d)

(3.5b) ω2(a + b, c; d)ω2(a, b; d)= ω2(b, c; d)ω2(a, b + c; d) 1 abcdp 1 (3.5c) ω (b; c,d)ω (a; c,d)ω (a + b; c,d)− =( 1) ω (a, b; c)ω (a, b; d)ω (a, b; c + d)− 1 1 1 − 2 2 2 1 1 (3.5d) ω1(a; b, c)ω1(a; c, b)− = ω2(b, a; c)ω2(a, b; c)− .

An S-factor system of parity p is a triple ω = (ω1,ω2,ω♯) where (ω1,ω2) is a B-factor 2 system of parity p and ω : (Z/q) k× is a function satisfying the following conditions: ♯ → 1 (3.5e) ω1(a; b, c)ω2(b, c; a)= ω♯(a, b)ω♯(a, c)ω♯(a, b + c)− 1 (3.5f) ω1(c; a, b)ω2(a, b; c)= ω♯(a, c)ω♯(b, c)ω♯(a + b, c)− .

We say that ω is symmetric if ω♯(a, b)= ω♯(b, a). Remark 3.6. A few remarks: (a) The B- and S- preﬁxes stand for “braiding” and “syllepsis” (see §4 for more). (b) The parity p of the factor system only intervenes in (3.5c). (c) Let Γp be the set of all S-factor systems of parity p for Z/q. Then Γ = Γ0 Γ1 is a Z/2-graded abelian group. ⊕

Example 3.7. Let ω1, ω2, and ω♯ be the constant functions taking the value 1. Then ω is an even S-factor system. We refer to this as the trivial factor system. n Example 3.8. Suppose that q is divisible by 4. Then n 2 is a well-deﬁned function Z/q Z/2. Put 7→ → c ab a b A (a; b, c)=1, A (a, b; c)=( 1)(2) , A (a, b)=( 1)(2)(2). 1 2 − ♯ − Then A is an odd symmetric S-factor system. This is perhaps the most important example for this paper. 14 STEVENVSAMANDANDREWSNOWDEN

Example 3.9. Suppose that ζ k. Deﬁne 4 ∈ ζ4 if a, b, and c are all odd B1(a; b, c)=1, B2(a, b; c)= . (1 otherwise Then (B , B ) is an odd B-factor system. Now suppose that ζ k, and put 1 2 8 ∈ ζ8 if a and b are both odd B♯(a, b)= . (1 otherwise

Then (B1, B2, B♯) is an odd symmetric S-factor system. This factor system has the advantage over the A factor system in that it does not require q to be divisible by 4. Example 3.10. Define C (r; s, t)=( 1)rst, C (r, s; t)=( 1)rst, C (r, s)=1. 1 − 2 − ♯ Then C is an even symmetric S-factor system. Example 3.11. Define D (r; s, t)=( 1)rst, D (r, s; t)=( 1)rst, D (r, s)=( 1)rs. 1 − 2 − ♯ − Then D is an even symmetric S-factor system. 2 Definition 3.12. Let ϕ: (Z/q) k× be an arbitrary function. Put → 1 1 ω1(r; s, t)= ϕ(r, s + t)ϕ(r, s)− ϕ(r, t)− , 1 1 ω2(r, s; t)= ϕ(r + s, t)ϕ(r, s)− ϕ(s, t)− ,

ω♯(r, s)= ϕ(r, s)ϕ(s,r).

Then (ω1,ω2,ω♯) is a symmetric even S-factor system, called the coboundary of ϕ. We denote this system by ∂(ϕ). Example 3.13. Suppose 4 q. Then C is the coboundary of the function c given by | rs c(r, s)=( 1)( 2 ). − Example 3.14. Suppose q is an arbitrary even integer and ζ4 k. Then D is the coboundary of the function d given by ∈ 1 if x or y is even d(x, y)= . (ζ4 if x and y are odd We now give a somewhat more involved example of a coboundary. Proposition 3.15. Set ǫ(n)= 1 if n 3 (mod 4) and ǫ(n)=1 otherwise. − ≡ (a) Suppose that ζ k and that q is divisible by 8. Deﬁne 4 ∈ n m a m −(2) (n, m)= ǫ(n) ζ4 . Then A = B ∂(a) holds as B-factor systems. (b) Suppose that· ζ k and that q is divisible by 16. Deﬁne 16 ∈ (n)m a m − 2 nm ′(n, m)= ǫ(n) ζ4 ζ16

Then A = B ∂(a′) holds as S-factor systems. · SUPERSYMMETRIC MONOIDAL CATEGORIES 15

Proof. (a) The ﬁrst component of the equation A = B ∂(a) amounts to the identity · a (b+c) a b a c b+c −(2) b −(2) c −(2) ǫ(a) ζ4 = ǫ(a) ζ4 ǫ(a) ζ4 , which is clear. The second component amounts to

a+b c b a c ( 2 )c ab c (2)c c (2)c B (a, b; c)ǫ(a + b) ζ− =( 1)(2) ǫ(b) ζ− ǫ(a) ζ− . 2 4 − 4 4 The quotient of the left side by the right side is (c)ab c c c abc ( 1) 2 B (a, b; c)ǫ(a + b) ǫ(a) ǫ(b) ζ− , − 2 4 which now only depends on a, b, c modulo 4. We will show it is 1 in all cases. If c is even, this c (2) c ab B ab reduces to (( 1) ζ4− ) =1. If c is odd, then this simplifies to 2(a, b;1)ǫ(a+b)ǫ(a)ǫ(b)ζ4− . − 2 B b If a = 0, we get ǫ(b) = 1. If a = 1, we get 2(1, b;1)ǫ(b + 1)ǫ(b)ζ4− . If a = 2, we get b B b ǫ(b +2)ǫ(b)( 1) . Finally, if a = 3, we get 2(1, b;1)ǫ(b 1)ǫ(b)ζ4. In all cases, we see that the expressions− are 1 for all choices of b by− inspection. − (b) Since a′ is equal to a times a quantity that is bilinear, the first two components of ∂(a′) agree with those of ∂(a). In particular, the equation A = B ∂(a′) holds on the first two components by part (a). On the ♯ component, it amounts to the· following identity

b a a b a (2)a ab b (2)b ab ǫ(b) ζ− ζ ǫ(a) ζ− ζ ( 1)(2)(2) = B (a, b). 4 16 4 16 − ♯ The quotient of the left side by the right side is

b a a b a b a (2) (2) ab b ( )( ) 1 ǫ(b) ζ− − ζ ǫ(a) ( 1) 2 2 B (a, b)− . 4 8 − ♯ We will show this is always 1. Suppose that a = 2c is even (we will assume that a, b are integers). Then the above simpliﬁes to (b)c 2bc2+bc bc (b)c ( 1) 2 ζ− ζ ( 1) 2 =1. − 4 4 − Since the expression is symmetric in a and b, it remains to prove that the expression is 1 when a =2c + 1 and b =2d + 1 are odd. In that case, it simpliﬁes to

(2d2+d)a (2c2+c)b cd c+d cd 1 ǫ(b)ζ− ζ− ( 1) ζ ζ ǫ(a)( 1) ζ− 4 4 − 4 8 − 8 d ad c bc c+d = ǫ(b)( 1) ζ− ( 1) ζ− ζ ǫ(a) − 4 − 4 4 = ǫ(a)( 1)cǫ(b)( 1)d =1. − − Remark 3.16. Let Cr(G, A) denote the group of r-cochains of a group G with values in the abelian group A. Note that C0(G, A)= A is the set of functions from a single point to A. These groups form a cochain complex that computes the group cohomology of G. Let r,s r s r,s r+s C = C (Z/q, k×) C (Z/q, Z). We have a natural map i: C C (Z/q, k×) given on pure tensors by ⊗ → ′ c (gr+1,...,gr+s) i(c c′)(g ,...,g )= c(g ,...,g ) . ⊗ 1 r+s 1 r One easily sees that i is an isomorphism (e.g., by considering cochains supported at a single point). Thus C1,2 C2,1 can be identified with the set of pairs (ω ,ω ) of functions (Z/q)3 ⊕ 1 2 → k×. For such a pair to form an even B-factor system, it must be killed by the differential of the total complex (this accounts for the first three conditions in the definition of factor 3 system). This suggests that even B-factor systems are closely related to H (Z/q Z/q, k×), and that there could be a more fundamental definition including C0,3 and C3,0×. We note 16 STEVENVSAMANDANDREWSNOWDEN

that 3-cocycles of Z/q can be used to twist the associativity constraint in a monoidal Z/q- supercategory, which might be relevant. We expect that even S-factor systems are similarly related to third abelian cohomology (as deﬁned in [Mc]). It is not clear to us exactly how odd factor systems relate to cohomology. 3.4. Braidings. Let A be a monoidal Z/q-supercategory with q even (we allow q = 0). Consider an even isomorphism of superfunctors β : Σ. Explicitly, β assigns to every ⊗→⊗◦ pair (A, B) of objects an even isomorphism β : A B B A such that if f : A A′ A,B ⊗ → ⊗ → and g : B B′ are homogeneous morphisms then the diagram → βA,B A B / B A ⊗ ⊗ f g g f ⊗ ⊗ βA′,B′ A′ B′ / B′ A′ ⊗ ⊗ f g commutes up to ( 1)| || |. Consider the two hexagons (H1) and (H2): − βA,B⊗C A (B C) / (B C) A α−1 ❧❧5 ⊗ ⊗ ⊗ ⊗ ❘❘❘ α−1 A,B,C ❧❧❧ ❘❘❘ B,C,A ❧❧❧ ❘❘❘ ❧❧❧ ❘❘❘ ❧❧❧ ❘❘) (H1) (A B) C B (C A) ❘ ⊗ ⊗ ❘❘❘ ❧❧❧5 ⊗ ⊗ ❘❘❘ ❧❧❧ ❘❘❘ ❧❧❧ βA,B 1C ❘❘❘ ❧❧1B βA,C ⊗ ❘) ❧❧❧ ⊗ / (B A) C −1 B (A C) ⊗ ⊗ αB,A,C ⊗ ⊗

βA⊗B,C (A B) C / C (A B) . α ❧❧5 ⊗ ⊗ ⊗ ⊗ ❘❘❘ α A,B,C ❧❧❧ ❘❘❘ C,A,B ❧❧❧ ❘❘❘ ❧❧❧ ❘❘❘ ❧❧❧ ❘❘) (H2) A (B C) (C A) B ❘ ⊗ ⊗ ❘❘❘ ❧❧❧5 ⊗ ⊗ ❘❘❘ ❧❧❧ ❘❘❘ ❧❧❧ 1A βB,C ❘❘❘ ❧❧βA,C 1B ⊗ ❘) ❧❧❧ ⊗ A (C B) / (A C) B ⊗ ⊗ αA,C,B ⊗ ⊗ Definition 3.17. We say that β is a braiding with respect to the even B-factor system ω if in (H1) the bottom path is equal to ω ( A ; B , C ) times the top path, and in (H2) 1 | | | | | | the bottom path is equal to ω2( A , B ; C ) times the top path, whenever A, B, and C are homogeneous objects. | | | | | | Definition 3.18. We say that β is a symmetry with respect to the even S-factor system ω if it is a braiding and further satisfies the condition βB,A βA,B = ω♯( A , B )idA B for all homogeneous objects A and B. ◦ | | | | ⊗ Remark 3.19. By taking ω to be the trivial factor system (Example 3.7), one recovers the usual definition of symmetry: the diagrams (H1) and (H2) commute, and βB,AβA,B = idA B. ⊗ 2 Remark 3.20. Suppose β is a symmetric with respect to ω, and let ϕ: (Z/q) k× be → an arbitrary function. Define β′ by β′ = ϕ( X , Y )β . Then β′ is a symmetry with X,Y | | | | X,Y respect to ω ∂ϕ, where ∂(ϕ) is as in Definition 3.12. · SUPERSYMMETRIC MONOIDAL CATEGORIES 17

Remark 3.21. The reason that even factor systems are the appropriate systems of scalars will be made clear in §3.8. Deﬁnition 3.22. Let A and B be braided monoidal Z/q-supercategories. A braided monoidal Λ-superfunctor F : A B is a monoidal Λ-superfunctor (F, Φ,ϕ) such that the following diagram commutes for→ all objects X,Y of A

βF (X),F (Y ) F (X) F (Y ) / F (Y ) F (X) . ⊗ ⊗ ΦX,Y ΦY,X

F (βX,Y ) F (X Y ) / F (Y X) ⊗ ⊗ A natural transformation of braided monoidal superfunctors is simply one of monoidal superfunctors, i.e., the braiding is not used in the deﬁnition. 3.5. Type I superbraidings. Suppose A is a monoidal Z/q-supercategory, with q even (we allow q = 0), admitting a Π-structure. Consider an even isomorphism of functors β : T. Explicitly, β assigns to every pair (A, B) of homogeneous objects an even ⊗→⊗◦ A B isomorphism βA,B : A B Π| || |(B) A such that if f : A A′ and g : B B′ are nonzero homogeneous⊗ morphisms→ of homogeneous⊗ objects then the→ diagram →

β A,B A B A B / Π| || |(B) A ⊗ ⊗ f g Π|A||B|(g) f ⊗ ⊗ β ′ ′ A ,B A B A′ B′ / Π| || |(B′) A′ ⊗ ⊗ f g commutes up to ( 1)| || |. Note that the existence of nonzero f,g forces A = A′ and − | | | | B = B′ ; if f or g is 0, then the power of Π may need to change, but the resulting diagram |will| commute| | for trivial reasons. The hexagons (H1) and (H2) admit obvious modiﬁcations (H1′) and (H2′) in this setting, as follows. Let A, B, and C be homogeneous objects. Then (H1′) is the diagram

ψ1 A (B C) / (Πb+c(B) C) A . ❙ −1 ♥⊗7 ⊗ ⊗ ⊗❙❙ −1 α ♥♥♥ ❙❙❙α ♥♥ ❙❙❙ ♥♥♥ ❙❙❙ ♥♥♥ ❙❙) b+c (H1′) (A B) C Π (B) (C A) P ⊗ ⊗ PP ❦❦❦5 ⊗ ⊗ PPP ❦❦❦ PPP ❦❦❦ β id PP ❦❦ ψ2 ⊗ P' ❦❦❦ (Πb(B) A) C / Πb(B) (A C) ⊗ ⊗ α−1 ⊗ ⊗ Here b = A B , c = A C , ψ is the composition | || | | || | 1

βA,B⊗C A (B C) / Πb+c(B C) A ∼ / (Πb+c(B) C) A, ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ψ2 is the composition

id βA,C ξ ξ id Πb(B) (A C) ⊗ / Πb(B) (Πc(C) A) ⊗ ⊗ / Πb+c(B) (C A) , ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ 18 STEVENVSAMANDANDREWSNOWDEN

and we have omitted indices for readability; recall ξ = ξn,m is the isomorphism Πn Πm. → The diagram (H2′) is:

β (A B) C / Πa+b(C) (A B) ♥7 ⊗ ⊗ ⊗ ⊗❙❙❙ α ♥♥♥ ❙❙❙ α ♥♥ ❙❙❙ ♥♥♥ ❙❙❙ ♥♥♥ ❙❙) (H2′) A (B C) (Πa+b(C) A) B ⊗ ⊗ PP ❦❦5 ⊗ ⊗ PPP ❦❦❦ PPP ❦❦❦ id β PPP ❦❦❦β id ⊗ P' ❦❦❦ ⊗ A (Πb(C) B) / (A Πb(C)) B ⊗ ⊗ α ⊗ ⊗ where a = A C and b = B C , and we have again omitted indices. It is not| correct|| | to ask for| the|| | above hexagons to commute: in fact, this leads to inconsistencies (see §3.8). Instead, we make the following definition: Definition 3.23. We say that β is a type I superbraiding with respect to the odd B- factor system ω if in (H1′) the bottom path is equal to ω ( A ; B , C ) times the top path, 1 | | | | | | and in (H2′) the bottom path is equal to ω ( A , B ; C ) times the top path. 2 | | | | | | Definition 3.24. We say that β is a type I supersymmetry with respect to the odd S-factor system ω if it is a type I superbraiding with respect to ω, and for all homogeneous objects A and B, the diagram

βA,B A B / Πab(B) A ❙ ⊗ ❙❙❙ ⊗ ❙❙❙ ❙❙ β ab ❙❙❙ Π (B),A ω♯(a,b)ξ ξ ❙❙ ⊗ ❙) Πab(A) Πab(B) ⊗ commutes, where a = A and b = B . | | | | Remark 3.25. If β satisfies the conditions to be a supersymmetry then the two hexagon axioms are equivalent conditions: starting from either hexagon, invert all of the arrows and relabel the objects to get the other hexagon. Example 3.26. We define a type Ia superbraiding (respectively, supersymmetry) to be a type I superbraiding (respectively, supersymmetry) with respect to the factor system A from Example 3.8. Example 3.27. We define a type Ib superbraiding (respectively, supersymmetry) to be a type I superbraiding (respectively, supersymmetry) with respect to the factor system B from Example 3.9.

1 Remark 3.28. One can convert between type Ia and type Ib superbraidings when BA− is a coboundary, in the sense of Deﬁnition 3.12 (see Remark 3.20). Proposition 3.15 gives 1 conditions that ensure BA− is a coboundary. Deﬁnition 3.29. Let A and B be monoidal Z/q-supercategories equipped with type I superbraidings with respect to the same factor systems. A type I superbraided monoidal SUPERSYMMETRIC MONOIDAL CATEGORIES 19 superfunctor is a monoidal superfunctor F : A B such that the following diagram commutes →

β F (X),F (Y ) X Y F (X) F (Y ) / Π| || |(F (Y ) F (X)) ⊗ ⊗ |X||Y | ΦX,Y Π (ΦY,X ) F (β ) X,Y X Y X Y F (X Y ) / F (Π| || |(Y X)) Π| || |(F (Y X)) ⊗ ⊗ ⊗ where the equality is the canonical isomorphism coming from Proposition 2.2. A natural transformation of such functors is simply one of monoidal superfunctors.

3.6. Type II superbraidings. The deﬁnition of type I superbraiding followed the general procedure of replacing Σ by T, and therefore has some conceptual clarity. However, these superbraidings can be somewhat inconvenient due to the Π’s that arise; for instance, if X is an odd-degree object then a type I superbraiding does not induce an endomorphism of X X, but rather a map X X Π(X) X. Of course, one can convert an even degree ⊗ X Y ⊗ → ⊗ map X Y Π| || |(Y ) X into a map X Y Y X of degree X Y . This leads to an alternate⊗ → formulation of⊗ the concept of superbraiding⊗ → ⊗ in which there| are|| | no Π’s, but the maps may not be even. We now give the details. Let A be a monoidal Z/q-supercategory, with q even (we allow q = 0); we do not require a Π-structure. Consider an isomorphism β : Σ such that if A and B are homogeneous then β is homogeneous of degree A B .⊗→⊗◦ Explicitly, β assigns to every pair of homogeneous A,B | || | objects (A, B) an isomorphism β : A B B A of degree A B such that if f : A A′ A,B ⊗ → ⊗ | || | → and g : B B′ are homogeneous morphisms of homogeneous objects then the diagram →

βA,B A B / B A ⊗ ⊗ f g g f ⊗ ⊗ βA′,B′ A′ B′ / B′ A′ ⊗ ⊗ A B ( f + g )+ f g A B ( f + g ) commutes up to ( 1)| || | | | | | | || |. The first part of the sign, namely ( 1)| || | | | | | , comes from the condition− that β is a natural transformation, while the second− part of the f g sign, namely ( 1)| || |, comes from the definition of Σ. − Definition 3.30. We say that β is a type II superbraiding with respect to the B-factor system ω if in (H1) the bottom path is equal to ω1( A ; B , C ) times the top path, and in (H2) the bottom path is equal to ω ( A , B ; C ) times| | | the| top| | path. 2 | | | | | | Definition 3.31. We say that β is a type II supersymmetry with respect to the S-factor system ω if it is a type II superbraiding with respect to ω and for all homogeneous objects A and B we have βB,AβA,B = ω♯( A , B ) idA B | | | | · ⊗ Remark 3.32. We define type IIa/IIb superbraidings/supersymmetries analogously to the type I case (see Examples 3.26 and 3.27).

Deﬁnition 3.33. Let A and B be monoidal Z/q-supercategories equipped with type II superbraidings with respect to the same factor systems. A type II superbraided monoidal 20 STEVENVSAMANDANDREWSNOWDEN

functor is a monoidal superfunctor F : A B such that the following diagram commutes →

βF (X),F (Y ) F (X) F (Y ) / F (Y ) F (X) ⊗ ⊗ ΦX,Y ΦY,X

F (βX,Y ) F (X Y ) / F (Y X) ⊗ ⊗ for all objects X and Y . A natural transformation of such functors is simply one of monoidal superfunctors. 3.7. Equivalence between type I and type II superbraidings. Fix a monoidal Z/q- supercategory (with q even) A with a Π-structure. The following proposition shows how to convert between the two types of superbraidings. Proposition 3.34. Suppose that β is a type II superbraiding with respect to the odd B-factor system ω. For homogeneous objects A and B, deﬁne βA,B′ to be the composition

0,|A||B| β ξ idA A,B B ⊗ A B A B / B A / Π| || |(B) A. ⊗ ⊗ ⊗

Then β′ is a type I superbraiding with respect to the odd B-factor system Cω, where C is as in Example 3.10. Moreover, this construction deﬁnes a bijection between the two types of superbraidings. Now suppose that β is a type II supersymmetry with respect to the odd S-factor system ω. Then β′ is a type I supersymmetry with respect to the odd S-factor system Dω, where D is as in Example 3.11. Again, this construction is bijective. We note that C = D as B-factor systems, and so one could use D in the ﬁrst part of the proposition as well.

Proof. First we check that β′ is a natural transformation, i.e., that the following square

β′ A,B A B A B / Π| || |(B) A ⊗ ⊗ f g Π|A||B|(g) f ⊗ ′ ⊗ β ′ ′ A ,B A B A′ B′ / Π| || |(B′) A′ ⊗ ⊗ f g commutes up to ( 1)| || | for all nonzero homogeneous morphisms f : A A′ and g : B B′ between homogeneous− objects. By deﬁnition, this expands to the diagram→ →

β A,B ξ 1 A B A B / B A ⊗ / Π| || |(B) A ⊗ ⊗ ⊗ f g g f Π|A||B|(g) f ⊗ ⊗ ⊗ β ′ ′ A ,B ξ 1 A B A′ B′ / B′ A′ ⊗ / Π| || |(B′) A′ ⊗ ⊗ ⊗ A B ( f + g )+ f g Since β is a type II superbraiding, the left square commutes up to ( 1)| || | | | | | | || |. By − A B ( f + g ) deﬁnition of how Π acts on morphisms, the right square commutes up to ( 1)| || | | | | | , f g − and so the original square commutes up to ( 1)| || |, as desired. − SUPERSYMMETRIC MONOIDAL CATEGORIES 21

Now we check that the hexagon axiom (H1′) holds. Consider the following diagram (for simplicity we ignore the associators)

ω1 βA,B⊗C ξ 1 · A B + A C A B C / B C A ⊗ / Π| || | | || |(B C) A ⊗ ⊗ ❥❥❥5 ⊗ ⊗▲▲▲ ⊗ ⊗ ❥❥❥❥ ▲▲ βA,B 1 ❥❥ ▲▲ = ⊗ ❥❥❥ ▲▲ ∼ ❥❥❥❥ 1 βA,C ▲▲ ❥❥ ⊗ ▲▲ ξ 1 1 ▲▲ A B + A C B A C ⊗ ⊗ ▲▲Π| || | | || |(B) C A ⊗ ⊗ ξ ξ 1 ▲▲ O ⊗ ⊗ ⊗ ⊗ ▲▲▲ ξ 1 1 ▲▲ ⊗ ⊗ ▲▲ ∼= ▲▲ 1 βA,C 1 ξ 1 & A B ⊗ A B A B A C Π| || |(B) A C / Π| || |(B) C A ⊗ ⊗ / Π| || |(B) Π| || |(C) A ⊗ ⊗ ⊗ ⊗ ⊗ ⊗

where the ω1 in the top left arrow is ω1( A ; B , C ). On the left side, the upper triangle commutes by the ﬁrst hexagon axiom for| |β.| The| | | bottom quadrilateral commutes up to A B C A B C ( 1)| || || |. On the right side, the bottom triangle commutes up to ( 1)| || || | by the sign− rule when we compose the bottom path. The top triangle commut−es: we can take Π(X Y ) = Π(X) Y , and ξ : X Y Π(X Y ) to be ξX idY , and we use this convention⊗ for the Π(⊗B C) above.⊗ Then→ the top⊗ right vertical map⊗ is the identity map, ⊗ the bottom right vertical map is ξ ξ 1. Hence we see that (H1′) holds for β′ by following the outer paths. ⊗ ⊗ Finally, we check that the hexagon axiom (H2′) holds for β′ by considering the following diagram (again we ignore associators for simplicity)

ω2 βA⊗B,C ξ 1 1 · A C + B C A B C / C A B ⊗ ⊗ / Π| || | | || |(C) A B ⊗ ⊗ ❥❥4 ⊗ ⊗ ❲❲❲❲ O ⊗ ⊗ βA,C 1 ❥❥❥ ❲❲❲ ξ 1 1 ⊗❥❥❥ ❲❲❲⊗❲ ⊗ 1 βB,C ❥❥❥ ❲❲❲❲ ξ 1 1 ⊗ ❥❥❥❥ ❲❲❲❲❲ ⊗ ⊗ ❥❥❥ ❲+ 1 ξ 1 B C B C A C B ⊗ ⊗ / A Π| || |(C) B / Π| || |(C) A B ⊗ ⊗ ⊗ ⊗ ⊗ ⊗

where the δ in the top left arrow is ω2( A , B ; C ), and the bottom right map is βA,Π|B||C|(C) 1. The left triangle commutes by the| second| | | hexagon| | axiom for β. The middle quadrilateral⊗ A B C commutes up to ( 1)| || || | and the right triangle commutes. Hence by considering the − outer paths, we see that β′ satisﬁes the second hexagon axiom. Finally, the diagrams above can be used to show that β is a type II superbraiding if β′ is a type I superbraiding and hence the stated construction gives a bijection between type I and type II superbraidings. Now we handle the last statement, and suppose that β is a type I supersymmetry. Consider the following diagram:

β A,B ξ 1 A B A B / B A ⊗ / Π| || |(B) A ⊗ ❖❖ ⊗ ⊗ ❖❖❖ ❖❖ β β ❖❖❖ B,A Π(B),A ω♯( A , B ) ❖❖ | | | | ❖' 1 ξ A B A B ⊗ / A Π| || |(B) ⊗ ❚❚❚ ⊗ ❚❚❚❚ ❚❚❚❚ ξ 1 ξ ξ ❚❚❚❚ ⊗ ⊗ ❚❚) A B A B Π| || |(A) Π| || |(B) ⊗ 22 STEVENVSAMANDANDREWSNOWDEN

The upper left triangle commutes since β is a type II supersymmetry, and the bottom triangle A B commutes by deﬁnition of composition. The square commutes up to ( 1)| ||| | since β is a − type II superbraiding, so we conclude that β′ is a superbraiding with respect to Dω.

3.8. Explanation of factor systems. We now explain the rationale behind the deﬁnition of B-factor systems. We work with type II superbraidings here, but this discussion can be converted to type I superbraidings or ordinary braidings as well. Let A be a monoidal Z/q- supercategory with q even. We assume, for simplicity, that the monoidal structure is strict. Suppose that we have an isomorphism βX,Y : Σ such that βX,Y is homogeneous of degree X Y on homogeneous objects, and⊗→⊗◦ that (H1) and (H2) commute up to scalars. Precisely,| for|| homogeneous| objects A, B, and C of degrees a, b, and c, we suppose that the diagrams

ω1(a;b,c) βA,B⊗C A B C · / B C A ⊗ ⊗PPP ♥♥6⊗ ⊗ PPP ♥♥♥ PPP ♥♥♥ βA,B idC PP ♥♥idB βA,C ⊗ P( ♥♥ ⊗ B A C ⊗ ⊗

ω2(a,b;c) βA⊗B,C A B C · / C A B ⊗ ⊗PPP ♥♥⊗6 ⊗ PPP ♥♥♥ PPP ♥♥♥ idA βB,C PP ♥♥βA,C idB ⊗ P( ♥♥ ⊗ A C B ⊗ ⊗ commute, where ω1 and ω2 are non-zero scalars. Note that these triangles are the hexagon axioms since we have assumed that the associators are the identity. Let A, B, C, and D be four homogeneous objects of degrees a, b, c, and d. Consider the tetrahedron

ABCD ♥♥ PPP ♥♥♥ PPP v♥♥♥ PP( (3.35a) BACD / BCDA PPP ♥♥6 PPP ♥♥♥ PP( ♥♥♥ BCAD

Here we have omitted the symbols, and each arrow is the unique β map that makes sense. ⊗ This tetrahedron has four faces that each commute up to an ω1-factor. This yields the identity (3.5a). Similarly, we have the commuting tetrahedron

ABCD ♥♥ PPP ♥♥♥ PPP v♥♥♥ PP( (3.35b) ABDC / DABC PPP ♥♥6 PPP ♥♥♥ PP( ♥♥♥ ADBC that yields the identity (3.5b). SUPERSYMMETRIC MONOIDAL CATEGORIES 23

Now consider the following diagram: ABCD / CDAB ✷ ❋ < F O ✷ ❋❋ ①①☞ ✷✷ ❋❋ ①① ☞☞ ✷ ❋❋ ①① ☞ ✷ ❋❋ ①① ☞ ✷✷ ❋❋ ①① ☞☞ ✷ ❋❋ ①① ☞ ✷✷ " ① ☞ ✷ ACDB ☞☞ (3.35c) ✷✷ 6 ☞☞ ✷✷ ☞☞ ✷✷ ☞☞ ( ACBD ✷✷ ☞☞ CADB P ✷ ☞ 6 PPP ✷ ☞ ♥♥♥ PP ✷✷ ☞☞ ♥♥ PP( ☞ ♥♥♥ CABD This diagram has six triangles and one square (composed of the bottom four vertices). If a, b, c, and d are all odd then both βA,C and βB,D are odd, and the square anti-commutes; this is the one place where the speciﬁcs of supercategories intervene. Otherwise, at least one of βA,C and βB,D are even, and the square commutes. Thus, in general, the square commutes up to ( 1)abcd, and we obtain (3.5c) with p = 1. (In the case of ordinary braidings, this sign would not− be present, and so we would obtain (3.5c) with p = 0.) Finally, consider the following diagram:

BAC ❲❲ ❣3/ BCA ♥7 ❲❲❲❣❲❣❣❣ PPP ♥♥♥ ❣❣❣❣❣ ❲❲❲❲❲ PP ♥♥♥ ❣❣❣❣❣ ❲❲❲❲❲ PPP ♥♥♥ ❣❣❣❣❣ ❲❲❲❲❲ PPP ♥♥❣❣❣❣❣ ❲❲❲❲❲P' (3.35d) ABC ❣❣ ❲+ CBA P❲❲❲❲ ❣❣❣3 7 PPP ❲❲❲❲ ❣❣❣❣ ♥♥♥ PP ❲❲❲❲❲ ❣❣❣❣❣ ♥♥ PPP ❲❲❲❲❲ ❣❣❣❣❣ ♥♥♥ PPP ❲❲❲❲❲❣❣❣❣❣ ♥♥♥ ' ❣❣❣❣❣ ❲❲❲❲ ♥ ACB +/ CAB

This diagram has four triangles, two that commute up to ω1-factors and two that commute up to ω2-factors. It also has two squares that commute by the naturality of β. For example, one of the squares is

βA,BC ABC / BCA

id βB,C βB,C id ⊗ ⊗ βA,CB ACB / CBA and this commutes up to ( 1)a(b+c)bc = 1. This diagram yields (3.5d). We thus see that the four− equations in the deﬁnition of a B-factor system are necessary for consistency of the hexagon axioms. Of course, one might wonder if there are additional equations that are needed that we have overlooked. In fact, this is not the case: this is best explained by the 2-categorical considerations in §4. (The discussion there will also explain the equations for S-factor systems.) 3.9. Grothendieck groups. Let A be a monoidal Z/q-supercategory with q even. Suppose that A◦ has an exact structure and is exact. Then induces a multiplication on K(A◦) and K (A) and gives them each the structure⊗ of a superring.⊗ If the monoidal structure admits a braiding± then all three of these superrings are commutative. If the monoidal structure admits a superbraiding (of any kind) then K+(A) is commutative while K (A) is supercommutative. Thus, via K , we can view supersymmetric monoidal supercategories− as a categoriﬁcation of the notion− of a supercommutative superalgebra. 24 STEVENVSAMANDANDREWSNOWDEN

4. Some 2-category theory In this section, we explain the 2-categorical underpinnings for supersymmetries and superbraidings. We do not intend for this discussion to be perfectly rigorous, and we allow ourselves to take some shortcuts and make some simpliﬁcations. We have mostly followed [Cr] for the theory of monoidal 2-categories, and refer there for more details; see also [BN] and [KV]. 4.1. Monoidal 2-categories. Let (C , ) be a semistrict monoidal 2-category. We refer to [Cr, §2.1] for the precise deﬁnition, and⊗ make a few remarks here. We require composition of 1-morphisms in C to be strictly associative (this is part of what “2-category” means). We also require to be strictly associative (this is part of what semistrict means). We ignore the unit of the⊗ monoidal structure in our discussion. An important piece of the monoidal structure is that the square

idA g A B ⊗ / A B′ ⊗ ⊗ f idB f id ′ ⊗ ⊗ B id ′ g A ⊗ A′ B / A′ B′ ⊗ ⊗ commutes up to a given 2-isomorphism if,g; precisely,

i : (f id ′ ) (id g) ∼ (id ′ g) (f id ). f,g ⊗ B ◦ A ⊗ −→ A ⊗ ◦ ⊗ B In what follows, we omit tensor symbols (i.e., we write AB in place of A B), and simply write A for the identity 1-morphism on an object A of C . ⊗ 4.2. Braidings on monoidal 2-categories. A braiding on C consists of a pseudonatural equivalence R : AB BA together with 2-isomorphisms A,B → ρA B,C : (BRA,C ) (RA,BC) RA,BC , ρA,B C : (RA,C B) (ARB,C ) RAB,C . | ◦ → | ◦ → satisfying the ﬁve axioms below (we have inverted the convention from [Cr, §2.2] using ρ in 1 place of ρ− ). Note that the above two 2-isomorphisms make the diagrams

RA,BC RAB,C ABC / BCA ABC / CAB ❏❏ t: ❏❏ t: ❏❏ tt ❏❏ tt ❏❏ tt ❏❏ tt ❏❏ tt ❏❏ tt RA,B C ❏$ tt BRA,C ARB,C ❏$ tt RA,C B BAC ACB commute up to isomorphism, and are essentially the hexagon axioms. (Remember that the associators are identity morphisms, and can thus be omitted.) We recall that if f : A A′ → and g : B B′ are 1-morphisms, then there is a given 2-isomorphism → R : (gf)R ∼ R ′ ′ (fg); f,g A,B −→ A ,B this is part of the data of the pseudonatural equivalence R. This isomorphism makes the square

RA,B AB / BA

fg gf

RA′,B′ A′B′ / B′A′ commute up to isomorphism. SUPERSYMMETRIC MONOIDAL CATEGORIES 25

Axiom 1. The tetrahedron in (3.35a) commutes for all objects A, B, C, and D, where the edges are the obvious 1-morphisms built from R and the faces are the 2-morphisms provided by ρ. Here “commutes” is taken in the 2-categorical sense. Explicitly, this means that the diagram Bρ 1 A|C,D◦ (BCRA,D)(BRA,C D)(RA,BCD) / (BRA,CD)(RA,BCD)

1 (ρ D) ρA|B,CD ◦ A|B,C ρA|BC,D (BCRA,D)(RA,BC D) / RA,BCD commutes. Axiom 2. The tetrahedron in (3.35b) commutes. This means that the diagram

(ρ C) 1 A,B|D ◦ (RA,DBC)(ARB,DC)(ABRC,D) / (RAB,DC)(ABRC,D)

1 (Aρ ) ρAB,C|D ◦ B,C|D ρA,BC|D (RA,DBC)(ARBC,D) / RABC,D commutes. Axiom 3. The diagram in (3.35c) commutes; the 2-isomorphism in the square is given by i. We write this out explicitly using three diagrams. First, we require that the diagram

1 1 Aρ ◦ ◦ B|C,D (CRA,DB)(RA,C DB)(ACRB,D)(ARB,C D) / (CRA,DB)(RA,CDB)(ARB,CD) ϕ ρ B 1 1 ρ B 1 A|C,D ◦ ◦ A|C,D ◦ 1 Aρ , ◦ B|C,D (RA,CDB)(ACRB,D)(ARB,C D) / (RA,CDB)(ARB,CD) commutes, and deﬁne ϕ to be the indicated arrow. Second, we require that the diagram

Cρ 1 1 A,B|D◦ ◦ (CRA,DB)(CARB,D)(RA,C BD)(ARB,C D) / (CRAB,D)(RA,CBD)(ARB,C D) ψ 1 1 ρ D 1 ρ D ◦ ◦ A,B|C ◦ A,B|C Cρ 1 , A,B|D◦ (CRA,DB)(CARB,D)(RAB,C D) / (CRAB,D)(RAB,C D) commutes, and again deﬁne ψ to be the indicated arrow. Finally, we require that the diagram

1◦i◦1 (CRA,DB)(RA,C DB)(ACRB,D )(ARB,C D) / (CRA,DB)(CARB,D )(RA,C BD)(ARB,C D)

ϕ ψ

ρA,B|CD ρAB|C,D (RA,CD B)(ARB,CD ) / RAB,CD o (CRAB,D )(RAB,C D)

commutes, where i = if,g with f = RA,C and g = RB,D. Axiom 4. The diagram in (3.35d) commutes; the 2-isomorphisms in the four triangles come from ρ, while the 2-isomorphisms in the two squares come from R. Explicitly, this means that the diagram ρ 1 1 ρ B,A|C ◦ ◦ A|B,C (RBA,C )(RA,BC) o (RB,C A)(BRA,C )(RA,BC) / (RB,C A)(RA,BC )

RRA,B ,idC RidA,RB,C 1 ρ−1 ρ−1 1 ◦ A,B|C A|C,B ◦ (CRA,B)(RAB,C ) / (CRA,B)(RA,C B)(ARB,C )o (RA,CB)(ARB,C ) 26 STEVENVSAMANDANDREWSNOWDEN commutes. Axiom 5. A number of conditions regarding the unit object; see [Cr, §2.2]. 4.3. Symmetric and sylleptic structures. There are two “levels” of commutativity for monoidal 1-categories: braided and symmetric. For monoidal 2-categories, there are three levels: braided, sylleptic, and symmetric. We now recall what the latter two mean. Fix a braiding R on C . A syllepsis is a 2-isomorphism (we have inverted the convention 1 from [Cr, §4], using v in place of v− )

vA,B : RB,A RA,B idA B ◦ → ⊗ such that the following diagrams commute (as before, the dotted arrows denote the deﬁnition of the common composition in the ﬁrst and third diagrams).

1 1 ρ ◦ ◦ A|B,C (RB,AC)(BRC,A)(BRA,C)(RA,BC) / (RB,AC)(BRC,A)(RA,BC ) , ϕ ρ 1 1 ρ 1 B,C|A◦ ◦ B,C|A◦ 1 ρ ◦ A|B,C , (RBC,A)(BRA,C )(RA,BC) / (RBC,A)(RA,BC )

ϕ (RB,AC)(BRC,A)(BRA,C )(RA,BC) / (RBC,A)(RA,BC ) ,

1 BvA,C 1 vA,BC ◦ ◦ vA,B C (RB,AC)(RA,BC) / idABC

1 1 ρ ◦ ◦ A,B|C (ARC,B)(RC,AB)(RA,C B)(ARB,C ) / (ARC,B)(RC,AB)(RAB,C ) , ψ ρ 1 1 ρ 1 C|A,B ◦ ◦ C|A,B ◦ 1 ρ ◦ A,B|C , (RC,AB)(RA,C B)(ARB,C ) / (RC,AB)(RAB,C )

ψ (ARC,B)(RC,AB)(RA,C B)(ARB,C ) / (RC,AB)(RAB,C ) .

1 vA,C B 1 vAB,C ◦ ◦ AvB,C (ARC,B)(ARB,C ) / idABC There are also conditions involving the unit object that we omit (see [Cr, §4]). A syllepsis is a symmetry if 1 v = v 1 agree as isomorphisms ◦ A,B B,A ◦ R R R R . A,B B,A A,B → A,B 4.4. The 2-category of supercategories. Fix an even positive integer q. For the remainder of this section, we let C be the 2-category whose objects are Z/q-supercategories equipped with a Π-functor. The 1-morphisms of C are superfunctors, and the 2-morphisms are even natural transformations. This is a 2-category in the sense that we have been using it, as composition of 1-morphisms in C is strictly associative. In this section, we denote supercategories by A, B, . . ., and objects of supercategories by a, b, . . ., to maintain consistency with the rest of this section. Let A and B be objects of C . We let AB = A B be the supercategory A ⊠ B defined in §2.5. For objects a A and b B, we write a ⊗ b, or simply ab, instead of a ⊠ b in this section. We equip A ∈B with the∈ Π-structure Π(⊗a b) = Π(a) b. Thus A B is again ⊗ ⊗ ⊗ ⊗ SUPERSYMMETRIC MONOIDAL CATEGORIES 27 an object of C . This defines a monoidal structure on C . However, it is not semistrict, since this product is not strictly associative. We will ignore this detail, and simply treat as if it were semistrict. This is the main reason we disclaimed at the start of this section⊗ that our discussion would not be entirely rigorous. In the remainder of this section, we investigate how Σ and T (as defined in §3.2) define braidings (or syllepses or symmetries) on C . We focus on the T case since it is most relevant to this paper, but the Σ case is very similar. To maintain consistency with the rest of this section, we write R in place of T. Thus, for objects A and B of C , we have the equivalence a b RA,B : A B B A defined on objects by RA,B(a b) = Π| || |(b) a, and on morphisms ⊗ → ⊗ f g a b ⊗ ⊗ by R (f g)=( 1)| || |Π| || |(g) f. A,B ⊗ − ⊗ 4.5. Braiding. To give R the structure of a braiding, we first must define the ρ isomor- 3 phisms. Fix functions ω1,ω2 : (Z/q) k×. Let A, B, and C be objects of C , and let a, b, and c be homogeneous objects of A, B→, and C. We have

a b a c ((BR )(R C))(a b c) = Π| || |(b) Π| || |(c) a A,C A,B ⊗ ⊗ ⊗ ⊗ and a ( b + c ) R (a b c) = Π| | | | | | (b) c a. A,BC ⊗ ⊗ ⊗ ⊗ We deﬁne a b a c a ( b + c ) ρA B,C : Π| || |(b) Π| || |(c) a Π| | | | | | (b) c a | ⊗ ⊗ → ⊗ ⊗ by

a c b c ((R B)(AR ))(a b c) = Π| || |(Π| || |(c)) a b A,C B,C ⊗ ⊗ ⊗ ⊗ and c ( a + b ) R (a b c) = Π| | | | | | (c) a b. AB,C ⊗ ⊗ ⊗ ⊗ a c b c c ( a + b ) We deﬁne ρA,B C to be ω2(a, b; c) times the identity map; note that Π| || | Π| || | = Π| | | | | | . We have thus deﬁned| the necessary data for a braiding on C . We now examine◦ the axioms. Axiom 1. The diagram becomes

ω1(a;c,d)1 ξ ξ 1 a b a c a d ⊗ ⊗ ⊗ a b a ( c + d ) Π| || |(b)Π| || |(c)Π| || |(d)a / Π| || |(b)Π| | | | | | (c)da .

ω1(a;b,c)ξ ξ 1 1 ω1(a;b,cd)ξ ξ 1 1 ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ω1(a;bc,d)ξ 1 ξ 1 a ( b + c ) a d ⊗ ⊗ ⊗ a ( b + c + d ) Π| | | | | | (b)cΠ| || |(d)a / Π| | | | | | | | (b)cda

The top composition is ω1(a; b,cd)ω1(a; c,d)ξ ξ ξ 1 while the bottom composition is ω (a; b, c)ω (a; bc,d)ξ ξ ξ 1. Hence the diagram⊗ ⊗ commutes⊗ if and only if 1 1 ⊗ ⊗ ⊗ ω1(a; b, c)ω1(a; bc,d)= ω1(a; b,cd)ω1(a; c,d) which is equivalent to (3.5a) (note we have multiplicative notation as opposed to additive notation here since we are using the objects rather than their degrees as input). 28 STEVENVSAMANDANDREWSNOWDEN

Axiom 2. The diagram becomes

ω (a,b;d) a d b d c d 2 ( a + b ) d c d Π| || |(Π| || |(Π| || |(d)))abc / Π | | | | | |(Π| || |(d))abc

ω2(b,c;d) ω2(ab,c;d) ω (a,bc;d) a d ( b + c ) d 2 ( a + b + c ) d Π| || |(Π | | | | | |(d))abc / Π | | | | | | | |(d)abc where each map is a scalar times the identity. Thus, commutativity of this diagram is equivalent to the identity

ω2(ab, c; d)ω2(a, b; d)= ω2(b, c; d)ω2(a, bc; d) which is (3.5b). Axiom 3. The ﬁrst diagram comes out to be the following:

1 a c b c a d b d ○ a c b ( c + d ) a d Π| || |(Π| || |(c))Π| || |(Π| || |(d))ab / Π| || |(Π| | | | | | (c))Π| || |(d)ab . ϕ ○2 ○3 4 , a ( c + d ) b c b d ○ a ( c + d ) b ( c + d ) Π| | | | | | (Π| || |(c))Π| || |(d)ab / Π| | | | | | (Π| | | | | | (c))dab

The four numbered maps are given by

a b d ( c +1) ○1 =( 1)| || || | | | ω (b; c,d)ξ ξ 1 1 − 1 ⊗ ⊗ ⊗ ○2 = ω (a; c,d)ξ ξ 1 1 1 ⊗ ⊗ ⊗ ○3 = ω (a; c,d)ξ ξ 1 1 1 ⊗ ⊗ ⊗ a b d ( c +1) ○4 =( 1)| || || | | | ω (b; c,d)ξ ξ 1 1 − 1 ⊗ ⊗ ⊗ To compute the signs, we use the identities Πk(ξn,m)=( 1)k(n+m)ξn+k,m+k and ξn,m = − Πk ξn+k,m+k. We thus see that the square commutes, and

a b c d ϕ =( 1)| || || || |ω (b; c,d)ω (a; c,d)ξ ξ 1 1. − 1 1 ⊗ ⊗ ⊗ Note that one has to apply the sign rule when computing the compositions, and this cancels a b d the ( 1)| || || | part of the sign. The second diagram is the following: −

ω (a,b;d) a c b c a d b d 2 a c b c d ( a + b ) Π| || |(Π| || |(c))Π| || |(Π| || |(d))ab / Π| || |(Π| || |(c))Π| | | | | | (d)ab ψ ω2(a,b;c) ω2(a,b;c) ω (a,b;d) , c ( a + b ) a d b d 2 c ( a + b ) d ( a + b ) Π| | | | | | (c)Π| || |(Π| || |(d))ab / Π| | | | | | (c)Π| | | | | | (d)ab

All maps are scalar multiples of the identity, so the diagram commutes and

ψ = ω (a, b; d)ω (a, b; c) id. 2 2 · SUPERSYMMETRIC MONOIDAL CATEGORIES 29

Finally, we have the third diagram. It is

id Π|a||c|(Π|b||c|(c))Π|a||d|(Π|b||d|(d))ab / Π|a||c|(Π|b||c|(c))Π|a||d|(Π|b||d|(d))ab

ϕ ψ Π|a|(|c|+|d|)(Π|b|(|c|+|d|)(c))dab Π|c|(|a|+|b|)(c)Π|d|(|a|+|b|)(d)ab ❱❱❱ ❤❤ ❱❱❱ω❱2(a,b;cd)id ω1(ab;c,d)ξξ11❤❤❤❤ ❱❱❱❱ ❤❤❤❤ ❱❱❱❱ ❤❤❤❤ ❱❱❱* t❤❤❤❤ Π(|a|+|b|)(|c|+|d|)(c)dab (We have omitted symbols in the bottom right map.) This diagram commutes if and only if (3.5c) holds. ⊗ Axiom 4. The diagram becomes

ω (b,a;c) c ( a + b ) a b 2 b c a c a b b c a ( b + c ) Π| | | | | | (c)Π| || |(b)a o Π| || |(Π| || |(c))Π| || |(b)a / Π| || |(c)Π| | | | | | (b)a

|a||b||c| ( 1) ξc ξ 1 − ⊗ ⊗ ω (a,b;c)−1 ( a + b ) c a b 2 a c b c a b a ( b + c ) b c Π | | | | | |(c)Π| || |(b)a / Π| || |(Π| || |(c))Π| || |(b)a o Π| | | | | | (Π| || |(c))ba

a c ( b +1) where the top right map is ( 1)| || | | | ω1(a; b, c)ξc ξ 1 and the bottom right map is a b ( c +1) 1 1 − 1 ⊗ ⊗ ( 1)| || | | | ω1(a; c, b)− ξc− ξ− 1. Taking into account signs from composing tensor − ⊗ ⊗ 1 products of maps, the right side composition is ω1(a; b, c)ω1(a; c, b)− times the identity, and hence the commutativity of this diagram is equivalent to (3.5d). We thus see that R and ρ define a braiding if and only if (ω1,ω2) is a B-factor system. In particular, we obtain braidings using the systems defined in Examples 3.26 and 3.27. 4.6. Symmetry. Now we attempt to define a syllepsis on R. Fix a function ω : (Z/q)2 ♯ → k×. We have a b a b R (R (a b))=Π| || |(a) Π| || |(b). B,A A,B ⊗ ⊗ We define vA,B : RB,ARA,B id to be the isomorphism ω♯(a, b)ξ ξ on an object a b. We check the commutativity of→ the four diagrams in order. The first⊗ is ⊗

ω1(a;b,c)1 ξ ξ a b ( a c ) a b a c ⊗ ⊗ a b a c a ( b + c ) Π| || |(Π | || | (a))Π| || |(b)Π| || |(c) / Π| || |(Π| || |(a))Π| | | | | | (b)c . ϕ ω2(b,c;a)id ω2(b,c;a)id

ω1(a;b,c)1 ξ ξ , a ( b + c ) a b a c ⊗ ⊗ a ( b + c ) a ( b + c ) Π| | | | | | (a)Π| || |(b)Π| || |(c) / Π| | | | | | (a)Π| | | | | | (b)c This formally commutes, and so ϕ = ω (a; b, c)ω (b, c; a)1 ξ ξ. 1 2 ⊗ ⊗ The second diagram is

a b a c a b a c ϕ a ( b + c ) a ( b + c ) Π| || |(Π| || |(a))Π| || |(b)Π| || |(c) / Π| | | | | | (a)Π| | | | | | (b)c .

|a||b||c| ( 1) ω♯(a,c)ξ 1 ξ ω (a,bc)ξ ξ 1 − ⊗ ⊗ ♯ ⊗ ⊗ ω (a,b)ξ ξ 1 a b a b ♯ ⊗ ⊗ Π| || |(a)Π| || |(b)c / abc This diagram commutes if and only if

ω1(a; b, c)ω2(b, c; a)ω♯(a, bc)= ω♯(a, b)ω♯(a, c), 30 STEVENVSAMANDANDREWSNOWDEN

which is equivalent to (3.5e). The third diagram is

ω (a,b;c)id a c b c a c b c 2 a c b c ( a + b ) c Π| || |(a)Π| || |(b)Π| || |(Π| || |(c)) / Π| || |(a)Π| || |(b)Π | | | | | |(c) . ψ ω1(c;a,b)ξ ξ 1 ω1(c;a,b)ξ ξ 1 ⊗ ⊗ ⊗ ⊗ ω (a,b;c)id , ( a + b ) c a c b c 2 ( a + b ) c ( a + b ) c Π | | | | | |(a)bΠ| || |(Π| || |(c)) / Π | | | | | |(a)bΠ | | | | | |(c) This formally commutes, and we ﬁnd that ψ = ω (c; a, b)ω (a, b; c)ξ ξ 1. 1 2 ⊗ ⊗ The fourth diagram is

a c b c a c b c ψ ( a + b ) c ( a + b ) c Π| || |(a)Π| || |(b)Π| || |(Π| || |(c)) / Π | | | | | |(a)bΠ | | | | | |(c) .

ω♯(a,c)ξ 1 ξ ω (ab,c)ξ 1 ξ ⊗ ⊗ ♯ ⊗ ⊗ ω (b,c)1 ξ ξ b c b c ♯ ⊗ ⊗ aΠ| || |(b)Π| || |(c) / abc This diagram commutes if and only if

ω1(c; a, b)ω2(a, b; c)ω♯(ab, c)= ω♯(a, c)ω♯(b, c), which is equivalent to (3.5f). Finally, the two isomorphisms 2 a b a b a b (R R R )(ab) = Π | || |(b)Π| || (a) Π| || |(b)a = R (ab) A,B B,A A,B → A,B are 1 v = ω (a, b)ξ ξ, v 1= ω (b, a)ξ ξ. ◦ A,B ♯ ⊗ B,A ◦ ♯ ⊗ These agree if and only if ω♯(a, b)= ω♯(b, a), i.e., γ is symmetric. We thus see that v deﬁnes a syllepsis if and only if (ω1,ω2,ω♯) is an S-factor system, and it deﬁnes a symmetry if and only if additionally ω♯ is symmetric. We thus obtain symmetries from the systems in Examples 3.26 and 3.27. Remark 4.1. A variant of the above discussion shows that even factor systems can be used to endow Σ with a braided, sylleptic, or symmetric structure.

5. Spin symmetric groups 5.1. S-sets and s-groups. An s-set is a triple (X, ,c) where X is a set, : X Z/2 is a function (the degree function), and c is an involution|·| of X such that cx|·|= x →for all | | | | x X. We write Xi for the set of elements of parity i. ∈Let X and Y be s-sets. A homogeneous map X Y of degree n is a function that → maps Xi into Yi+n and is compatible with c. We let Hom(X,Y )i denote the set of maps of degree i, and let Hom(X,Y ) = Hom(X,Y ) Hom(X,Y ) , which is naturally an s-set. We 0 ∐ 1 thus have a category SSet of s-sets. We let SSet◦ be the subcategory where the morphisms are even (degree 0). Let X and Y be s-sets. We deﬁne the product of X and Y , denoted X Y , as follows. The underlying set is the quotient of the usual cartesian product X Y by the×× involution c c. That is, elements of X Y are represented by ordered pairs (x, y),× and the pairs (x, y) and× (cx, cy) represent the same×× element. The degree function is given by (x, y) = x + y . The involution c is given by c(x, y)=(cx,y)=(x,cy). We deﬁne an isomorphism| | X | Y| | |Y X ×× → ×× SUPERSYMMETRIC MONOIDAL CATEGORIES 31

x y by (x, y) c| || |(y, x). The product gives SSet◦ the structure of a symmetric monoidal category. 7→ ×× An s-group is a group object in the monoidal category SSet◦. Explicitly, an s-group is a triple (G, ,c) where G is a group, : G Z/2 is a group homomorphism, and c G is a central element|·| satisfying c2 = 1 and|·| c =0.→ A homomorphism of s-groups f : G∈ H is a group homomorphism such that f(c|)=| c and f(g) = g for all g G. → If G and H are s-groups then G H is naturally an| s-group,| | | with multiplication∈ deﬁned by h g′ ×× (g, h)(g′, h′)= c| || |(gg′, hh′). Two elements g and h of an s-group G are said to supercom- h g mute if gh = c| || |hg. Similarly, two subsets A and B of G are said to supercommute if every element of A supercommutes with every element of B. If H1 and H2 are supercommuting s-subgroups of G then there is a homomorphism H H G given by (h , h ) h h . 1×× 2 → 1 2 7→ 1 2 Example 5.1. Let X be an object in a supercategory, and let Auth(X) be the group of all homogeneous automorphisms of X. Then Auth(X) is an s-group, taking c = 1 and the − h degree map to be the usual degree map. For a super vector space V , we write GL (V ) in place of Auth(V ).

5.2. Spin symmetric groups. Recall that the braid group Bn is the group generated by elements s1,...,sn 1 modulo the following two types of relations: − the braid relations s s s = s s s for 1 i n 2; and • i i+1 i i+1 i i+1 ≤ ≤ − the commutativity relations sisj = sjsi for i j 2. • | − | ≥ 2 The symmetric group Sn is isomorphic to the quotient of Bn by the relations si = 1; the isomorphism takes si to the transposition (i i + 1). δ For δ Z/2, we define Bn to be the group generated by elementss ˜1,..., sñ 1,c modulo the following∈ relations: − the element c is centrale and c2 = 1; • the braid relationss ˜is˜i+1s˜i =s ˜i+1s˜is˜i+1 for 1 i n 2; and • the (super)commutativity relationss ˜ s˜ = cδs˜≤s˜ for≤ i− j 2. • i j j i | − | ≥ (δ,ǫ) δ 2 ǫ For ǫ Z/2, define Sn to be the quotient of Bn by the relationss ˜i = c . It is well-known ∈ (δ,ǫ) that c = 1 in Sn (for δ = 1 and ǫ = 0, this follows from Proposition 8.1, for instance), and 6 δ e e thus c = 1 in Bn as well. We therefore have central extensions 6 e 1 c Bδ B 1. e →h i→ n → n → and 1 c Se(δ,ǫ) S 1 →h i→ n → n → where c is cyclic of order two. h i (δ,ǫ) e (δ,ǫ) ℓ We define the degree function on Sn to be the composition Sn Sn Z/2, where ℓ is the length modulo 2 of a permutation, i.e., 0 for even permutations→ a→nd 1 for odd (δ,ǫ) (1,0) permutations. In this way, Sn hase the structure of an s-group.e We let Sn = Sn , and refer to this as the spin-symmetric group. e (δ,ǫ) e e Remark 5.2. For n = 0, 1, the central extension Sn is trivial for all ǫ and δ, while for n = 2, 3 it is independent of δ. For n 4, the four central extensions are non-isomorphic, 2 ≥ 2 and the group H (Sn, Z/2) of central extensions is isomorphice to (Z/2) via (δ, ǫ). 2 The calculation of H (Sn, Z/2) can be done as follows. First, H1(Sn, Z) is the abelian- ization of S , which is just Z/2 for n 2 and trivial otherwise. Next, H (S , Z) is dual n ≥ 2 n 32 STEVENVSAMANDANDREWSNOWDEN

2 to H (Sn, C×), which is Z/2 for n 4 and trivial otherwise [HH, Theorems 2.7, 2.9]. Our earlier claim then follows from the universal≥ coefficients theorem. 5.3. The τ˜ elements. Let τ S be the permutation defined by n,m ∈ n+m i + m if 1 i n τ (i)= ≤ ≤ . n,m i n if n +1 i n + m ( − ≤ ≤ In a sense, τn,m induces the natural bijection of sets [n] [m] [m] [n]. Let σn,i Sn+m be the (n + 1)-cycle (i i +1 i + n). Then we have∐ → ∐ ∈ ··· σn,i = sisi+1 si+n 1, τn,m = σn,m σn,1. ··· − ··· We define σñ,i =s ˜is˜i+1 s˜i+n 1, τñ,m =σ ñ,m σñ,1, ··· − ··· as elements of Sn+m. Theτ ñ,m elements are very important, and are essentially responsible for all examples of supersymmetric monoidal structures that we write down. e N n m Proposition 5.3. We have τñ,mτ˜m,n = c with N = 2 2 . Proof. For any i, j, k, we have σ˜i,kσ˜j,i+k 1 = (˜sks˜k+1 s˜i+k 1)(˜si+k 1s˜i+k s˜i+j+k 2) − ··· − − ··· − (i 1)(j 1) = c − − σ˜j 1,i+kσ˜i 1,k − − where we have canceled thes ˜i+k 1 and commuted the resultingσ ˜i 1,k pastσ ˜j 1,i+k. By applying this relation successively,− this implies that − − (i)(j 1) σ˜i,kσ˜j,i+k 1σ˜j,i+k 2 σ˜j,i = c 2 − σ˜j 1,i+kσ˜j 1,i+k 1 σ˜j 1,i+1. − − ··· − − − ··· − Finally, we successively use this relation to get

τñ,mτ˜m,n = (˜σn,mσñ,m 1 σñ,1)(˜σm,nσ˜m,n 1 σ˜m,1) − ··· − ··· (m 1)(n) = c − 2 (˜σn,mσñ,m 1 σñ,2)(˜σm 1,n+1σ˜m 1,n σ˜m 1,1) − ··· − − ··· − . . m n = c( 2 )(2). There is a natural injective homomorphism of s-groups (5.4) j : S S S n,m n×× m → n+m defined by (˜si, 1) s˜i and (1, s˜i) sñ+i. (To see that this map is injective, note that it is injective modulo 7→c , and so the kernel7→ ise containede ine c , but c itself is not in the kernel.) h i h i Proposition 5.5. For g S and h S , we have ∈ n ∈ m 1 nm g +nm h + g h τñ,mjn,m(g, h)˜τn,m− = c | | | | | || |jm,n(h, g). e e Proof. First consider the case h = 1, m = 1, and g =s ˜i where 1 i n 1. Then τ˜ =σ ˜ =s ˜ s˜ s˜ and hence ≤ ≤ − n,1 n,1 1 2 ··· n 1 τñ,1s˜iτñ,−1 =s ˜1s˜2 sñs˜isñsñ 1 s˜1 ··· − ··· n i 1 = c − − s˜ s˜ (˜s s˜ s˜ ) s˜ 1 2 ··· i+1 i i+1 ··· 1 n 2 n = c − s˜i+1 = c s˜i+1 SUPERSYMMETRIC MONOIDAL CATEGORIES 33 where in the second equality, we used the commuting relationss ˜ s˜ = cs˜ s˜ for j = n, n j i i j − 1,...,i+2, and in the third equality, we used the braid relations ˜i+1s˜is˜i+1 =s ˜is˜i+1s˜i and then again the commuting relations for j = i 1, i 2,..., 1. Since conjugation is a group homo- − − 1 n g morphism, it follows immediately thatτ ñ,1jn,m(g, 1)˜τn,−1 = c | |jm,n(1,g) for any g Sn and 1 nm g 1 nm h ∈ henceτ ñ,mjn,m(g, 1)˜τn,m− = c | |jm,n(1,g). Similarly,τ ñ,mjn,m(1, h)˜τn,m− = c | |jm,n(h, 1). Finally, we have e 1 1 τñ,mjn,m(g, h)˜τn,m− =τ ñ,mjn,m(g, 1)jn,m(1, h)˜τn,m− nm g +nm h = c | | | |jm,n(1,g)jm,n(h, 1) nm g +nm h + g h = c | | | | | || |jm,n(h, g).

Proposition 5.6. For any m, n, p, we have jn,m+p(1, τ˜m,p)jn+m,p(˜τm,n, 1)=τ ˜m,n+p.

Proof. By definition, we have jn,m+p(1, σ˜m,i)=˜σm,n+i, so j (1, τ˜ )j (˜τ , 1) = j (1, σ˜ σ˜ )j (˜σ σ , 1) n,m+p m,p n+m,p m,n n,m+p m,p ··· m,1 n+m,p m,n ··· m,1 =σ ˜ σ˜ σ˜ σ˜ =τ ˜ . m,n+p ··· m,n+1 m,n ··· m,1 m,n+p e 5.4. The category S˜. We define a homogeneous Z-supercategory S˜ as follows. There is a single homogeneous object [n] of degree n, for each n 0, and no homogeneous objects of ≥ negative degree. We put End˜([n]) = k[S ]/(c + 1) and Hom˜([n], [m]) = 0 for n = m. We S n S 6 now define a monoidal structure on S˜. On objects, we define [n] [m] = [n + m]. We define ⊗ e ⊗

: End˜([n]) End˜([m]) End˜([n + m]) ⊗ S ⊗ S → S to be the linear map induced by jn,m. The unit object is [0]. The monoidal structure is strict (e.g., the associator maps are the identities). One readily verifies that this satisfies the requisite axioms for a monoidal Z-supercategory. For n, m N, define β :[n] [m] [m] [n]tobeτ ˜ . ∈ n,m ⊗ → ⊗ n,m Proposition 5.7. β defines a type IIa supersymmetry on S˜. Proof. Proposition 5.5 exactly shows that β is a natural transformation (see the first paragraph of §3.6). Proposition 5.6 exactly shows that the hexagon (H1) commutes. Propo- k n m sition 5.3 shows that β is symmetric, i.e., βm,nβn,m = ( 1) with k = 2 2 , which also implies the second hexagon axiom. This completes the proof.− Now suppose that ζ16 k. For n Z, define ǫ(n) = 1 if n 3 (mod 4) and ǫ(n)=1 ∈ ∈ − ≡ n m m (2) nm otherwise. Define β′ :[n + m] [n + m] to be the map ǫ(n) ζ− ζ τ˜ . n,m → 4 16 n,m

Proposition 5.8. β′ defines a type IIb supersymmetry on S˜. Proof. This follows from Proposition 3.15. Let q be a positive integer. We can then regard S˜ as Z/q-graded, and the monoidal structure carries over. If q is divisible by 4 then β defines a type IIa supersymmetry, while if q is any even integer then β′ defines a type IIb supersymmetry (the key point here is that the scalar βn,m′ βm,n′ only depends on the parities of n and m; this is the purpose of the somewhat complicated scalar factor in the definition of β′). 34 STEVENVSAMANDANDREWSNOWDEN

Remark 5.9. In the category S˜, the object [n] admits an odd degree automorphism if and only if n 2. Thus S˜ does not admit a Π-structure, since the objects [0] and [1] do not. ≥ Remark 5.10. In our discussion of supersymmetric monoidal categories, we have worked exclusively with linear categories since the examples we care about are linear. However, this is not necessary. One can define a “non-linear supercategory” to be a category enriched in SSet◦, and then go on to define non-linear analogs of many of the other notions we have discussed. The category S˜ defined above is in fact the linearization of what we would call the groupoid of spin-sets, which is likely an object of fundamental importance. 5.5. Actions of spin-symmetric groups. Let A be a type IIa supersymmetric monoidal Z/q-supercategory with 4 q. In what follows, we assume that A is strict (i.e., the associativity isomorphisms are| identity maps) for simplicity. (In fact, this comes at no loss of generality due to a version of Mac Lane’s coherence theorem in this setting; see the discussion in [BE] following Definition 1.4.) Let X be a homogeneous object of degree p Z/q n n ∈ and fix a positive integer n. For 1 i n 1, define σ : X⊗ X⊗ to be the map ≤ ≤ − i → n (i 1) 2 (n i 1) id βX,X id (i 1) 2 (n i 1) n X⊗ = X⊗ − X⊗ X⊗ − − ⊗ ⊗ X⊗ − X⊗ X⊗ − − = X⊗ ⊗ ⊗ −−−−−−−→ ⊗ ⊗ 2 Note that σi has degree p (since p = p modulo 2). We then have: Proposition 5.11. The automorphisms σ satisfy the following two relations: p (a) σiσj =( 1) σjσi for i j > 1. − p | − | (b) σiσi+1σi =( 1) σi+1σiσi+1. p − (c) σ2 =( 1)(2). i − Proof. Relation (a) follows immediately from how composition works in product categories and the fact that each σi has degree p. We now verify (b). It suffices to treat the case n = 3. We must show that the following diagram commutes up to ( 1)p: − σ2 σ1 X X X / X X X / X X X ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ σ1 σ2

σ2 σ1 X X X / X X X / X X X ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ By the first hexagon axiom, σ1σ2 = βX X,X . It thus suffices to show that the following diagram commutes up to ( 1)p: ⊗ − βX⊗X,X X X X / X X X ⊗ ⊗ ⊗ ⊗ βX,X idX idX βX,X ⊗ ⊗ βX⊗X,X X X X / X X X ⊗ ⊗ ⊗ ⊗ This follows from the definition of β (see §3.6), and the fact that X = βX,X = p and id = 0. Finally, (c) follows from the definition of supersymmetry. | | | | | X | We introduce the following notation: 0 (0,0) 1 (1,0) 2 (0,1) 3 (1,1) Sn = Sn , Sn = Sn , Sn = Sn , Sn = Sn . p p The index p in Sn is regarded as an element of Z/4. The above proposition shows that Sn n e e e e e ip e e e p acts on X⊗ if we let the generators ˜ act by ( 1) and c act by ( 1) . In fact: i − − e e SUPERSYMMETRIC MONOIDAL CATEGORIES 35

p A A Proposition 5.12. The group Sn naturally acts on the functor p pn given by X n → 7→ X⊗ . e Proof. We just need to verify naturality. Given a homogeneous morphism f : X Y be- n n n → tween homogeneous objects, we get a morphism f ⊗ : X⊗ Y ⊗ , and we check that it → supercommutes with the action of σi. We have

n i 2 (i 2) σ f ⊗ = (1 β 1)(f ⊗ f ⊗ f ⊗ − ) i ⊗ Y,Y ⊗ ⊗ ⊗ i f Y i 2 (i 2) =( 1) | || |(f ⊗ β f ⊗ f ⊗ − ) − ⊗ Y,Y ⊗ i f Y + f i 2 (i 2) =( 1) | || | | |(f ⊗ f ⊗ β f ⊗ − ) − ⊗ X,X ⊗ f i 2 (i 2) =( 1)| |(f ⊗ f ⊗ f ⊗ − )(1 β 1) − ⊗ ⊗ ⊗ X,X ⊗ f n =( 1)| |f ⊗ σ . − i Remark 5.13. There are similar results to the above in the superbraiding case and the type IIb case.

5.6. Universality of S˜. Fix an integer q divisible by 4 (we allow q = 0). In this section, we regard the category S˜ as a type IIa supersymmetric monoidal Z/q-supercategory via the supersymmetry β deﬁned in §5.4. The following theorem shows that S˜ is the universal supersymmetric monoidal Z/q-supercategory equipped with an object of degree 1:

Theorem 5.14. Let A be a type IIa supersymmetric monoidal Z/q-supercategory. Let B be the category of supersymmetric monoidal Z/q-superfunctors S˜ A. Then the functor → Φ: B A◦ given by Φ(F )= F ([1]) is an equivalence of k-linear categories. → 1

Proof. Suppose that X is a degree 1 object of A. We have seen that Sn naturally acts on n n X⊗ , and so we can build a functor Ψ : S˜ A by Ψ ([n]) = X⊗ , which is easily seen X → X to be a supersymmetric monoidal Z/q-functor. The assignment X eΨX is functorial by 7→ Corollary 5.12. The resulting functor Ψ: A◦ B is easily seen to be quasi-inverse to Φ. 1 → Remark 5.15. Given p Z/q, one can build a supersymmetric monoidal Z/q-supercategory ∈ S˜p using the Sp groups, specializing to S˜ when p = 1. The category S˜p is the universal supersymmetric monoidal Z/q-category with an object of degree p. e 5.7. Type IIa′ supersymmetries. Let A be a type IIa monoidal Z/4-supercategory. We p n have seen that Sn acts on X⊗ when X is homogeneous of degree p Z/4. This may seem 2 ∈ strange since it identifies H (Sn, Z/2) with Z/4 (as sets). In fact, this is due to a choice we made, namely, the choice of factor system. To make this clear, define a type IIa′ supersymmetry to be a type II supersymmetry with respect to ACD, where A, C, and D are as in Examples 3.8, 3.10, and 3.11. Suppose A is a type IIa′ supersymmetric monoidal Z/4-category and X is an object of degree p Z/4. p n ∈ Then one can show that Sn− acts on X⊗ . We note that if ζ4 k then one can pass between type IIa and type IIa′ superbraidings: ∈ X Y indeed, if β is a type IIa superbraidinge then putting βX,Y′ = ζ4| || |βX,Y , one finds that β′ is a type IIa′ superbraiding. (The ζ4 factor in this expression realizes CD as a coboundary.) 36 STEVENVSAMANDANDREWSNOWDEN

6. Linear spin-species 6.1. S-representations. Let G be an s-group. An s-representation of G on a superspace V is a homomorphism of s-groups ρ: G GLh(V ). More explicitly, an s-representation ρ associates to each g G an automorphism→ ρ(g) of V such that ∈ (a) ρ(c)= 1, (b) ρ(g) is− homogeneous of degree deg(g), and (c) ρ(gh)= ρ(g)ρ(h). Note that an s-representation is simply a representation of G on V in the usual sense satisfying certain conditions. A morphism of s-representations f : V W of parity p is a p g → map of super vector spaces of parity p such that f(gv)=( 1) | |gf(v) for all v V . We thus have an SVec-category Rep(G) of representations of G−, which is abelian and∈ admits a Π-structure (as we see below). Let G and H be s-groups and let V and W be s-representations of them. We deﬁne an s-representation of G H on V W by the formula ×× ⊗ v h (g, h)(v w)=( 1)| || |gv hw. ⊗ − ⊗ Let j : G H H G be the usual isomorphism of s-groups, and regard W V as an s-representation×× → of G×× H via j; explicitly, ⊗ ×× g h + w g (g, h)(w v)=( 1)| || | | || |hw gv. ⊗ − ⊗ Then the usual isomorphism τ : V W W V (as deﬁned in §2.1) is an isomorphism of G H s-representations. ⊗ → ⊗ ×× Remark 6.1. Note that when V and W are two s-representations of G there is no natural s-representation of G on V W , due to the lack of a diagonal morphism G G G. However, one can regard V ⊗W as a representation of the ordinary group G (or even→ G/××c ) by g(v w) = gv gw. Note⊗ that there is no sign here, even if g and v are both of oddh i degree.⊗ This action⊗ of G/ c is by even transformations. h i 6.2. Π-structures. Let V be a super vector space, and let π k[1] be the element 1 k. There are (at least) three natural choices of Π-structure on V :∈ ∈ Let Π (V )= V [1] and let ξ : V Π (V ) be the map v v. • n n → n 7→ Let Πℓ(V )= k[1] V , and let ξℓ : V Πℓ(V ) be the map v π v. • ⊗ → 7→ ⊗ v Let Π (V )= V k[1], and let ξ : V Π (V ) be the map v ( 1)| |v π. • r ⊗ r → r 7→ − ⊗ We refer to Πℓ and Πr as the “left” and “right” Π-structures, and Πn as the “neutral” Π- structure (since it does not favor left or right). The neutral Π-structure may seem to be the most natural choice, but there is a practical issue with it: if v is an element of V then it is also an element of V [1], and thus the symbol v becomes ambiguous. The reason for the sign in the right Π-structure is so that the following diagram commutes: V V k ⊗ ξr id ξr ⊗ Π (V ) V Π (k) r ⊗ r Here ξr : k Πr(k)= k[1] is the map 1 π. Since this is an odd map, we have (id ξr)(v v→ 7→ ⊗ ⊗ 1) = ( 1)| |v π. The sign in ξr also ensures that the canonical even degree isomorphism Π (V )− Π (V⊗) is just the symmetry τ of the tensor product. ℓ → r SUPERSYMMETRIC MONOIDAL CATEGORIES 37

Suppose that V is an s-representation of an s-group G. We define s-representations of g G on Π (V ) = k[1] V and Π (V ) = V k[1] by g(π v)=( 1)| |π gv and g(v ℓ ⊗ r ⊗ ⊗ − ⊗ ⊗ π) = gv π. One readily verifies that the maps ξr : V Πr(V ) and ξℓ : V Πℓ(V ) are odd degree⊗ isomorphisms of s-representations. Furthermor→ e, the canonical even→ degree isomorphism Π (V ) Π (V ) is one of s-representations. (We do not bother defining Π for ℓ → r n s-representations.) We thus have two Π-structures for s-representations: Πr and Πℓ.

6.3. Induction. Let H G be s-groups and let V be an s-representation of H. Consider ⊆ G the induced representation W = IndH (V ) = k[G] k[H] V . We give W the structure of a super vector space by declaring g v to be homogeneous⊗ of degree g + v whenever v V is homogeneous. In this way, W is⊗ an s-representation of G. As usual,| | induction| | is the∈ left adjoint of restriction. Consider the following situation: H and H are s-subgroups of G. • 1 2 V1 (resp. V2) is an s-representation of H1 (resp. H2). • 1 τ is an element of G such that τH1τ − = H2. • τ τ i: V1 V2 is an even-degree isomorphism of s-representations of H1, where V2 is the • → 1 s-representation of H1 on V2 deﬁned by h v = (τhτ − )v. In other words, we have 1 · i(hv)=(τhτ − )i(v) for h H . ∈ 1 Deﬁne G τ G 1 τ ϕ: Ind (V ) Π| | Ind (V ), ϕ(g v)= gτ − i(v) π| | H1 1 → r H2 2 ⊗ ⊗ ⊗ We then have: Proposition 6.2. The map ϕ is an even degree isomorphism of s-representations of G.

Proof. We ﬁrst check that ϕ is well-deﬁned, i.e., that ϕ(gh v) = ϕ(g hv) for h H1. Omitting the factors of π for space, we have ⊗ ⊗ ∈

1 1 1 1 ϕ(gh v)= ghτ − i(v)= gτ − τhτ − i(v)= gτ − i(hv)= ϕ(g hv). ⊗ ⊗ ⊗ ⊗ ⊗ Thus ϕ is well-deﬁned. For g′ G we have ∈ 1 τ g′ϕ(g v)= g′(gτ − i(v) π| |)= ϕ(g′g v), ⊗ ⊗ ⊗ ⊗ 1 and so ϕ is a map of s-representations. Finally, interchanging the roles of τ and τ − gives a map in the opposite direction that is the inverse of ϕ. Thus ϕ is an isomorphism.

6.4. The category of linear spin-species. A representation of S is a sequence (Vn)n 0, ∗ ≥ where Vn is an s-representation of the s-group Sn over k. A morphism of S -representations ∗ V W is a sequence f = (fn)n 0 where fn : Vn Wn is a (not necessarilye homogeneous) → ≥ → morphism of Sn-representations. We say thate an S -representation V is evene (resp. odd) if ∗ Vn = 0 for all odd (resp. even) values of n. This gives Rep(S ) the structure of a supercate- ∗ gory. For an integere q, we can regard Rep(S ) to bee an (inhomogeneous) Z/q-supercategory ∗ by declaring an object (Vn)n 0 to have degree d Z/q ifeVn = 0 for all n d (mod q). ≥ ∈ 6≡ Every object decomposes as a direct sume of homogeneous objects, and this decomposition is ﬁnite if q = 0. 6 38 STEVENVSAMANDANDREWSNOWDEN

6.5. Monoidal structure. Let n and m be non-negative integers. Suppose that V and W are two representations of S . We define V W to be the representation of S given by ∗ ∗ ⊗ e Sn (V W )n = Ind e e (Vi Wj), e ⊗ Si Sj ⊗ e i+j=n ×× M where here S S is regarded as a subgroup of S via the map j defined in (5.4). Given i×× j n n,m a third S -representation U, we have a natural identification ∗ e e ee Sn ((U V ) W )n = Ind e e e (Ui Vj Wk), e ⊗ ⊗ Si Sj Sk ⊗ ⊗ i+j+k=n ×× ×× M and similarly for U (V W ). This yields a natural associativity isomorphism (U V ) W ⊗ ⊗ ⊗ ⊗ → U (V W ) that satisfies the pentagon axiom. The S -representation that is k in degree 0 ⊗ ⊗ ∗ and 0 in other degrees is naturally a unit object for . Thus Rep(S ) has the structure of a monoidal Z/q-supercategory. ⊗e ∗ e 6.6. Supersymmetry. In this section, we assume q is divisible by 4 and construct a supersymmetry on Rep(S ). We begin with the following observation: ∗ Proposition 6.3. Let V and W be s-representations of S and S . Regard S S as a e n m n×× m subgroup of Sn+m via jn,m, and similarly with n and m reversed. Then e e e e i : V W (W V )τñ,m e V,W ⊗ → ⊗ v w +nm v +nm w v w ( 1)| || | | | | |w v ⊗ 7→ − ⊗

is an isomorphism of representations of Sn Sm, where the superscript denotes the conjugate representation. Precisely, this means ×× e e 1 i (j (g, h)(v w))=τ ˜ j (g, h)˜τ − i (v w) V,W n,m ⊗ n,m n,m n,m V,W ⊗ for all v V and w W and all g S and h S . ∈ ∈ ∈ n ∈ m Proof. We have e e h v i (j (g, h)(v w))=( 1)| || |i (gv hw) V,W n,m ⊗ − V,W ⊗ h v +( g + v )( h + w )+nm( g + v )+nm( h + w ) =( 1)| || | | | | | | | | | | | | | | | | | hw gv − ⊗ g h +nm g +nm h + v w +nm v +nm w + g w =( 1)| || | | | | | | || | | | | | | || |hw gv − ⊗ 1 =τ ˜ j (g, h)˜τ − i (v w). n,m n,m n,m V,W ⊗ In the last equality, we used Proposition 5.5, and that c acts by 1. − As discussed in §6.3, we have an isomorphism e e Sn+m nm Sn+m (6.4) Inde e (V W ) Πr Ind e e (W V ) Sn Sm Sm Sn ×× ⊗ → ×× ⊗ coming from conjugation byτ ñ,m, and the isomorphism iV,W (maintaining the notation from the proposition). Definition 6.5. Given homogeneous objects V and W of Rep(S ), we define ∗ V W β : V W Π| || |(W V ) V,W ⊗ → r ⊗ e SUPERSYMMETRIC MONOIDAL CATEGORIES 39 to be the isomorphism induced by the isomorphisms (6.4) using the elementsτ ñ,m and the isomorphisms i. Explicitly, for v V and w W and g S , we have ∈ n ∈ m ∈ n+m nm( v + w )+ v w 1 nm β (g v w)=( 1) | | | | | || |gτ˜− w v π , V,W ⊗ ⊗ − n,m ⊗e ⊗ ⊗ e Sn+m where here g v w denotes an element of Ind e e (V W ). Sn Sm ⊗ ⊗ ×× ⊗ Proposition 6.6. β defines a type Ia supersymmetry on . ⊗ Proof. First, we check that β is a natural transformation. Let f : V V ′ and g : W W ′ n → n m → m be nonzero homogeneous morphisms. Pick v V , w W and σ S . Then we have ∈ n ∈ m ∈ n+m V W Π| || |(g f)β (σ v w) ⊗ Vn,Wm ⊗ ⊗ e nm( v + w )+ v w + f ( σ +nm+ w )+ g ( σ +nm) 1 nm =( 1) | | | | | || | | | | | | | | | | | στ˜− g(w) f(v) π . − n,m ⊗ ⊗ ⊗ On the other hand, we have

β ′ ′ (f g)(σ v w) Vn,Wm ⊗ ⊗ ⊗ g ( σ + v )+ f σ = β ′ ′ (( 1)| | | | | | | || |σ f(v) g(w)) Vn,Wm − ⊗ ⊗ g ( σ + v )+ f σ +nm( f + v + g + w )+( f + v )( g + w ) 1 nm =( 1)| | | | | | | || | | | | | | | | | | | | | | | | | στ˜− g(w) f(v) π . − n,m ⊗ ⊗ ⊗ f g This agrees with the ﬁrst expression up to ( 1)| || |, which is what we wanted to show. − n m Next, it follows from Proposition 5.3 that β β = ( 1)(2)( 2 ). Hence we only need V,W W,V − to check that one of the hexagon axioms hold, and we check (H1′). So pick non-negative integers m, n, p and let U, V, W be representations of Sm, Sn, and Sp, respectively. We need to check that (1V βU,W )(βU,V 1W ) = βU,V W . (Here we identify U with the S - ⊗ ⊗ ⊗ ∗ representation that is U in degree m and 0 in other degrees,e ande similarlye for V and W . We note that it suﬃces to check the axiom on such objects.) The upper path in the hexagon is the following composition: Indm+n+p(Indm+n(U V ) W ) m+n,p m,n ⊗ ⊗ α−1 U,V,W Indm+n+p(U Indn+p(V W )) −−−−→ m,n+p ⊗ n,p ⊗ β U,V ⊗W Indm+n+p(Indn+p(V W ) U) k[m(n + p)] −−−−→ n+p,m n,p ⊗ ⊗ ⊗ α−1 V,W,U Indm+n+p(V Indp+m(W U)) k[m(n + p)] −−−−→ n,m+p ⊗ p,m ⊗ ⊗

Here we have written n in place of Sn in the inductions for readability. We compute the composition explicitly. Let g Sm+n+p and let u U, v V , and w W be homogeneous elements. We have ∈ e ∈ ∈ ∈ e g (1 u v) w ⊗ ⊗ ⊗ ⊗ g u (1 v w) 7→ ⊗ ⊗ ⊗ ⊗ m(n+p)( u + v + w )+ u ( v + w ) 1 m(n+p) ( 1) | | | | | | | | | | | | gτ˜− (1 v w) u π 7→ − m,n+p ⊗ ⊗ ⊗ ⊗ ⊗ m(n+p)( u + v + w )+ u ( v + w ) 1 m(n+p) ( 1) | | | | | | | | | | | | gτ˜− v (1 w u) π 7→ − m,n+p ⊗ ⊗ ⊗ ⊗ ⊗ 40 STEVENVSAMANDANDREWSNOWDEN

The lower path in the hexagon is the following composition: Indm+n+p(Indm+n(U V ) W ) m+n,p m,n ⊗ ⊗ βU,V idW ⊗ Indm+n+p(Indm+n(V W ) k[mn] U) −−−−−−→ m+n,p n,m ⊗ ⊗ ⊗ m+n+p m+n ∼ Ind (Ind (V W ) U) k[mn] −→ m+n,p n,m ⊗ ⊗ ⊗ α−1 V,W,U Indm+n+p(V Indn+p(U W )) k[mn] −−−−→ n,m+p ⊗ n,p ⊗ ⊗ idV βU,W ⊗ Indm+n+p(V Indm+p(W U) k[mp]) k[mn] −−−−−−→ n,m+p ⊗ p,m ⊗ ⊗ ⊗ m+n+p ∼ Ind (V Indm+p(W U)) k[m(n + p)] −→ n,m+p ⊗ p,m ⊗ ⊗ We have g (1 u v) w ⊗ ⊗ ⊗ ⊗ mn( u + v )+ u v 1 mn ( 1) | | | | | || |g (τ − v u π ) w 7→ − ⊗ m,n ⊗ ⊗ ⊗ ⊗ mn( u + v )+ u v +mn w 1 mn ( 1) | | | | | || | | |gτ˜− (1 v u) w π 7→ − m,n ⊗ ⊗ ⊗ ⊗ ⊗ mn( u + v )+ u v +mn w 1 mn ( 1) | | | | | || | | |gτ˜− v (1 u w) π 7→ − m,n ⊗ ⊗ ⊗ ⊗ ⊗ mn( u + v )+ u v +mn w +mp( u + w )+ u w 1 1 mp mn ( 1) | | | | | || | | | | | | | | || |gτ˜− v (˜τ − w u π ) π 7→ − m,n ⊗ ⊗ m,p ⊗ ⊗ ⊗ ⊗ mn( u + v )+ u v +mn w +mp( u + w )+ u w +mp v 1 1 m(n+p) ( 1) | | | | | || | | | | | | | | || | | |gτ˜− j(1, τ˜− ) v (1 w u) π 7→ − m,n m,p ⊗ ⊗ ⊗ ⊗ ⊗ where here j = jn,m+p. In the final step, we used the identity 1 mp v 1 v (˜τ − w u)=( 1) | |(1, τ˜− )(v (1 w u)) ⊗ m,p ⊗ ⊗ − m,p ⊗ ⊗ ⊗ m+p 1 in V Indp,m (W U), and then moved (1, τ˜m,p− ) into the first tensor factor via j. Since 1 ⊗ 1 ⊗1 τ˜m,n− j(1, τ˜m,p− )=˜τm,n− +p (Proposition 5.6) and the signs in the two compositions agree, the result follows. 6.7. Universal property. Let Rep(S )pf be the full subcategory of Rep(S ) on finite length projective objects. If char(k) = 0 then∗ the projective condition is automatic.∗ The category Rep(S )pf is “nice” in the sense definede in §8.1: it is additive, Karoubian,e and admits a ∗ Π-structure. Let V be the “standard” representation of S : it is k in degree 1 and 0 in othere degrees. ∗ e Theorem 6.7. The pair (Rep(S )pf , V) is the universal nice type Ia supersymmetric monoidal Z-supercategory with an object of∗ degree 1, in the following sense. Suppose C is a nice supersymmetric monoidal Z-supercategorye and M is an object of C of degree 1. Then there exists a supersymmetric monoidal superfunctor Φ: Rep(S )pf C and an isomorphism ι: Φ(V) M. Furthermore, (Φ, ι) is unique up to isomorphism.∗ → → e Proof. One can prove this directly; however, we simply note that Rep(S )pf is the “nice ∗ envelope” of S˜, and so inherits the universal property of S˜ (Theorem 5.14). e 6.8. Exterior powers of categories. One can generalize the above discussion to use coefficients in a supercategory A. This leads to the notion of the exterior power of A, as defined in [GK]. The following discussion is intended as an extended comment, and so we omit many details. SUPERSYMMETRIC MONOIDAL CATEGORIES 41

Let A be a nice supercategory (see §8.1), and let A⊠ˆn be its n-fold self-product with respect to the tensor product ⊠ˆ defined in §8.6 below. The symmetric group Sn acts in a natural ⊠ˆn n way on A , and so Sn does too through the quotient map Sn Sn. A Sym -object of ⊠ˆn → A is an object M of A equipped with even isomorphisms ασ : σ∗(M) M for σ Sn → ⊠ˆ ∈ satisfying the usual cocyclee conditions. A n-object of A ise an object M of A n equipped with isomorphisms βσ : σ∗(M) M for σ Sn satisfying the cocycle conditions and such → V∈ n n that βc = 1 and βσ is homogeneous of the same degree as σ. We let Sym (A) and (A) − n n denote the categories of Sym - and -objectse of A. We define Sym(A) and (A) to be the Z-supercategories with these homogeneous pieces. V One can endow Sym(A) with a symmetricV monoidal structure and (A)V with a supersymmetric monoidal structure, similar to the definitions used above for Rep(S ). We note V ∗ that Rep(S ) itself can be viewed as (SVec). These categories satisfy natural universal properties: e.g.,∗ given a nice supersymmetric monoidal Z-supercategory B, givinge a super- symmetrice monoidal Z-functor (A) VB is the same as giving a superfunctor A B. → → V7. Queer vector spaces We assume k contains ζ throughout this section, and put √2= ζ (1 ζ ). 16 8 − 4 7.1. Queer vector spaces. Let V be a super vector space. A queer structure on V is an odd isomorphism ν : V V such that ν2 =1. A queer vector space is a vector space equipped with a queer structure.→ Suppose that (U, µ) and (V, ν) are queer vector spaces. A homogeneous morphism of queer spaces (U, µ) (V, ν) is a homogeneous linear map f → f : U V such that fµ = ( 1)| |νf. In this way, we have a supercategory QVec of queer vector→ spaces. − 7.2. The half tensor product. Let (U, µ) and (V, ν) be two queer vector spaces. Then µ ν defines an even automorphism of U V squaring to 1. We define the half tensor ⊗ 1 ⊗ − product of U and V , denoted 2− (U V ), to be the ζ -eigenspace of µ ν. We note that ⊗ 4 ⊗ the ( ζ4)-eigenspace is isomorphic to the ζ4-eigenspace, via µ 1 or 1 ν. The half tensor product− is an ordinary super vector space; it does not come with⊗ a preferred⊗ queer structure. 7.3. The queer monoidal category. Motivated by the above discussion, we define the queer category Queer to be the Z/2-supercategory with Queer0 = SVec and Queer1 = QVec. We now define a monoidal operation on Queer. We define on objects as follows: ⊙ ⊙ Given U, V Queer , we let U V = U V . • ∈ 0 ⊙ ⊗ Let U Queer0 and V Queer1, and let ν be the queer structure on V . Then • U V ∈carries the queer structure∈ 1 ν. We define U V to be the queer vector space⊗ (U V, 1 ν). We similarly define⊗ V U. ⊙ ⊗ ⊗ ⊙ 1 Given U, V Queer , we define U V to be the half tensor product 2− (U V ). • ∈ 1 ⊙ ⊗ The definition of on morphisms is straightforward. In this way, we have defined a superfunctor ⊙ : Queer ⊠ Queer Queer. ⊙ → The unit object for this monoidal structure is the unit object of Queer0, i.e., the super vector space k. The isomorphisms λ and ρ are similar to SVec, and we omit the details. We now discuss the associativity constraint α for . For this, we will require the following result: ⊙ 42 STEVENVSAMANDANDREWSNOWDEN

Proposition 7.1. Let (U, µ), (V, ν), and (W, σ) be queer vector spaces. Then µσ 1 is an 1 1 − automorphism of U V W that maps U 2− (V W ) isomorphically to 2− (U V ) W . ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ 2 1 Proof. Since (µσ) = 1, the inverse of µσ 1 is 2 (µσ + 1). Now suppose that x 1 − − − ∈ U 2− (V W ). Thus νσx = ζ4x; applying σ to each side, we find νx = ζ4σx. We thus have⊗ ⊗ − µν(µσ 1)x = νσx µνx = ζ x + ζ µσx = ζ (µσ 1)x, − − − − 4 4 4 − 1 which shows that (µσ 1)x belongs to 2− (U V ) W . By symmetry, σµ 1 maps 1 − 1 ⊗ ⊗ − 2− (U V ) W to U 2− (V W ). Since µσ 1 and σµ 1 are inverses (up to a factor of 2), these⊗ maps⊗ are isomorphisms.⊗ ⊗ − − Let U, V, W Queer be homogeneous objects. We define an isomorphism ∈ α : U (V W ) (U V ) W U,V,W ⊙ ⊙ → ⊙ ⊙ as follows. We identify the domain and target above with subspaces of U V W . If at least one of U, V , or W has degree 0, then the domain and target are identified⊗ ⊗ with the same subspace, and we take α to be the identity map. Suppose now that U, V , and W are each degree 1, and let µ, ν, and σ be the queer structures on them. We then take α = 1 (µσ 1), which is an appropriate isomorphism by Proposition 7.1. U,V,W √2 − Proposition 7.2. The pentagon axiom holds. Proof. Let U,V,W,X Queer be homogeneous objects. We let µ,ν,σ,τ denote queer structures on these objects,∈ respectively, whenever they have degree 1. We must show that the diagram (U V ) (W X) ⊙✐4 ⊙ ⊙❯❯ αU,V,W ⊙X ✐✐✐✐ ❯❯❯❯αU⊙V,W,X ✐✐✐✐ ❯❯❯❯ ✐✐✐✐ ❯❯❯❯ ✐✐✐ ❯❯* U (V (W X)) ((U V ) W ) X ⊙ ⊙ ⊙ ⊙ ⊙O ⊙ idU αV,W,X αU,V,W idX ⊙ ⊙ αU,V ⊙W,X U ((V W ) X) / (U (V W )) X ⊙ ⊙ ⊙ ⊙ ⊙ ⊙ commutes. We identify each space above with a subspace of U V W X. If at most two of U, V , W , and X are degree 1 then all maps in the diagram⊗ are⊗ the identity⊗ map, and so it commutes. Now suppose exactly three of the objects have degree 1. Then in each of the two paths in the pentagon, exactly one α is not the identity, and these two α’s are equal; 1 for example, if U = 0 then αU V,W,X = idU αV,W,X = (ντ 1). Finally, suppose that | | ⊙ ⊙ √2 − all four objects have degree 1. Then all edges in the pentagon except the bottom left and bottom right are the identity. We must show (µσ 1)(ντ 1)=2 on U (V (W X)). − − ⊙ ⊙ ⊙ Suppose x belongs to this space. Then we have µνx = ζ4x, and so νx = ζ4µx; similarly, στx = ζ4x, and so τx = ζ4σx. We thus find ντx = ζ νσx = ζ σνx = σµx = µσx. 4 − 4 − Thus (µσ 1)(ντ 1)x =(µσ 1)( µσ 1)x =2x, − − − − − since (µσ)2 = 1. This completes the proof. − SUPERSYMMETRIC MONOIDAL CATEGORIES 43

7.4. Supersymmetry. We now aim to deﬁne a supersymmetry on . We let τ be the standard symmetry on SVec. We begin with the following observation.⊙ Proposition 7.3. Let (U, µ) and (V, ν) be queer vector spaces. Then the diagram

µ ν U V ⊗ / U V ⊗ ⊗ τ τ ν µ V U ⊗ / V U ⊗ ⊗ commutes up to 1. In particular, τ maps the ζ4-eigenspace of µ ν on U V to the ( ζ )-eigenspace− of ν µ on V U. ⊗ ⊗ − 4 ⊗ ⊗ Proof. The first statement follows from the fact that τ is a natural transformation and µ and ν are odd (see the first paragraph of §3.4). The second statement follows from the first. Let U, V Queer. We define an isomorphism ∈ β : U V V U U,V ⊙ → ⊙ as follows. If at least one of U or V has degree 0 then βU,V = τU,V . Now suppose that both 1 U and V have degree 1, and let ν be the queer structure on V . Then βU,V = ζ16− (ν 1)τU,V . This maps U V into V U by Proposition 7.3 and properties of the half tensor⊗ product (see §7.2). ⊙ ⊙ Proposition 7.4. β defines a type IIb supersymmetry on . ⊙ Proof. One easily sees that β is a natural transformation, and it is evident that βU,V has degree U V when U and V are homogeneous objects. We will use the following convention in the remainder| || | of the proof: for an endomorphism µ of a super vector space U, we write µ still for the induced endomorphism of U V or V U. Note that, under this convention, ⊗ ⊗ µ commutes with τU,V . Now, let U, V Queer be homogeneous objects. If either U or V has degree 0 then ∈ βV,U βU,V is the identity. Now suppose that both U and V have degree 1, and let µ and ν be the queer structures. Then 1 1 1 βV,U βU,V = ζ8− µτV,U ντU,V = ζ8− µντV,U τU,V = ζ8− µν = ζ8. In the final step, we used the fact that µν = ζ on U V . We thus see that β β = 4 ⊙ V,U U,V B♯( U , V )idU V in all cases, which verifies the requisite symmetry condition for β. We| | now| | verify⊗ the hexagon axiom (H1). Let U, V, W Queer be homogeneous objects, and let µ,ν,σ be the queer structures on the objects of∈ degree 1. We must show that the diagram

βU,V ⊙W U (V W ) / (V W ) U α−1 ❦❦5 ⊙ ⊙ ⊙ ⊙ ❙❙❙ α−1 U,V,W ❦❦❦ ❙❙❙ V,W,U ❦❦❦ ❙❙❙ ❦❦❦ ❙❙❙ ❦❦❦ ❙❙) (U V ) W V (W U) ❙ ⊙ ⊙ ❙❙❙ ❦❦❦5 ⊙ ⊙ ❙❙❙ ❦❦❦ ❙❙❙ ❦❦❦ βU,V 1C ❙❙❙ ❦❦1B βU,W ⊙ ❙) ❦❦❦ ⊙ / (V U) W −1 V (U W ) ⊙ ⊙ αV,U,W ⊙ ⊙ 44 STEVENVSAMANDANDREWSNOWDEN

commutes. (Recall that B1 = 1 identically, which is why the diagram must commute exactly.) If at most one of the objects has degree 1, then this is the usual hexagon in SVec, which commutes. Now suppose exactly two objects have degree 1. Then all the associators are the identity (or, more precisely, induced by the standard associator on the ordinary tensor products). If U = 0 then each of the β’s is just a τ, and the diagram commutes. Otherwise, | | one of βU,V or βU,W is the usual τ, and the other involves the same queer structure as βU,V W ; for instance, if V = 0 then ⊙ | | 1 1 βU,V W = ζ16− στU,V W , βU,V = τU,V , βU,W = ζ16− στU,W . ⊙ ⊙ Thus the diagram commutes (the queer structure can be moved past all the τ’s and then one has the ordinary hexagon axiom in SVec). Finally, suppose that all objects have degree 1. The top path is (induced by) the map 1 1 (µν 1)τU,V W (σµ 1) = τU,V W (µν 1)(σµ 1). 2 − ⊗ − 2 ⊗ − − The bottom path is (induced by) the map 1 1 1 1 ζ− στ (σν 1)ντ = ζ− τ τ σ(σν 1)ν √2 8 U,W − U,V √2 8 U,W U,V − Since τU,V W = τU,W τU,V , it suffices to show that ⊗ 1 1 1 (µν 1)(σµ 1) = ζ− σ(σν 1)ν 2 − − √2 8 − as functions on (U V ) W . This equation is equivalent to ⊙ ⊙ 1 νσ µν σµ +1= √2ζ− (1 σν). − − 8 − Since µν = ζ on the source, we have µ = ζ ν, and so the left side of the above equation is 4 − 4 σν ζ + ζ σν +1=(1 ζ )(1 σν), − − 4 4 − 4 − 1 and so the result follows, as 1 ζ4 = √2ζ8− . Since β satisfies the symmetry− condition, the second hexagon axiom is a consequence of the first one. Thus the result is proved. 7.5. Further comments. The pentagon and hexagon axioms in Queer are rather complicated. In fact, there is an alternate way of defining the category Queer that makes these axioms more transparent. Recall that the Clifford algebra Cln is generated by n supercommuting odd elements that each square to 1. One formulation of Bott periodicity (discussed in Appendix A) states that Cln and Cln+2 are Morita equivalent. We can identify QVec with the category of Cl1-modules. The half tensor product is then the composition Mod Mod Mod SVec, Cl1 × Cl1 → Cl2 → where the first functor takes (V, W ) to V W , with obvious Cl2-structure, and the second functor is the Morita equivalence provided⊗ by Bott periodicity. It is the second step that introduces the complications. To obtain a “cleaner” version of the queer category, we can simply omit the Morita equivalence above. More precisely, we could redefine Queeri to be the 2-colimit of the categories

ModCln taken over n of the same parity as i. The tensor product would then be more natural and the axioms would be more transparent. This new approach also points the way to a vast generalization: one can start with an arbitrary Z/2-supercategory A and construct a “queer version” by considering Cleven-modules in A0 and Clodd-modules in A1. In fact, one can also work over more general base fields where SUPERSYMMETRIC MONOIDAL CATEGORIES 45 the periodicity in Clifford algebras may be 4 or 8 instead of 2. We call this generalization “Clifford eversion,” and carry out the details in the subsequent section. 8. Clifford eversion 8.1. Nice supercategories. Let C be a supercategory. We say that C is additive if all finite direct sums exist. (The direct sum of two objects is an object equipped with even inclusion and projection maps satisfying the usual conditions.) We say that C is Karoubian if every even idempotent e: A A in C has an image. We say that C is nice if it is additive, Karoubian, and admits→ a Π-structure. We say that a Λ-supercategory is nice if each graded piece is. We will mostly be concerned with nice supercategories in this section. We note that if M is an object in a nice supercategory and V is a finite dimensional super vector r s r s space then we can make sense of V M, e.g., k | M = M ⊕ Π(M)⊕ . (In fact, this construction does not require the Karoubian⊗ condition.)⊗ ⊕

8.2. The Cliﬀord algebra. The Cliﬀord algebra Cln is the k-superalgebra generated by odd-degree variables α ,...,α subject to the relations α2 = 2 and α α = α α for 1 n i i j − j i i = j. If 2 is a square in k then one can renormalize αi so that it squares to 1 (as we did in6 §7.5), which is a more common convention. We use our convention so that the following proposition works without need of √2:

Proposition 8.1. There is a group homomorphism S Cl× given by n → n 1 s˜i (αi+1 αi), c 1. 7→ 2 − e 7→ − Proof. Let σ = 1 (α α ). We have i 2 i+1 − i σ2 = 1 (α2 α α α α + α2)=1, i 4 i+1 − i+1 i − i i+1 i as the middle terms cancel and the outer terms are each equal to 2. It is clear that σi and σ anti-commute if i j > 1. Finally, we have j | − | σ σ σ = 1 (α α )= σ σ σ , i i+1 i 2 i − i+2 i+1 i i+1 so the braid relation holds. We thus see that the σ’s satisfy the relations deﬁning the spin- symmetric group, and the claim follows.

8.3. Cliﬀord modules. Let A be a supercategory. A Cln-module in A is a pair (M, ρ) consisting of an object M of A and a morphism of k-superalgebras ρ: Cl End(M). n → Explicitly, a Cln-module consists of an object M together with odd-degree supermorphisms α ,...,α : M M such that α2 = 2 and α α = α α for i = j. Suppose that M and 1 n → i i j − j i 6 N are two Cln-modules in A. A morphism f : M N of parity p is a morphism in A of parity p compatible with the Cl -structures, i.e., such→ that fα =( 1)pα f. We let Cl (A) n i − i n be the supercategory of Cln-modules in A. One easily sees that if A is additive or Karoubian then so is Cln(A). Moreover, if A admits a Π-structure then so does Cln(A), via

Π(M; α1,...,αn) = (Π(M);Π(α1),..., Π(αn)).

We thus see that if A is nice then so is Cln(A). The following statement is clear, and we omit its proof. Proposition 8.2. We have an equivalence Cl (Cl (A)) Cl (A) deﬁned by m n → n+m ((M; α ,...,α ); α′ ,...,α′ ) (M; α ,...,α ,α′ ,...,α′ ). 1 n 1 m 7→ 1 n 1 m 46 STEVENVSAMANDANDREWSNOWDEN

Given a Cl -module (M; α ,...,α ) in A and an element σ S , put n 1 n ∈ n

σ (M; α1,...,αn)=(M; ασ(1),...,ασ(n)). ∗

This construction deﬁnes a left action of Sn on the category Cln(A). We let Sn act via the homomorphism S S . The following proposition shows that every object of Cl (A) is n → n n canonically equivariant for the Sn action: e e Proposition 8.3. Let (M; α1,...,αe n) be a Cln-module. Then Sn naturally acts on M, and permutes the α’s via the surjection S S . In particular, for σ S , we have a natural n → n ∈ n isomorphism of Cln-modules e e e σ : (M; α ,...,α ) (M; α ,...,α ) 1 n → σ(1) σ(n) that has the same parity as σ.

Proof. The action comes from Proposition 8.1 using the homomorphism ϕ: Sn Cln×. We σ → claim that ϕ(σ)αi = ( 1)| |ασ−1(i)ϕ(σ) for all σ Sn. It is enough to check this when σ =s ˜ . Then we have (for− the third line, k / j, j +1∈ ) e j ∈{ } e ϕ(˜s )α = 1 (α α )α = 1 α α 1= 1 α (α α )= α ϕ(˜s ), j j 2 j+1 − j j 2 j+1 j − − 2 j+1 j+1 − j − j+1 j ϕ(˜s )α = 1 (α α )α =1 1 α α = 1 α (α α )= α ϕ(˜s ), j j+1 2 j+1 − j j+1 − 2 j j+1 − 2 j j+1 − j − j j ϕ(˜s )α = 1 (α α )α = 1 α (α α )= α ϕ(˜s ). j k 2 j+1 − j k − 2 k j+1 − j − k j

8.4. Periodicity. Throughout this section, we ﬁx a value p such that Clp is isomorphic as a superalgebra to a matrix superalgebra. (By Theorem A.4, we know that p = 8 is possible for any k and that we can take p = 2 if ζ4 k.) Let Up be a Clp-module such that the functor SVec Mod given by V V U∈is an equivalence, and let ǫ Cl be an even → Clp 7→ ⊗ p p ∈ p idempotent such that the functor ModClp SVec given by M ǫpM is a quasi-inverse. Let A be a nice supercategory. The following→ is clear. 7→

Proposition 8.4. The functor A Cl (A) given by M M U is an equivalence. → p 7→ ⊗ p We let Φ : Cl (A) Cl (A) be the equivalence induced by the identiﬁcation Cl (A)= n n → n+p n+p Clp(Cln(A)). For a non-negative integer i, deﬁne Cl(i;p)(A) to be the 2-colimit of the system

Φi Φi+p Cl (A) / Cl (A) / Cl (A) / i i+p i+2p ··· Explicitly, an object of Cl (A) is a Cl -module in A, for some integer n 0 such that (i;p) n ≥ n i (mod p). If M is a Clm-module and N is a Cln-module, with m n i (mod p), then≡ ≡ ≡ (r m)/p (r n)/p A A HomCl(i;p)( )(M, N) = lim HomClr( )(Φ − (M), Φ − (N)), −→ where the colimit is taken as r over integers r such that r i (mod p). It is clear that the natural functor Cl (A) →Cl ∞ (A) is an equivalence, and so,≡ in a sense, these limit i → (i;p) categories are nothing new. However, the categories Cl(i;p)(A) will sometimes be easier to work with, especially when dealing with products. SUPERSYMMETRIC MONOIDAL CATEGORIES 47

8.5. Clifford eversion. As before, let p be such that Clp is a matrix superalgebra, and fix an integer q divisible by p (we allow q = 0). Note that q is necessarily even. Definition 8.5. The Clifford eversion of a nice Z/q-supercategory A, denoted Cl(A), is the nice Z/q-supercategory given by Cl(A)i = Cl(i;p)(Ai).

Remark 8.6. We defined Cl(A)i to be Cl(i;p)(Ai). In some ways, it is more natural to use Cl(i;q)(Ai), but there is one problem with this: it is not defined if q = 0 and i< 0. Remark 8.7. The motivation for the definition of Clifford eversion comes from the queer category; see §7.5. We let Clk be the k-fold iterate of Cl. To study this, it is convenient to introduce a slight (k) (k) variant. Define Cl (A) to be the Z/q-supercategory given by Cl (A)i = Cl(ki;p)(A). Thus Cl(1) = Cl. The equivalence in Proposition 8.2 induces an equivalence Cl (Cl (B)) (i;p) (j;p) → Cl(i+j;p)(B) for any nice supercategory B. It follows that there is a natural equivalence Cl(k)(Cl(ℓ)(A)) Cl(k+ℓ)(A). In particular, we see that there is a natural equivalence → Clk(A) Cl(k)(A) for any k 0. → ≥ A p A Proposition 8.8. We have a canonical equivalence ∼= Cl ( ). Proof. Combine the natural equivalences A Cl(p)(A) and Clp(A) Cl(p)(A). → → Suppose A is a nice monoidal Z/q-supercategory. We define a natural monoidal structure on Cl(A) as follows. Let r, s Z/q and let r′ and s′ be non-negative integers congruent to ∈ r and s modulo p. We have a functor Cl ′ (A ) ⊠ Cl ′ (A ) Cl ′ ′ (A ) given by r r s s → r +s r+s (M; µ ,...,µ ′ ) (N; ν ,...,ν ′ ) (M N; µ ,...,µ ′ , ν ,...,ν ′ ). 1 r ⊗ 1 s 7→ ⊗ 1 r 1 s Here µi is shorthand for µi 1, and similarly for νj. Note that M is an object of Ar and N is an object of A , so M⊗ N is an object of A . Thus the above object does indeed s ⊗ r+s belong to Cl(A)r+s. The diagram

1 Φ ′ ⊗ s Clr′ (Ar) ⊠ Cls′ (As) / Clr′ (Ar) ⊠ Cls′+p(As)

Φr′+s′ Clr′+s′ (Ar+s) / Clr′+s′+p(Ar+s)

commutes, since in both cases the extra endomorphisms for Clp are appended to the end of the list. The diagram

Φ ′ 1 s ⊗ Clr′ (Ar) ⊠ Cls′ (As) / Clr′+p(Ar) ⊠ Cls′ (As)

Φr′+s′ Clr′+s′ (Ar+s) / Clr′+s′+p(Ar+s) does not commute, but does commute up to canonical even isomorphism. Precisely, let (M; µ ) and (N; ν ) be objects of Clr′ (Ar) and Cls′ (As), and let (Uq; λ1,...,λp) be the • • Clp-module inducing periodicity. Then the top path in the above diagram yields (M N; µ ,λ , ν ), while the bottom path yields (M N; µ , ν ,λ ). These two objects are iso-⊗ • • • ⊗ • • • morphic via Proposition 8.3, using jr′,p+s′ (1, τ˜p,s′ ) Sr′+s′+p. Note that this isomorphism is even since p is even. ∈ e 48 STEVENVSAMANDANDREWSNOWDEN

It follows from the above discussion that we have a well-defined functor Cl(A) ⊠ Cl(A) Cl(A). The associativity and unit constraints on Cl(A) are defined in the obvious manner,→ and one readily verifies that the axioms of a monoidal Z/q-supercategory are satisfied. We now investigate how the above construction interacts with braidings. In what follows, we assume 4 q. Suppose we have an isomorphism β : Σ. Given objects (M; α ) | ⊗→⊗◦ • and (N; α′ ) as above, we define • ′ β(′M;α•),(N;α ) : (M; α ) (N; α′ ) (N; α′ ) (M; α ) • • ⊗ • → • ⊗ • to be the composition −1 βM,N τ˜r,s (M N; α ,α′ ) (N M; α ,α′ ) (N M; α′ ,α ). ⊗ • • −−−→ ⊗ • • −−→ ⊗ • • The second map uses the action of S on N M, as in Proposition 8.3. We note that r+s ⊗ β′ = β + rs. For the following, we use the notation from §3.3. | | | | Proposition 8.9. Maintain the abovee notation.

(a) If β is a symmetry (or braiding) with respect to ω Γ0 then β′ is a type II supersymmetry (or superbraiding) with respect to DAω. ∈ (b) If β is a type II supersymmetry (or superbraiding) with respect to ω Γ1 then β′ is a symmetry (or braiding) with respect to Aω. ∈

Proof. Suppose that β is a braiding with respect to ω = (ω1,ω2). We show that β′ is A a type II superbraiding with respect to ω. We have already observed that βX,Y′ is homogeneous of degree X Y if X and Y are homogeneous objects. We now verify that | || | β′ is a natural transformation. Suppose that f : (M; µ ,...,µ ) (M ′; µ′ ,...,µ′ ) and 1 r → 1 r g : (N; ν1,...,νs) (N ′; ν1′ ,...,νs′ ) are homogeneous morphisms of homogeneous objects of Cl(A). Consider the→ diagram:

−1 βM,N τ˜r,s (M N; µ , ν ) / (N M; µ , ν ) / (N M; ν ,µ ) ⊗ • • ⊗ • • ⊗ • • f g g f g f ⊗ ⊗ ⊗ −1 βM′,N′ τ˜r,s (M ′ N ′; µ′ , ν′ ) / (N ′ M ′; µ′ , ν′ ) / (N ′ M ′; ν′ ,µ′ ) ⊗ • • ⊗ • • ⊗ • • f g The ﬁrst square commutes up to ( 1)| || |, as this is simply what it means that β is a natural − ( f + g )rs transformation, while the second square commutes up to ( 1) | | | | . This shows that β′ is natural. − We now look at the hexagon axioms. Let (M; µ ), (N; ν ), and (P, π ) be homogeneous objects of Cl(A) of degrees r, s, and t. Consider the• following• diagram: •

βM,NP x (MNP ; µ,ν,π) / (NP M; µ,ν,π) / (NP M; ν,π,µ) O βN,M z (NMP ; µ,ν,π) (NP M; ν,µ,π) ❯❯❯ ✐✐4 ❯❯❯❯ ✐✐✐✐ ❯❯❯❯ ✐✐✐✐ y ❯❯❯ ✐✐✐ βM,P ❯❯* ✐✐✐ (NMP ; ν,µ,π)

Here we have omitted and symbols, and x, y, and z are the appropriate elements of S ⊗ • ∗ appearing in the deﬁnition of β′. This is the hexagon (H1) for β′, where we have omitted e SUPERSYMMETRIC MONOIDAL CATEGORIES 49 the associators. Put

(α1,...,αr+s+t)=(µ1,...,µr, ν1,...,νs, π1,...,πt).

We regard each αi as acting on MNP , or any of the other tensor products in the above diagram, in the natural manner. We let Sr+s+t act on MNP via the α’s (as in Proposition 8.1). Then 1 1 rst 1 1 x =τ ˜− , y = j (˜τ −e , 1), z =( 1) y j (1, τ˜− ) y− . r,s+t r+s,t r,s − · · s,r+t r,t · 1 The expressions for x and y are clear. By definition, z is id τ˜− , but where thisτ ˜ N ⊗ r,t r,t element is built from the µ’s and π’s, i.e., the elements α1,...,αr and αr+s+1,...,αr+s+t. This agrees with the above formula; note that 1 rs yαiy− =( 1) α −1 − τr,s (i) by Proposition 8.3. One readily verifies that βM,P commutes with the αi’s (in the evident sense), and thus with y as well. Since βM,P βN,M = ω1(r; s, t)βM,NP (this is the first hexagon axiom for β) and js,r+t(1, τ˜r,t) jr+s,t(˜τr,s, 1)=τ ˜r,s+t (Proposition 5.6), we see that the above ·rst diagram commutes up to ( 1) ω (r; s, t), which verifies the first hexagon axiom for β′ with − 1 respect to DAω. The second hexagon axiom is similar. We thus see that β′ is a type II superbraiding. Now suppose that β is symmetric, and let us show that β′ is supersymmetric. Maintaining the above notation, we have

β(′N;ν),(M;µ)β(′M;µ),(N;ν) = yβN,M xβM,N 1 1 where x =τ ˜r,s− , and y is deﬁned likeτ ˜s,r− but with respect to αr+1,...,αr+s,α1,...,αr; in rs 1 1 rs 1 other words, y =( 1) xτ˜− x− =( 1) τ˜− . Since β and x commute, we have − s,r − s,r N,M r s rs 1 1 (2)(2)+rs β(′N;ν),(M;µ)β(′M;µ),(N;ν) =( 1) τ˜s,r− τ˜r,s− βN,M βM,N =( 1) ω♯( M , N )idM N , − − | | | | ⊗ where we have used Proposition 5.3. We thus see that β′ is a supersymmetry with respect to DAω. We have thus proved statement (a). Statement (b) is similar, though some additional signs come in: in the hexagon axiom, we pick up an additional ( 1)rst when commuting β and y, and in the symmetry axiom we pick up an additional− ( 1)rs when commuting M,P − βN,M andτ ˜r,s. Remark 8.10. If one only cares about (super)braidings, one can phrase the above proposition using B-factor systems instead of S-factor systems.

The above proposition can be summarized pictorially as follows: for ω Γ0, Clifford eversion acts by ∈ ω DAω Dω Aω ω . → → → → →··· That is, given a (super)symmetric monoidal category for one factor system, the Clifford eversion will be a (super)symmetric monoidal category for the next factor system. The list of factor system repeats with periodicity 4, while Clifford eversion itself repeats with period p (which is a divisor of 8). We have seen that there is a canonical equivalence A Clp(A). If A is (super)braided or (super)symmetric then so is Clp(A), and with respect to→ the same factor system (recall that q is divisible by 4). We now show that this equivalence is appropriately monoidal. 50 STEVENVSAMANDANDREWSNOWDEN

Proposition 8.11. The equivalence A Clp(A) is monoidal, (super)braided monoidal, or (super)symmetric monoidal, depending on→ what structure A has. Proof. The definition of the monoidal structure and (super)braiding on Cl(A) generalizes in the obvious manner to Cl(k)(A) for any k. The equivalences Cl(k)(Cl(ℓ)(A)) Cl(k+ℓ)(A) are → easily seen to be compatible with these structures; it follows that the equivalence Clk(A) → Cl(k)(A) is as well. The equivalence A Cl(p)(A) induced by Clifford periodicity is also compatible with these structures. → Combining the above results, we find:

Proposition 8.12. Suppose q is divisible by 4 and let ω Γ0. Then the 2-category of symmetric monoidal Z/q-supercategories with respect to ω is∈ equivalent to the 2-category of type II supersymmetric monoidal Z/q-supercategories with respect to Aω. Specializing to the canonical choice ω = 1, we find: Corollary 8.13. The 2-category of symmetric monoidal Z/q-supercategories is equivalent to the 2-category of type IIa supersymmetric monoidal Z/q-supercategories, via Clifford eversion. Remark 8.14. We have assumed 4 q throughout the above discussion. There are analogous results if p =2 and 2 q, assuming| one has sufficiently many roots of unity in k. | 8.6. Envelopes. Let C be a supercategory. We define its additive envelope Cadd as follows. The objects are symbols (formal direct sums) A A where the A are objects of C. A 1 ⊕···⊕ r i morphism f : A A B B is an element (f ) HomC(A , B ), thought 1 ⊕···⊕ r → 1 ⊕···⊕ s i,j ∈ i,j i j of as a matrix. We say that f is homogeneous of parity p if each fi,j is homogeneous of parity Q p. Given another morphism g : B1 Bs C1 Ct, the composition gf is defined ⊕···⊕ → Cadd⊕···⊕ by (gf)i,j = k gk,jfi,k. One readily verifies that is an additive supercategory. We have a fully faithful functor i: C Cadd that satisfies a universal property: given a superfunctor j : C A toP an additive supercategory→ A, there is a unique (up to isomorphism) superfunctor add→ j′ : C A such that j = j′i. In particular, i is an equivalence if C is additive. The Karoubian→ envelope of a supercategory A, denoted Akar, consists of objects (A, e) where A is an object of C and e is an even idempotent of A. A morphism (A, e) (B, f) in Akar is a morphism ϕ: A B such that fϕ = ϕ = ϕe. (The identity of (A, e)→ is given by e: A A.) One easily sees→ that Akar is a Karoubian supercategory. We have a fully faithful superfunctor→ i: A Akar defined by i(A)=(A, 1) that satisfies a universal property: given an superfunctor j : A→ B to a Karoubian supercategory B, there is a unique (up to → kar isomorphism) superfunctor j′ : A B such that j = j′i. In particular, i is an equivalence if A is Karoubian. → Lemma 8.15. Let F : A B be a fully faithful superfunctor of supercategories. Then F add : Aadd Badd and F kar→: Akar Bkar are also fully faithful. → → Proof. This is clear for F add by the explicit description for morphisms given above. Similarly, a morphism (A, e) (A′, e′) in the Karoubian envelope is just a morphism A A′ satisfying some conditions which→ are preserved by F , so F kar is also fully faithful. → Let A and B be nice Z/q-supercategories. The product A ⊠ B will typically not be additive or Karoubian, and thus not nice. We define A ⊠ˆ B to be the Karoubian envelope SUPERSYMMETRIC MONOIDAL CATEGORIES 51 of the additive envelope of A ⊠ B. This is again a nice Z/q-supercategory, and ⊠ˆ endows the 2-category of nice Z/q-supercategories with a monoidal structure. We note that if A is a nice monoidal Z/q-supercategory then the monoidal product A ⊠ A A canonically → extends to a superfunctor A ⊠ˆ A A, by formal properties of envelopes. → 8.7. 2-categorical aspects. We now explain some 2-categorical aspects of Clifford eversion. As in §4, we do not intend this discussion to be completely rigorous. Fix a positive integer q divisible by 4, and let C be the 2-category of nice Z/q-supercategories, which we regard as monoidal under the product ⊠ˆ . Let ω Γ0 be an even S-factor system. Then Σ and ω induce a symmetric monoidal structure on∈C , just as T and an odd S-factor system do (see §4). We write (C , Σ,ω) to denote the resulting symmetric monoidal 2-category. Of course, we similarly get a symmetric monoidal 2-category from T and an odd S-factor system. The following is the main statement we want to explain, though we do not give a complete proof: Theorem 8.16. Clifford eversion defines an equivalence of symmetric monoidal 2-categories Cl: (C , Σ,ω) (C , T, Aω). → Lemma 8.17. Let A and B be nice Z/q-supercategories. We have a canonical equivalence

iA B : Cl(A) ⊠ˆ Cl(B) Cl(A ⊠ˆ B). , → Proof. Let n,m jA B : Cl (A) ⊠ Cl (B) Cl (A ⊠ˆ B) , n m → n+m be the functor defined by (M; µ ,...,µ ) ⊠ (N; ν ,...,ν ) (M ⊠ N; µ ,...,µ , ν ,...,ν ), 1 n 1 m 7→ 1 n 1 m where here we write µi in place of µi id on the right side, and similarly for νj. n,m ⊗ A We claim that jA,B is fully faithful. Faithfullness is clear. Let A, A′ be objects of Cln( ) and let B, B′ be objects of Cl (B). A homogeneous morphism A B A′ B′ in m ⊗ → ⊗ Cl (A ⊠ˆ B) is an element f = g h HomA(A, A′) HomB(B, B′), where the n+m k k ⊗ k ∈ ⊗ gk are homogeneous and linearly independent, and similarly for the hk, such that f P f f(µ 1)=( 1)| |(µ′ 1)f and f(1 ν )=( 1)| |(1 ν′ )f i ⊗ − i ⊗ ⊗ j − ⊗ j hk f for all i, j. The first condition becomes k( 1)| |gkµi hk = k( 1)| |µi′ gk hk, so − ⊗ gk − ⊗ by linear independence of the hk, we deduce that gkµi = ( 1)| |µi′ gk for all i, k. (Note: P h − P n,m f = g + h .) Similarly, we deduce that h ν = ( 1)| k|ν′ h for all j, k. Hence j is | | | k| | k| k j − j k A,B full. Thus each gk defines a morphism A A′ in Cln(A) and each hk defines a morphism B → n,m B B′ in Clm( ), and so f is in the image of jA,B . This proves the claim. → n,m One easily sees (using an argument we have already made) that the jA,B are compatible with the periodicity equivalences. They therefore induce a superfunctor

jA B : Cl(A) ⊠ Cl(B) Cl(A ⊠ˆ B) , → that is also fully faithful. Let

iA B : Cl(A) ⊠ˆ Cl(B) Cl(A ⊠ˆ B) , → be the functor induced by jA,B by taking envelopes; the target does not change since it is already additive and Karoubian. By Lemma 8.15, we see that iA,B is fully faithful. Iterating this construction, we obtain a fully faithful functor k k k k iA B : Cl (A) ⊠ˆ Cl (B) Cl (A ⊠ˆ B) , → 52 STEVENVSAMANDANDREWSNOWDEN

for any k 0. Consider the diagram ≥ ip p 1 p 1 A,B p 1 Cl − (Cl(A)) ⊠ˆ Cl − (Cl(B)) / Cl − (Cl(A ⊠ˆ B)) . ❱❱❱ ❥4 ❱❱❱❱ ❥❥❥❥ ❱❱❱❱ ❥❥❥ p−1 ❱❱❱ ❥❥❥p−1 i ❱❱ ❥❥ Cl (iA,B) Cl(A),Cl(B) ❱❱* ❥❥ p 1 Cl − (Cl(A) ⊠ˆ Cl(B)) This diagram commutes, essentially by definition. The two categories in the top row are A ⊠ˆ B p each identified with by Proposition 8.8, and iA,B is easily seen to be identified with the identity functor; thus the top line is an equivalence. Since the two diagonal functors are fully faithful and their composite is an equivalence, each is an equivalence. Since Cl is an p 1 equivalence of 2-categories, again by Proposition 8.8, and Cl − (iA,B) is an equivalence, it follows that iA,B is an equivalence. Lemma 8.18. The diagram

iA,B Cl(A) ⊠ˆ Cl(B) / Cl(A ⊠ˆ B)

TCl(A),Cl(B) Cl(ΣA,B) iB,A Cl(B) ⊠ˆ Cl(A) / Cl(B ⊠ˆ A) commutes up to a canonical even isomorphism.

Proof. Let (M; µ ) be(N; ν ) be objects of Clm(A) and Cln(B). Then • • iA,B((M; µ ) (N; ν ))=(M N; µ , ν ) • ⊗ • ⊗ • • where here we have written µ in place of µ 1 and ν in place of 1 ν . The functor i i ⊗ j ⊗ j ΣA B : A ⊠ˆ B B ⊠ˆ A induces a functor Cl (ΣA B): Cl (A ⊠ˆ B) Cl (B ⊠ˆ A) that , → m+n , m+n → m+n takes the above object to (N M; µ , ν ), where now µi means 1 µi and νj means νj 1. ⊗ • • ⊗ ⊗ Note that the order of the µ and ν does not change, since Clm+n(ΣA,B) simply applies ΣA,B to all components of the input. On the other hand, T((M; µ ) (N; ν ))=Πmn(N; ν ) (M; µ ). • ⊗ • • ⊗ • Applying iB,A to this yields Πmn(N M; ν ,µ ). ⊗ • • We have a natural even degree isomorphism (N M; µ , ν ) Πmn(N M; ν ,µ ) ⊗ • • → ⊗ • • 1 1 coming from Proposition 8.3 using the elementτ ˜m,n− defined in §5.3: note thatτ ˜m,n− has parity mn. We know that Clifford eversion is an equivalence of 2-categories C C . The main content of Theorem 8.16 is that it admits a symmetric monoidal structure.→ A definition of braided monoidal 2-functor is given in [BN, §4], though this definition is not the most general possible, and only concerns braidings (not symmetries or syllepses). The above lemmas define much of the structure required for Cl to be a braided monoidal 2-functor (though we have not discussed the modification α in [BN, Definition 16]). To complete the proof that Cl is braided monoidal, we would need to verify the quite complicated conditions in loc. cit. We leave this (together with the problem of upgrading from braided monoidal to symmetric SUPERSYMMETRIC MONOIDAL CATEGORIES 53 monoidal) as an open problem. We note that the particular factor systems appearing in Theorem 8.16 have not appeared in the above discussion; however, we know these factor systems must be the correct ones due to Proposition 8.9: precisely, if A is a symmetric monoidal object in (C , Σ,ω), i.e., a symmetric monoidal supercategory with respect to ω, then by Proposition 8.9, Cl(A) is a type II supersymmetric monoidal supercategory with respect to ADω, and therefore (can be converted to) a type I supersymmetric monoidal supercategory with respect to Aω (via Proposition 3.34), i.e., a symmetric monoidal object of (C , T, Aω).

1 1 8.8. Grothendieck groups and eversion. Fix a Cl1-module (k | ,ξ) in SVec. Explicitly, 1 0 0 1 let e0 be a basis vector for k | and let e1 be a basis vector for k | and let ξ be the map given by e e and e 2e . 0 7→ 1 1 7→ 0 Let A be a nice supercategory. For an object M of A, let F (M) be the Cl1-module 1 1 (M k | , 1 ξ) in A; this defines a functor F : A Cl1(A). Let G: Cl1(A) A be the forgetful⊗ functor,⊗ i.e., G(M,α)= M. → → Proposition 8.19. We have natural even isomorphisms G F = id Π and F G = id Π. ◦ ⊕ ◦ ⊕ Proof. The statement for G F is clear. Suppose that (M,α) is a Cl1-module in A. Then 1 1 ◦ M k | is naturally a Cl2-module via α = α 1 and ξ = 1 ξ. One easily verifies that ⊗ 1 1 ⊗ 1 1⊗ 2 ξα defines an even isomorphism (M k | ,α) (M k | ,ξ) of Cl1-modules; this is a − ⊗1 1 → ⊗ variant of Proposition 8.3. Since (M k | ,α) is naturally isomorphic to (M,α) Π(M,α) ⊗ ⊕ in Cl1(A) by an even degree isomorphism, the result follows.

Now suppose that A is equipped with an exact structure. Then Cln(A) has a natural exact structure as well. Proposition 8.20. The forgetful functor Cl (A) A induces an isomorphism 1 → K (Cl (A))[ 1 ] K (A)[ 1 ]. + 1 2 → + 2 Proof. This follows from Proposition 8.19 since id Π induces multiplication by 2 on K . ⊕ +

Now suppose that A is a Z/2-supercategory with an exact structure and that ζ4 k. We take p = 2 in Clifford periodicity, so that the Clifford eversion Cl(A) is also a Z∈/2- supercategory. We define a Z[1/√2]-linear map

i: K (Cl(A)) Z[1/√2] K (A) Z[1/√2] + ⊗ → + ⊗ as follows. If A is an object of Cl(A)0 = A0, then i([A]) = [A]. If (A, α) is an object of Cl1(A) = Cl1(A1), then we deﬁne i([(A, α)]) = [A]/√2. Thus, on the odd piece we are just scaling the map induced by the forgetful functor by 1/√2. It follows from Proposition 8.20 that i is an isomorphism of abelian groups. The following proposition explains the reason for the √2 factor: Proposition 8.21. If A has a monoidal structure then i is a ring isomorphism.

Proof. Let A, A′ be objects of A0 and let (M,α), (M ′,α′) be objects of Cl1(A1). Then A A′ is the same when computed in Cl(A) or A, so i([A][A′]) = i([A])i([A′]). Similarly, A ⊗ (M,α)=(A M, 1 α), so again i([A][(M,α)]) = i([A])i([(M,α)]). ⊗ ⊗ ⊗ 54 STEVENVSAMANDANDREWSNOWDEN

Finally, via Cl (A) = A,(M,α) (M ′,α′) is identiﬁed with the ζ -eigenspace N of α α′. 2 ∼ ⊗ 4 ⊗ The ζ -eigenspace and ζ -eigenspace of M M ′ are interchanged by α 1, and hence we 4 − 4 ⊗ ⊗ see that (M,α) (M ′,α′) = N Π(N), so ⊗ ∼ ⊕ 1 i([(M,α)][(M ′,α′)]) = [N]= [M][M ′]= i([(M,α)])i([(M ′,α′)]). 2

9. Schur–Sergeev duality We work over k = C in this section for simplicity.

9.1. Polynomial representations. For an integer n 0, we let q(n) gl(n n) denote A B≥ ⊂ | the super Lie subalgebra of matrices of the form where A, B are n n matrices. B A × This is the queer Lie superalgebra. Alternatively, if Vn is a vector space of dimension n, 1 1 q(Vn) is the commutator of the Cl1-module structure on Vn n = Vn k | in gl(Vn n). There | ⊗ | are natural embeddings Vn n Vn+1 n+1 which induce embeddings q(n) q(n + 1), and we define q( ) to be the direct| → limit. | → ∞ d Let V = lim Vn n; this is the standard representation of q( ). The tensor powers Vn⊗n n | ∞ | are semisimple,−→ and the centralizer of q(n) is the Hecke–Clifford algebra Hd (defined below) d d H when n d [WW, Theorem 4.7]. The inclusion maps Vn⊗n Vn⊗+1 n+1 are d-linear and ≫ d | → | hence the tensor powers V⊗ are semisimple. The simple constituents of this representation are naturally indexed by strict partitions of d; we let Lλ denote the simple corresponding to the partition λ [WW, Theorem 4.8]. We say that a representation of q( ) is polynomial if ∞ pol it is a direct sum of Lλ’s. The category of polynomial representations is denoted Rep (q). It is naturally a symmetric monoidal Z-supercategory. The object Lλ is homogeneous of degree λ . | | Remark 9.1. Even though Π(V) = V, we do not have Π(V ) = V for all objects V in pol ∼ ∼ Rep (q). More precisely, the endomorphism algebra of the irreducible Lλ, where λ is a strict partition, is Cl1 if and only if ℓ(λ) is odd [CW, Lemma 1.40].

9.2. The Hecke–Clifford algebra. The symmetric group Sn acts on the algebra Cln by g α = α . The Hecke–Clifford algebra is H = C[S ]⋉Cl . This is a superalgebra via · i g(i) n n n the grading deg(g)=0 for g Sn and deg(αi) = 1. We define a category Rep(H ) whose ∈ ∗ objects are sequences (Vn)n 0 where Vn is a finite-dimensional super vector space with a ≥ homogeneous action of Hn such that Vn = 0 for n 0. We define Π: Rep(H ) Rep(H ) by (ΠV ) = Π(V ). There is a natural identification≫ of H = H H with∗ a→ subalgebra∗ n n n,m n ⊗ m of Hn+m, and we define a monoidal structure on Rep(H ) via ∗ n H ⊠ (V W )n = n Hi,n i (Vi Wn i). ⊗ ⊗ − − i=0 M We define a symmetry β′ on this tensor product by

v w 1 a v w ( 1)| || |aτi,n− i w v ⊗ ⊗ 7→ − − ⊗ ⊗ where a Hn, v Vi, and w Wn i. We omit the straightforward check that this is well-deﬁned.∈ ∈ ∈ − SUPERSYMMETRIC MONOIDAL CATEGORIES 55

The Hecke–Cliﬀord algebra is closely related to the queer superalgebra. Let ξ : V V n → be the given odd endomorphism. The algebra Cln has a right action on V⊗ via

vi + + vn (v v )α =( 1)| +1| ··· | |(v ξ(v ) v ), 1 ⊗···⊗ n i − 1 ⊗··· i ···⊗ n where the vi are homogeneous elements of V. The symmetric group Sn also has a right action n on V⊗ , in the usual manner. Together, these define a right action of the Hecke–Clifford n algebra Hn on V⊗ . The following is a typical formulation of Schur–Sergeev duality: H pol n Proposition 9.2. The functor Rep( ) Rep (q) given by (Nn)n n V⊗ Hn Nn is an equivalence of symmetric monoidal ∗Z-supercategories.→ 7→ ⊗ L n Proof. Let Φ denote the functor under discussion. The explicit decomposition of V⊗ given in [WW, Theorem 4.8] shows that Φ is an equivalence of abelian categories. Let M and N be Hm and Hn-modules, respectively. Then (m+n) Φ(M N)= V⊗ H (H H (M N)) ⊗ ⊗ m+n m+n ⊗ m,n ⊗ m n = (V⊗ V⊗ ) H (M N) ∼ ⊗ ⊗ m,n ⊗ m n = (V⊗ H M) (V⊗ H N) ∼ ⊗ m ⊗ ⊗ n = Φ(M) Φ(N), ⊗ which shows that Φ is monoidal. Finally, we verify the symmetry. Let β be the symmetry pol m on Rep (q). Pick v x Φ(M) and v′ y Φ(N) where x M, y N, v V⊗ , and n ⊗ ∈ ⊗ ∈ ∈ ∈ ∈ v′ V⊗ are homogeneous. Then ∈ ( v + x )( v′ + y ) β((v x) (v′ y))=( 1) | | | | | | | | (v′ y) (v x) ⊗ ⊗ ⊗ − ⊗ ⊗ ⊗ v v′ + x v′ + x y =( 1)| || | | || | | || |(v′ v) (y x). − ⊗ ⊗ ⊗ On the other hand, under the identification Φ(M) Φ(N) = Φ(M N), (v x) (v′ y) v′ x ⊗ ∼ ⊗ ⊗ ⊗ ⊗ maps to ( 1)| || |(v v′) (1 x y) where 1 is the unit of H . If we apply the symmetry − ⊗ ⊗ ⊗ ⊗ m+n β′ in Rep(H ) to the second factor, then we get ∗ v′ x + x y 1 v′ x + x y + v v′ ( 1)| || | | || |(v v′) (τ − y x)=( 1)| || | | || | | || |(v′ v) (1 y x), − ⊗ ⊗ m,n ⊗ ⊗ − ⊗ ⊗ ⊗ ⊗ which agrees with the previous expression, so Φ is symmetric. 9.3. Sergeev–Schur duality. Let β be the type II supersymmetry on Rep(S ) obtained from the type I supersymmetry described in §6.6 by the general process in §∗3.7. Thus, explicitly, if M is a representation of Sm and N is a representation of Sn thene e e Sn+m Sn+m βM,N : Ind e e (M N) Inde e (N M) Sm Sne Sn Sm e ×× ⊗ → ×× ⊗ is given by v w +mn g +mn 1 g v w ( 1)| || | | | gτ˜− w v. ⊗ ⊗ 7→ − m,n ⊗ ⊗ By results in the aforementioned sections, β is a type II supersymmetry with respect to the factor system DA. Define β∗ by (mn) mn β∗ =( 1) 2 ζ β M,N − 4 M,N mn where m = M and n = N . Note that ( 1)( 2 )ζmn = d(m, n) where d is as defined in | | | | − 4 Example 3.14, and so β∗ is a type II supersymmetry with respect to A. The following is our reformulation of Sergeev–Schur duality, and the main result of this section: 56 STEVENVSAMANDANDREWSNOWDEN

Theorem 9.3. We have a natural equivalence of symmetric monoidal Z-supercategories Reppol(q) = Cl(Rep(S )), ∼ ∗ where here Rep(S ) is equipped with the supersymmetry β∗. ∗ e The proof will take the rest of the section. We omit the proof for the following fact, which e is a straightforward calculation. Lemma 9.4. We have an isomorphism of superalgebras C[S ]/(c + 1) Cl H n ⊗ n → n given by 1 α α and s˜ 1 ζ4 s (α α ). ⊗ i 7→ i j ⊗ 7→e2 j j − j+1 Proposition 9.5. We have an equivalence of symmetric monoidal Z-supercategories Cl(Rep(S )) = Rep(H ). ∗ ∼ ∗ Proof. Via the isomorphism in Lemma 9.4, an H -module M is the same thing as a Cl - e n n module in Rep(Sn). This gives an equivalence Φ: Rep(H ) Cl(Rep(S )) which preserves ∗ → ∗ the Z-grading. We claim that this is a monoidal equivalence. Let M be an Hm-module H H and let N be ane n-module. We have Clm Cln = Clm+n, so that em+n Hm,n (M N) e ⊗ ∼ ⊗ ⊗ Sm+n is isomorphic, as a vector space, to Ind e e (M N). The action of Clm is via M and Sm Sn ⊗ the action of Cln is via N, so that the action×× of Clm+n is via the tensor product action, and hence Φ is a monoidal equivalence. We now check symmetry. We have homomorphisms ϕ: S H , s˜ ζ4 s (α α ) n+m → n+m i 7→ 2 i i+1 − i ψ : S H , s˜ 1 (α α ). e n+m → n+m i 7→ 2 i+1 − i The existence of ψ comes from Proposition 8.1, while the existence of ϕ comes from Lemma 9.4. e Observe that ϕ(˜si) skew-commutes with αj, for all i and j, and so ϕ(˜si) and ψ(˜sj) skew- commute for all i and j. We also have ϕ(˜si)= ζ4siψ(˜si), and so ϕ(˜si)ψ(˜si)= ζ4si. It follows that we have mn mn τ =( 1)( 2 )ζ ϕ(˜τ )ψ(˜τ ). m,n − 4 m,n m,n The vector space isomorphism e Sm+n H Inde e (M N) m+n Hm,n (M N) Sm Sn ×× ⊗ → ⊗ ⊗ is given by g v w ϕ(g) v w. We thus have ⊗ ⊗ 7→ ⊗ ⊗ v w 1 β′ (ϕ(g) v w)=( 1)| || |ϕ(g)τ − w v M,N ⊗ ⊗ − m,n ⊗ ⊗ v w +(mn) mn 1 1 =( 1)| || | 2 ζ− ϕ(g)ψ(˜τ )− ϕ(˜τ )− w v − 4 m,n m,n ⊗ ⊗ v w +(mn)+mn g mn 1 1 =( 1)| || | 2 | |ζ− ψ(˜τ )− ϕ(gτ˜− ) w v − 4 m,n m,n ⊗ ⊗ 1 = ψ(˜τ )− β∗(g v w). m,n ⊗ ⊗ The expression on the ﬁnal line is exactly the symmetry on Cl(Rep(S )), and so the result follows. ∗ e Theorem 9.3 now follows from Propositions 9.5 and 9.2. SUPERSYMMETRIC MONOIDAL CATEGORIES 57

9.4. Symmetric functions. In this section, λ =(λ1 λr) denotes an integer partition, ℓ(λ) denotes the number of nonzero entries of λ,≥···≥ and λ = λ . | | i i Let Λ be the ring of symmetric functions in an alphabet x1, x2,... . Deﬁne symmetric P functions qk(x) by the identity

k 1+ xit qk(x)t = . 1 xit k 0 i 1 X≥ Y≥ − Let Γ be the subring of Λ generated by q1, q2,... . They satisfy the relations, for n> 0, n i ( 1) qiqn i =0. − − i=0 X So Γ Z[ 1 ] is generated by q , q , q ,... , which are algebraically independent. For a, b, deﬁne ⊗ 2 1 3 5 b i Q(a,b) = qaqb +2 ( 1) qa+iqb i − − i=1 X (with the convention that q = 0 if i < 0). Note that Q = Q . Let λ =(λ > > i (b,a) − (a,b) 1 ··· λ2r 0) be a strict partition (here ℓ(λ)=2r if λ2r > 0 and otherwise ℓ(λ)=2r 1) we deﬁne≥ −

Qλ = Pf(Q(λi,λj ))1 i,j 2r ≤ ≤ where Pf denotes the Pfaﬃan of a skew-symmetric matrix. The Qλ form a Z-basis for Γ [Ma, (8.9)] and Γ Z[1/√2] is a model for the Grothendieck groups Reppol(q), Rep(S ), and Rep(H ) (in each⊗ case, the ring structure comes from the monoidal product) as we now∗ ∗ explain. In particular, in each case, the simple objects are indexed by strict partitions. e pol ℓ(λ)/2 For Rep (q) and Rep(H ), the classes of the simple objects Lλ are 2−⌊ ⌋Qλ [CW, ∗ (ε(λ) ℓ(λ))/2 Theorem 3.48] and for Rep(S ), the classes of the simple objects Nλ are 2 − Qλ where ε(λ) 0, 1 is the reduction∗ modulo 2 of λ ℓ(λ)[HH, Theorem 8.6]. ∈{ } | | − In particular, when λ is even,e we have [Lλ] = [Nλ], and when λ is odd, we have | | | | 1 [N ] if ℓ(λ) is even √2 λ [Lλ]= . (√2[Nλ] if ℓ(λ) is odd

Recall from Proposition 9.5 that we have an equivalence Rep(Hn) Cl1(Rep(Sn)) when n is odd. Hence, from Proposition 8.21, we have an isomorphism ≃ e K (Cl (Rep(S ))) Z[1/√2] K (Rep(S )) Z[1/√2] + 1 n ⊗ → + n ⊗ [(A, α)] [A]/√2. e 7→ e To match this with the Q-functions, we note that N is a queer object if and only if λ λ | | − ℓ(λ) is odd [Jo, Corollary 4.19]. Hence when λ is odd, the simple objects of Cl1(Rep(Sn)) are | | 1 1 N ℓ(λ) is even N k | ℓ(λ) is odd e { λ | }∪{ λ ⊗ | } and the correspondence becomes 1 [N ] if ℓ(λ) is even √2 λ [Lλ] 1 1 1 . 7→ [N k | ] if ℓ(λ) is odd ( √2 λ ⊗ 58 STEVENVSAMANDANDREWSNOWDEN

Appendix A. Bott periodicity

Let k be a field of characteristic different from 2. In this paper, the Clifford algebra Clk 2 is the k-superalgebra generated by odd elements α1,...,αk subject to the relations αi = 2 for all i and αiαj = αj αi for all i = j. Our goal here is to establish a version of Bott periodicity (Theorem−A.4) for these Clifford6 algebras. While this is well-known, we have provided the details since most references work over k = R and our definition of the algebra is not standard. Here we follow the proof in [Tr]. Given a superalgebra A, the opposite superalgebra Aop has the same underlying vector a b space but multiplication of homogeneous elements is given by a op b = ( 1)| || |b a. In particular, the presentation of Clop is the same as Cl except that the· α square− to 2 instead· k k i − of 2. Also, note that Clk Clℓ ∼= Clk+ℓ where we emphasize that we are use the tensor product ⊗ x2 y1 of superalgebras, so that on homogeneous elements (x y )(x y )=( 1)| || |x x y y . 1 ⊗ 1 2 ⊗ 2 − 1 2 ⊗ 1 2 op 1 1 Lemma A.1. We have an isomorphism of superalgebras Cl Cl = End(k | ). 1 ⊗ 1 ∼ 1 1 Proof. Let e0, e1 be a homogeneous basis for k | . We represent α1 1 by the operator e e , e 2e and 1 α by the operator e e , e 2e . ⊗ 0 7→ 1 1 7→ 0 ⊗ 1 0 7→ 1 1 7→ − 0 Let H denote the quaternion algebra concentrated in degree 0, i.e., H is the associative k-algebra generated by even elements i and j such that ij = ji and i2 = j2 = 1. − − Lemma A.2. We have an isomorphism of superalgebras Cl = H Clop. 3 ∼ ⊗ 1 op 1 1 1 Proof. We define H Cl1 Cl3 by i 1 2 α1α2, j 1 2 α1α3, and 1 α1 2 α1α2α3. It is readily verified⊗ to be surjective→ and⊗ both7→ algebras⊗ have7→ dimension 8 and⊗ hence7→ it is an isomorphism. Lemma A.3. op We have an isomorphism of superalgebras Cl4 ∼= Cl4 . Proof. By Lemma A.1 and A.2, we have

op 1 1 Cl = Cl Cl = H Cl Cl = H End(k | ). 4 ∼ 3 ⊗ 1 ∼ ⊗ 1 ⊗ 1 ∼ ⊗ op 1 1 1 1 op a b Next, we have H = H via i i and j j and End(k | ) = End(k | ) via ∼ 7→ − 7→ − ∼ c d 7→ a c where the matrices are given with respect to a homogeneous basis of k1 1. b− d | 8 8 Theorem A.4 (Bott periodicity). We have an isomorphism of superalgebras Cl8 = End(k | ). ∼1 1 Furthermore, if k contains a primitive 4th root of unity ζ4, then Cl2 ∼= End(k | ). Proof. By Lemma A.1 and A.3, we have

op 1 1 4 8 8 Cl = Cl Cl = Cl Cl = End(k | )⊗ = End(k | ). 8 ∼ 4 ⊗ 4 ∼ 4 ⊗ 4 ∼ ∼ The second statement follows from Lemma A.1 since we have Cl = Clop via α ζ α . 1 ∼ 1 1 7→ 4 1 m n The periodicity comes when we consider Morita equivalence. Since End(k | ) is a simple superalgebra for any m, n, its category of (super)modules is equivalent to the category of

super vector spaces, and hence we get ModClk+8 ModClk for any k and ModClk+2 ModClk when ζ k. ≃ ≃ 4 ∈ SUPERSYMMETRIC MONOIDAL CATEGORIES 59

References [BN] John C. Baez, Martin Neuchl. Higher-dimensional algebra I: braided monoidal categories. Adv. Math. 121 (1996), 196–244. arXiv:q-alg/9511013v1 [BGH+] Paul Bruillard, Cesar Galindo, Tobias Hagge, Siu-Hung Ng, Julia Yael Plavnik, Eric C. Rowell, Zhenghan Wang. Fermionic modular categories and the 16-fold way. J. Math. Phys. 58 (2017), no. 4, 041704, 31 pp. arXiv:1603.09294v3 [BE] Jonathan Brundan, Alexander P. Ellis. Monoidal supercategories. Comm. Math. Phys. 351 (2017), no. 3, 1045–1089. arXiv:1603.05928v3 [CW] Shun-Jen Cheng, Weiqiang Wang. Dualities and Representations of Lie Superalgebras. Graduate Studies in Mathematics 144, American Mathematical Society, Providence, RI, 2012. [Cr] Sjoerd E. Crans. Generalized centers of braided and sylleptic monoidal 2-categories. Adv. Math. 136(2) (1998), 183–223. [EL] Alexander P. Ellis and Aaron D. Lauda. An odd categorification of Uq(sl2). Adv. Math. 265 (2014), 169–240. arXiv:1307.7816v2 [FH] William Fulton, Joe Harris. Representation Theory: A First Course. Graduate Texts in Mathemat- ics 129, Springer-Verlag, New York, 1991. [GK] Nora Ganter, Mikhail Kapranov. Symmetric and exterior powers of categories. Transform. Groups 19 (2014), no. 1, 57–103. arXiv:1110.4753v1 [Ho] Peter Hoffman. A projective analogue of Schur’s tensor power construction. Comm. Algebra 21 (1993), no. 7, 2211–2249. [HH] P. N. Hoffman, J. F. Humphreys. Projective Representations of the Symmetric Groups. Q-functions and Shifted Tableaux. Oxford Mathematical Monographs. Oxford Science Publications. The Claren- don Press, Oxford University Press, New York, 1992. [JS1] AndréJoyal, Ross Street. Braided monoidal categories. Macquarie Math Reports 860081 (1986). Available at: http://maths.mq.edu.au/~street/JS1.pdf [JS2] A. Joyal, R. Street. Braided tensor categories. Adv. Math. 102 (1993), no. 1, 20–78. [Jo] Tadeusz Józefiak. Schur Q-functions and cohomology of isotropic Grassmannians. Math. Proc. Cam- bridge Philos. Soc. 109 (1991), no. 3, 471–478. [KKT] Seok-Jin Kang, Masaki Kashiwara, Shunsuke Tsuchioka. Quiver Hecke superalgebras. J. Reine Angew. Math. 711 (2016), 1–54. arXiv:1107.1039v2 [Ka] Mikhail Kapranov. Supergeometry in mathematics and physics. arXiv:1512.07042v2 [KV] M. M. Kapranov, V. A. Voevodsky. 2-categories and Zamolodchikov tetrahedra equations. In Proc. Symp. Pure Math. 56 Part 2 (1994), AMS, Providence, pp. 177-259. Available at https://www.math.ias.edu/vladimir/node/71 [Ke] G. M. Kelly. Basic Concepts of Enriched Category Theory. Reprints in Theory and Applications of Categories, No. 10, 2005. [Ma] I. G. Macdonald. Symmetric Functions and Hall Polynomials, second edition, Oxford Mathematical Monographs, Oxford, 1995. [Mc] S. Mac Lane. Cohomology theory of abelian groups. Proc. International Congress of Mathemati- cians, Volume II (1950), 8–14. [MP] Scott Morrison, David Penneys. Monoidal categories enriched in braided monoidal categories. Int. Math. Res. Not. IMRN (2019), no. 11, 3527–3579. arXiv:1701.00567v1 [Na] M. L. Nazarov. Young’s orthogonal form of irreducible projective representations of the symmetric group. J. London Math. Soc. (2) 42 (1990), no. 3, 437–451. [Tr] Todd Trimble, The super Brauer group and super division algebras. Available at http://math.ucr.edu/home/baez/trimble/superdivision.html [Us] Robert Usher. Fermionic 6j-symbols in superfusion categories. J. Algebra 503 (2018), 453–473. arXiv:1606.03466v1 [WW] Jinkui Wan, Weiqiang Wang. Lectures on spin representation theory of symmetric groups. Bull. Inst. Math. Acad. Sin. (N.S.) 7 (2012), no. 1, 91–164. arXiv:1110.0263v2 60 STEVENVSAMANDANDREWSNOWDEN

Department of Mathematics, University of California, San Diego, CA Email address: [email protected] URL: http://math.ucsd.edu/~ssam/ Department of Mathematics, University of Michigan, Ann Arbor, MI Email address: [email protected] URL: http://www-personal.umich.edu/~asnowden/