Arxiv:1911.02925V3 [Math.GT]

Home , Torsion (algebra)

arXiv:1911.02925v3 [math.GT] 8 Jun 2021 0struhtepoern ok fJns itn n Resh and Witten, Jones, of works pioneering the through 80’s oyoilo nt [ knots of polynomial hoy hs r h ocalled so the fiel quantum are to These topology dimensional theory. low relates that domain d.O h n ad n a en oooia nainsus invariants topological as define referred can are one These hand, one topology. the algebraic On ods. ahmtclysekn,tecntuto fqatminva quantum of construction the speaking, Mathematically o genus Seifert the [ detect circle they the over together, fibers all manifold taken non-inver and when mutation and, detect sometimes they instance, for (non-abe a [ with twisting group by damental strengthened further and be information can topological they deep contain invariants These WSIGKPREGIVRAT I O ACLSAND CALCULUS FOX VIA INVARIANTS KUPERBERG TWISTING oooia nainso nt n -aiod a ebuil be may 3-manifolds and knots of invariants Topological .Introduction 1. pedxA h o-bla case non-abelian The References products semidirect via A. calculus Appendix Fox invariants Kuperberg 5. torsion Twisted Reidemeister and manifolds 4. Sutured superalgebras 3. Hopf 2. nteohrhn,oecnbidivrat sn h ol of tools the using invariants build can one hand, other the On oooyoinain eso htteeivrat r co are invariants these if that that, show and We calculus orientation. homology a opimo h udmna ru noAut( into 3-man group sutured fundamental balanced the of of morphism invariants topological are These nepaaino u o aclsfrua ntrso part a of terms in formulas calculus Fox sut our of of torsion explanation Reidemeister an relative twisted the of finement etpouto nivltr ofsuperalgebra Hopf involutory an of product rect Abstract. nains When invariants. 25 esuyKpregivrat o uue aiod ntec the in manifolds sutured for invariants Kuperberg study We .Tersligtitdivrat unott eextremely be to out turn invariants twisted resulting The ]. 1 n,mr eeal,teRieese oso f3-manifo of torsion Reidemeister the generally, more and, ] H H sa xeiragba eso htti nain speciali invariant this that show we algebra, exterior an is is EDMITRTORSION REIDEMEISTER N gae,te a eetne nacnnclwyt polynomia to way canonical a in extended be can they -graded, 10 AILL DANIEL , unu invariants quantum 11 1. ]. Introduction Contents PZNEUMANN OPEZ ´ lsia invariants classical 1 H n osbywt Spin a with possibly and ) H ihisatmrhs ru Aut( group automorphism its with hc eeitoue uigthe during introduced were which , rd3mnfls eas give also We 3-manifolds. ured flsedwdwt homo- a with endowed ifolds siiitdb i nte90’s, the in Lin by initiated as , iiiyo oekos[ knots some of tibility in ersnaino h fun- the of representation lian) clrHp group-algebra. Hopf icular ptdvaafr fFox of form a via mputed insrle nteter of theory the on relies riants tki-uav[ etikhin-Turaev ,esnily rmtometh- two from essentially, t, hoyadrepresentation and theory d ntadwehra3- a whether and knot a f n nld h Alexander the include and n h ol fclassical of tools the ing unu topology quantum s fasemidi- a of ase c tutr and structure e oare- a to zes 14 , H 35 19 powerful, ). l d [ lds , vast a , , 36 20 , , 33 46 45 34 33 27 18 10 ]. ]. 6 1 ] 2 DANIEL LOPEZ´ NEUMANN

(braided) monoidal categories, Hopf algebras and, in particular, quantum groups. For instance, the Jones polynomial of links and the Witten-Reshetikhin-Turaev invariants (WRT) of closed 3-manifolds are built through the monoidal category of representations of the quantum group Uq(sl2) [39]. One can also build invariants of closed 3-manifolds directly from an arbitrary finite dimensional Hopf algebra H, through the theory of Hopf algebra (co)integrals. This method was introduced by Kuperberg [21, 22] and Hennings [13], in different settings. These two approaches are now known to be essentially equivalent [4] and they relate to WRT when H = Uq(sl2) at a root of unity [5]. There exists also a theory of quantum invariants of pairs (M, ρ), where M is either a closed 3-manifold or a link complement and ρ is a homomorphism of π1(M) into some group G. These are obtained by extending the previous methods to monoidal categories or Hopf algebras graded by G [40, 42]. It turns out that the above two families of invariants are not disjoint, even though they are built from very different methods. More precisely, some classical invariants can be real- ized as quantum invariants, in general through the representation theory of an appropiate Hopf algebra H. For instance, the Alexander polynomial of links in the three-sphere can be obtained through such methods if H = Uq(gl(1|1)) [34, 37] or Uq(sl2) at q = i [28]. The abelian Reidemeister torsion of closed 3-manifolds has been shown to be obtained via an “unrolled” version of Uq(sl2) at q = i [3]. More recently, a special instance of the SL(2, C)-twisted torsion of the complement of a link L ⊂ S3 has been obtained through the representation theory of a graded object associated to Uq(sl2) (in its unrestricted version) [27]. In another direction, the author showed that the abelian relative torsion of balanced sutured 3-manifolds can be obtained through a generalization of the Hopf algebraic approach of Kuperberg, specialized to the Borel part of Uq(gl(1|1)) [26]. However, neither of these works capture the more general aspects of Reidemeister torsion theory, such as GL(n, C)-twisted Alexander polynomials, torsion of links in arbitrary homology spheres, etc. Understanding Reidemeister torsion as part of quantum topology seems an important issue in order to find topological applications of quantum invariants, which often are weaker than their classical counterparts. In this paper, we take a step in this direction by showing that several aspects of Rei- demeister torsion theory, such as Fox calculus, Spinc refinements and twisted polynomials, are general Hopf algebra constructions and therefore belong to the realm of quantum topology. The twisted relative Reidemeister torsion of balanced sutured 3-manifolds, hence also twisted Alexander polynomials of links, is shown to be a special case of such construction. We achieve this through our sutured manifold extension of involutory Kuperberg invariants [26] restricted to a semidirect product K[Aut(H)] ⋉ H. The key idea is to consider this semidirect product relative to K[Aut(H)], or equivalently, as a Hopf group-algebra graded by Aut(H). Thus, our results indicate that Reidemeister torsion naturally belongs to the world of homotopy quantum field theory (HQFT) of Turaev [40, 42].

Background. To describe our results in detail we recall a few notions and previous work. First, recall that a balanced sutured 3-manifold is a pair (M, γ) where M is a 3-manifold with non-empty boundary and γ is a collection of annuli in ∂M dividing the boundary into two homeomorphic pieces R−(γ) and R+(γ) [12]. For instance, the complement of a link in an arbitrary closed 3-manifold can be considered as a sutured manifold by letting γ consist on two meridians on each boundary component. A fundamental topological invariant of TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 3

ρ balanced sutured manifolds is the twisted relative Reidemeister torsion τ (M,R−(γ)) ∈ K, which depends on (M, γ) together with a homomorphism ρ : π1(M) → GL(V ) for some finite dimensional vector space V over a field K [41]. If h : π1(M) → H1(M) is ρ⊗h the Hurewicz map, one can extend the torsion to τ (M,R−(γ)) ∈ K[H1(M)]. This extension generalizes the twisted Alexander polynomials of links and for ρ ≡ 1, it is the Euler characteristic of a Floer homology invariant of sutured manifolds [8]. The torsion is actually defined up to a ± det ρ(g) indeterminacy with g ∈ π1(M), but this can be corrected by picking a Spinc structure s and a homology orientation ω (i.e. an orientation of the vector space H∗(M,R−(γ); R)) [8, 41] so it has the form ρ τ (M,R−(γ), s,ω) ∈ K. Recall also that given an arbitrary finite dimensional Hopf (super)algebra H over a K Kup K field , Kuperberg defines a topological invariant IH (Y,f) ∈ of a closed oriented 3-manifold Y endowed with a framing f of its tangent bundle [22]. This relies on the Heegaard diagrammatic presentation of closed 3-manifolds along with the theory of Hopf algebra (co)integrals (e.g. [32]). When H is the Borel of Uq(sl2) at a root of unity, this invariant is related to WRT invariants [4, 5]. However, the appearance of framings makes 2 the computation of this invariant quite challenging. If H is involutory, i.e. S = idH where S is the antipode of H, Kuperberg’s invariant is independent of the framing [21] hence we denote it by Kup K IH (Y ) ∈ . The involutory invariant admits an extension to pairs (Y, ρ), where Y is a closed 3-manifold and ρ : π1(Y ) → G is a homomorphism into some group G [44]. This relies on a finite-type involutory Hopf G-coalgebra H = {Hα}α∈G (which amounts to a Hopf algebra graded by G) [40, 43] and is denoted ρ K IH (Y ) ∈ . This specializes to IKup(Y ) if ρ is trivial. It has to be noted that the assumptions of [21] and H1 [44] imply semisimplicity of the relevant Hopf algebras [23]. Recently, a candidate for an unframed version of Kuperberg’s invariant has been proposed only assuming unimodularity of H [6]. However, neither of these unframed approaches can be directly related to WRT, since the Borel of Uq(sl2) at a root of unity is non-involutory and non-unimodular. Now, in [26], the author generalized Kuperberg’s involutory invariant to balanced sutured 3-manifolds. This construction relied on sutured Heegaard diagrams [15] along with relative versions of the Hopf algebra (co)integrals. More precisely, the algebraic input consisted essentially of an involutory Hopf superalgebra H (possibly of infinite dimension) relative to a pair of Hopf subalgebras A, B ⊂ H, where A is the domain of the cointegral and B is the target of the integral. When B = K the outpute is an invariant Iρe(M,γ, s,ω) ∈ K He where (M, γ) is a balanced sutured 3-manifold, ρ : π (M) → G(A) is a homomorphism e 1 into the group-likes of A, s ∈ Spinc(M, γ) and ω is a homology orientation. In contrast to the above unframed approaches [6, 21, 44], here H may be non-unimodular and the appearance of s and ω is related to this, indeed, Iρ depends on s,ω up to ±a∗(ρ(g)) where He ∗ × e a : G(A) → K is the distinguished group-like of H rel A and g ∈ π1(M). When H is the e ρ Borel of Uq(gl(1|1)) with an appropriate relative integral, we showed that I is a refinement He e e e 4 DANIEL LOPEZ´ NEUMANN

ρ × of the abelian torsion τ (M,R−(γ)), i.e. ρ : H1(M) → K , and we did this by relating the coproduct of H to Fox calculus.

Main results.e In the present work, we show that a semidirect product H = K[Aut(H)]⋉H ∗ × fits into the setting of [26] with A = K[Aut(H)] and a = rH : Aut(H) → K (where rH ρ K is as in Definition 2.3) and we study the resulting invariant IK[Aut(H)]⋉eH (M,γ, s,ω) ∈ , where ρ : π1(M) → Aut(H) is a homomorphism and s,ω are as above. We show that when the definining formula of this invariant is rewritten only in termse of the structure tensors of H and the homomorphism ρ, then one gets a formula very similar to the original one of Kuperberg [21] but in which ρ twists the structure tensors via Fox calculus. Since the ρ algebraic input only depends on H, we denote this formula by IH (M,γ, s,ω) and call it a ρ twisted Kuperberg invariant. We develop the formula for IH independently of [26] in Section 4. Thus, we can state our first main result as follows.

Theorem 1. Let H be a finite dimensional involutory Hopf superalgebra over a field K with a two-sided cointegral and integral. Let (M, γ) be a balanced sutured 3-manifold, c ρ : π1(M) → Aut(H) a group homomorphism, s ∈ Spin (M, γ) and ω an orientation of H (M,R (γ); R). Then the invariant Iρ at a semidirect product coincides with the Fox ∗ − He ρ calculus invariant IH : e ρ ρ IK[Aut(H)]⋉H (M,γ, s,ω)= IH (M,γ, s,ω). ρ We also provide an explanatione of the Fox calculus formula that defines IH in terms of Hopf group-algebras (the dual notion of a Hopf group-coalgebra). We show that the semidirect product K[Aut(H)] ⋉ H determines a Hopf Aut(H)-algebra H for which ρ ρ IH (Y )= IH (M0, γ0), where the left hand side is the (dual version of the) invariant of Virelizier [44] and the right 3 3 hand side is ours for M0 = Y \ B , where B is an open 3-ball embedded in Y and γ0 is a single suture in ∂M0, see Proposition 5.2. Moreover, we relate the relative integral approach of [26] to the Hopf group-(co)algebra approach. Therefore, our construction can also be understood as a generalization to sutured manifolds of [44] restricted to a particular Hopf group-algebra (which in our case may be non-semisimple). In addition, the Hopf group- algebra approach clarifies the appearance of Spinc structures and homology orientations, see Remark 5.3. Our construction has the additional feature that it can be extended to define polynomial invariants provided H is N-graded. Indeed, in such a case, any ρ : π1(M) → Aut(H) can be combined with the Hurewicz map h : π1(M) → H1(M; Z) to define a homomorphism ρ ⊗ h : π1(M) → Aut(H ⊗K K[H1(M)]). Hence, there is an invariant ρ⊗h K IH (M,γ, s,ω) ∈ [H1(M)] ρ which we call a twisted Kuperberg polynomial and this specializes to IH (M,γ, s,ω) via the augmentation map aug : K[H1(M)] → K. In particular, this procedure defines (Hopf algebraic) twisted multivariable polynomial invariants of links in arbitrary homology spheres. When ρ ≡ 1, this link invariant can be considered as a “polynomial deformation” of the Kuperberg invariant of the underlying closed 3-manifold, see Corollary 4.11. TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 5

In our second main theorem we show that all the above procedures generalize existing ones in Reidemeister torsion theory. Indeed, let Λ(V ) be the exterior algebra of an n- dimensional vector space V over a field K. This is an N-graded involutory Hopf superalgebra with Aut(Λ(V )) =∼ GL(V ) and the distinguished group-like of the associated semidirect × product is rΛ(V ) = det : GL(V ) → K . c Theorem 2. For an arbitrary (M, γ), ρ : π1(M) → GL(V ), s ∈ Spin (M, γ) and ω as above we have ρ ρ−T IΛ(V )(M,γ, s,ω)= τ0 (M,γ, s,ω) ρ−T where τ0 denotes the refinement of the twisted Reidemeister torsion τ (M,R−(γ)) of Subsection 3.7 1. Similarly, we have − ρ⊗h ρ T ⊗h K IΛ(V )(M,γ, s,ω)= σ(τ0 (M,γ, s,ω)) ∈ [H1(M)] −1 where σ : K[H1(M)] → K[H1(M)] is the K-linear map characterized by σ(f) = f ,f ∈ H1(M). This theorem generalizes to the non-abelian setting the result of [26] mentioned above. It turns out that there is a much simpler proof of the general case, relying on the universal property of the Hopf superalgebra Λ(V ). It has to be noted that the dual of the Borel of Uq(gl(1|1)) at a root of unity of order n considered in [26] is a semidirect product K[Z/nZ] ⋉ Λ(K) as in the present work (where Z/nZ acts over K by multiplication by an n-th root of unity). For link complements, the right hand side is equivalent to the twisted Alexander polynomials (see Corollary 3.6), hence we get: Corollary 1. Let L be an ordered, oriented m-component link in an homology sphere Y and ρ : π1(ML) → GL(V ) a homomorphism, where ML = Y \ L. Then the twisted Kuperberg polynomial of L at an exterior algebra is equivalent to the twisted Alexander polynomial: m ∗ −T det(t ρ(a ) − I ) σ(Iρ ⊗h(M , γ , s,ω)) ˙= i=1 i i n · ∆ρ (t ,...,t ) ∈ K[t±1,...,t±1] Λ(V ) L L ∆ρ L 1 m 1 m Q L,0 ρ where ∆L,0 is the 0-th twisted Alexander polynomial of L (which is always non-zero). Here n1 nm ˙= stands for equality up to multiplication by ± det(ρ(g)) · t1 · · · tm for some g ∈ π1(ML) and ni ∈ Z, and σ is defined as above. Future directions. As mentioned above, Kuperberg invariants of closed 3-manifolds can be defined out of an arbitrary finite dimensional Hopf algebra, but without the involutory condition, the 3-manifolds need to be framed. Finding an involutory Hopf algebra for which our extension of Kuperberg invariants gives something different than Reidemeister torsion seems unlikely (see Subsection 4.5 for concrete reasons). Thus, our results should be generalized to non-involutory Hopf algebras. On the other hand, one knows from the works of Kirk-Livingston [20] and Friedl-Vidussi [10,11] that twisted Alexander polynomials are an extremely powerful invariant and contain deep topological information. It would be interesting to see whether some of these results extend to our twisted Kuperberg polynomials for other (N-graded) Hopf algebras. Of course, for this to capture topological information beyond torsion one should use non-involutory

1The right hand side is computed at the inverse-transpose because we follow the conventions of [9] for the torsion. 6 DANIEL LOPEZ´ NEUMANN

Hopf algebras. More generally, it would be interesting to study possible topological applications of quantum invariants obtained through group-graded objects (Hopf group-coalgebras, modular G-categories, etc), and ultimately, of Turaev’s homotopy quantum field theories. Structure of the paper. Sections 2 and 3 consist of background material on Hopf (super)algebras, sutured manifolds and (twisted) Reidemeister torsion. In Section 4 we give ρ the Fox calculus formula for IH (M,γ, s,ω) and study some of its properties, notably, that it extends to K[H1(M)] for N-graded Hopf superalgebras and that it recovers our refinement of torsion when H is an exterior algebra (Theorem 2). This section is independent of our previous work [26]. In Section 5 we prove Theorem 1 and relate our construction to the Hopf group-algebra approach of [44]. Finally, we devote to an Appendix some (rather trivial) technical details concerning the case of non-abelian ρ, which was not considered in [26]. Acknowledgments. I would like to thank my PhD supervisor, Christian Blanchet, for all his support and suggestions during the course of my PhD. I would also like to thank Anna Beliakova, Andrés Fontalvo Orozco, Krzysztof Putyra and Alexis Virelizier for many interesting conversations. Finally, I would like to thank the referee for valuable comments. This project has received funding from the European Union’s Horizon 2020 Research and Innovation Programme under the Marie Sklodowska-Curie Grant Agreement No. 665850.

2. Hopf superalgebras In this section, we recall the notions of the theory of Hopf (super)algebras we need. For more details, see [32]. In what follows we assume all vector spaces are deﬁned over a ﬁeld K. 2.1. Basic notions and notation. By a super-vector space, we mean a vector space V endowed with a direct sum decomposition V = V0 ⊕ V1. A vector v ∈ V is said to be homogeneous if v ∈ V0 or v ∈ V1. If v ∈ Vi, we say that v has degree i, denoted |v| = i (mod 2). If V,W are super-vector spaces, then V ⊗ W is a super-vector space with

(V ⊗ W )i := Vj ⊗ Wi−j j M where the indices are taken mod 2. The category of super-vector spaces and their degree zero linear maps forms a symmetric monoidal category with symmetry map

τV,W : V ⊗ W → W ⊗ V v ⊗ w 7→ (−1)|v||w|w ⊗ v.

We will use tensor network notation for the morphisms of this category (see [21, 22]), that is, we denote the tensor factors of the domain of a linear map as incoming arrows and those of the target as outcoming arrows with the convention that the left to right direction in a tensor V1 ⊗ V2 ⊗ . . . ⊗ Vn corresponds to the top to bottom direction in tensor network notation. Moreover, since the (super)vector spaces will usually be clear from the context, we drop them from the notation. For instance, a tensor T : V ⊗ V → W ⊗ W ⊗ W is denoted by

T TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 7 where the rightmost arrows correspond, from top to bottom, to the tensor factors of W ⊗ W ⊗ W read from left to right (and similarly for the leftmost arrows). A vector v ∈ V , a ∗ ∗ covector v ∈ V and the identity map idV : V → V are denoted by v ,v∗ , .

Composition of tensors is denoted by joining the corresponding outcoming/incoming arrows, the tensor product is denoted by stacking one ﬁgure over another and the symmetry τV,W of the category is denoted by a crossing pair of arrows. For instance, if T1 : V → V ⊗ V and T2 : V ⊗ V → V then (T2 ⊗ idV ) ◦ (idV ⊗ T1), T1 ⊗ T2 and τV,V ◦ T1 are respectively denoted by

T1 T2 , , T1 . T1 T2

2.2. Hopf superalgebras. A Hopf superalgebra is a super-vector space H endowed with degree zero tensors

m ,,1 ∆ ,ǫ , S where each arrow corresponds to H (that is, m : H ⊗ H → H, 1 ∈ H, etc). These tensors satisfy the usual Hopf algebra axioms, except that the algebra property for the coproduct ∆ involves the symmetry τH,H of super-vector spaces (denoted by a crossing pair of arrows): ∆ m m ∆ = ∆ m .

Though we mostly use tensor network notation, in some places we also use Sweedler’s notation for the coproduct, that is, we write ∆(h)= h(1) ⊗ h(2) (omitting the sum sign) for any h ∈ H. Any (ungraded) Hopf algebra can be seen as a Hopf superalgebra concentrated in degree zero. In what follows we reserve the term Hopf algebra exclusively for the ungraded case. If H is a Hopf superalgebra, then we deﬁne Hop = (H,mop, 1, ∆,ǫ,S−1) and Hcop = op −1 op op ∗ (H,m, 1, ∆ ,ǫ,S ) where m = m ◦ τH,H and ∆ = τH,H ◦ ∆. The dual H is also a Hopf superalgebra in the usual way. We denote by Aut(H) the group of Hopf superalgebra automorphisms of H. op A Hopf superalgebra is said to be involutory if S ◦S = idH . Itis commutative if m = m and it is cocommutative if ∆=∆op. A commutative or cocommutative Hopf superalgebra is always involutory, see e.g. [32, Corollary 7.1.11]. A Hopf superalgebra H is N-graded if it has a direct sum decomposition H = ⊕n∈NHn such that

m(Hi ⊗ Hj) ⊂ Hi+j, ∆(Hn) ⊂ Hi ⊗ Hj, S(Hn) ⊂ Hn, i j n +X= for each i, j, n ∈ N and such that the N-degree reﬁnes the mod 2 degree, that is, Hi = ⊕n≥0H2n+i for each i = 0, 1 (mod 2). To distinguish from mod 2 degree, we set |x|0 := n ∈ N for nonzero x ∈ Hn. 8 DANIEL LOPEZ´ NEUMANN

Example 2.1. Let V be a finite dimensional vector space. The exterior algebra Λ(V ) on ⊗n ⊗0 V is the quotient of the tensor algebra T (V ) = ⊕n≥0V (where V = K) by the ideal generated by the elements of the form v ⊗ w + w ⊗ v with v, w ∈ V . This becomes an N-graded superalgebra by letting V ⊂ Λ(V ) be in degree one and it is a Hopf superalgebra if we set ∆(v) = 1 ⊗ v + v ⊗ 1, ǫ(v) = 0, S(v)= − v for any v ∈ V and extend ∆,ǫ (resp. S) by letting them be superalgebra homomorphisms (resp. antihomomorphism). This is a commutative, cocommutative (hence involutory) Hopf superalgebra. Since Prim(Λ(V )) = V (where Prim(H)= {h ∈ H | ∆(h)= h ⊗ 1 + 1 ⊗ h}) it follows that the group of automorphisms of Λ(V ) is isomorphic to GL(V ). Given a subgroup G ⊂ Aut(H), the semidirect-product K[G] ⋉ H is a Hopf superalgebra defined as follows. As a super-coalgebra, it is the tensor product K[G] ⊗ H, where K[G] is the group-algebra of G (this is a Hopf algebra with ∆(α)= α⊗α, ǫ(α)=1and S(α)= α−1 for each α ∈ G). Its product is defined by −1 (α1 ⊗ h1) · (α2 ⊗ h2) := α1α2 ⊗ α2 (h1)h2, −1 −1 and the antipode by S(α1 ⊗ h1) := α1 ⊗ α1 (SH (h1)) for any α1, α2 ∈ G and h1, h2 ∈ H. It has to be noted that for semisimple H (i.e. semisimple as a K-algebra), Aut(H) is a finite group by [31], hence K[Aut(H)] ⋉ H is finite-dimensional. However, for non-semisimple H, K[Aut(H)] ⋉ H is infinite-dimensional in general (for instance if H = Λ(V )).

2.3. Integrals and cointegrals. Let H = (H,m, 1, ∆,ǫ,S) be a ﬁnite dimensional Hopf superalgebra over a ﬁeld K. A right cointegral in H is an element cr ∈ H such that

cr m = ǫ cr .

A left cointegral of H is deﬁned as a right cointegral of Hop. A right integral is an element ∗ µr ∈ H such that

µr ∆ = µr 1 .

Equivalently, a right integral over H is a right cointegral of the dual Hopf superalgebra H∗. If H is finite dimensional then there is a non-zero right cointegral and it is unique up to scalar [32, Theorem 10.2.2], the same holds for left or right integrals as well. Remark 2.2. It is not always true that left (co)integrals are also right (co)integrals. If H is an (ungraded) Hopf algebra, then one says that H is unimodular if this is the case, but this definition is not completely appropriate in the super-case. Indeed, one can define unimodularity for finite tensor categories, e.g. the category of representations of a Hopf (super)algebra (see [7, Definition 6.5.7]). Then one shows that for Hopf algebras, unimodularity of Rep(H) is equivalent to the cointegral being two-sided. Nevertheless, for Hopf superalgebras, unimodularity of its category of representations is equivalent to the cointegral being two-sided and of mod 2 degree zero. For instance, an exterior algebra over an odd dimensional vector space is non-unimodular, even though cointegrals are two-sided. TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 9

Now let cr be a non-zero right cointegral in H and let α ∈ Aut(H). Clearly, α(cr) is a right cointegral so by uniqueness, there is a scalar λ ∈ K× (depending on α) such that α(cr) = λcr. Similarly, if µr is a right integral on H, uniqueness of integrals implies that ′ ′ K× µr ◦ α = λ µr for some λ ∈ . Since µr(cr) 6= 0 [32, Theorem 10.2.2, b)] it follows that λ′ = λ.

Deﬁnition 2.3. For any α ∈ Aut(H), we denote rH (α) := λ, where λ is the scalar deﬁned × above. This is a group homomorphism rH : Aut(H) → K .

Note that a Hopf algebra is semisimple if and only if ǫ(cr) 6= 0 by [32, Theorem 10.3.2] and so rH ≡ 1 in this case. Hence, the homomorphism rH is relevant only for non-semisimple H. Example 2.4. Let Λ(V ) be the exterior algebra on a ﬁnite dimensional vector space V , recall that Aut(Λ(V )) = GL(V ). If X1,...,Xn is a basis of V , then the (two-sided) cointegral of Λ(V ) is the product c = X1 ...Xn so that

α(c)= α(X1) ...α(Xn) = det(α)X1 ...Xn = det(α)c K× for any α ∈ GL(V ). Thus, rΛ(V ) : GL(V ) → is just the determinant. Now suppose H is N-graded and R is a commutative K-algebra with unit (we do not suppose R is a domain). Consider the Hopf superalgebra HR := H ⊗K R over R. Given an element α ∈ Aut(H) and f ∈ R we deﬁne a R-linear Hopf endomorphism α ⊗ f of HR by (1) α ⊗ f(h ⊗ f ′) := α(h) ⊗ (f |h|0 · f ′) ′ × where h ∈ H is homogeneous and f ∈ R. This is an automorphism of HR if f ∈ R . Now let c, µ be respectively a right cointegral and integral for H. Then cR := c ⊗ 1 and µR := µ ⊗ idR are respectively a cointegral and integral for HR respectively. If α ⊗ f is deﬁned as above one has α ⊗ f(c ⊗ 1) = α(c) ⊗ f |c|0

|c|0 = rH (α)c ⊗ f

|c|0 = rH (α)f (c ⊗ 1). × This implies that the homomorphism rHR : Aut(HR) → R satisﬁes

|c|0 (2) rHR (α ⊗ f) := rH (α) · f . 2.4. Hopf G-algebras. Let G be a group. A Hopf G-algebra is a family of coalgebras H = {(Hα, ∆α,ǫα)}α∈G indexed by G endowed with coalgebra morphisms

mα1,α2 : Hα1 ⊗ Hα2 → Hα1α2

− for each α1, α2 ∈ G, a unit 1 ∈ H1G and maps Sα : Hα → Hα 1 for each α ∈ G satisfying graded versions of the associativity, unitality and antipode axioms (see [43] or [40, 42] for more details in the dual setting). We employ similar tensor network notation for Hopf G-algebras, the G-labels only appearing as subscripts of the structure maps. Note that − H1G is a Hopf algebra in the usual sense. We say that H is involutory if Sα 1 Sα = idHα for each α ∈ G. We say that H is of ﬁnite type if each Hα is ﬁnite dimensional. These notions extend in an obvious way to super-vector spaces.

Let H = {Hα}α∈G be a ﬁnite type Hopf G-algebra. Since the dual of a Hopf G-algebra is a Hopf G-coalgebra, the existence and uniqueness theorems of integrals of [43] have 10 DANIEL LOPEZ´ NEUMANN analogous statements in the G-algebra case. Thus there exists a unique right cointegral of H (up to scalar), that is, a family c = {cα}α∈G, where cα ∈ Hα for each α ∈ G, satisfying

cα1

mα1α2 = ǫα2 cα1α2 .

∗ ∗ ∗ ∗ Moreover, there exists a unique family g = {gα}α∈G where gα ∈ Hα satisfying

= ∗ c mα1α2 gα1 α1α2 cα2

∗ ∗ ∗ g∗ and gα1α2 ◦ mα1α2 = gα1 ⊗ gα2 for each α1, α2 ∈ G. We call the comodulus of H.

Example 2.5. Let H be a ﬁnite dimensional Hopf (super)algebra and let G = Aut(H). Consider the semidirect product Hopf (super)algebra K[G] ⋉ H. Let

Hα := {h · α | h ∈ H}⊂ K[G] ⋉ H for each α ∈ G. Then H := {Hα}α∈G is a Hopf G-(super)algebra with the structure morphisms induced from the semidirect product. A right cointegral is given by cα := c · α ∈ Hα where c is a right cointegral of H. The comodulus of H is given by ∗ ∗ gα(h · α) := rH (α)g (h) ∗ ∗ for h ∈ H, α ∈ G, where g ∈ H is the comodulus of H and rH is the homomorphism of Deﬁnition 2.3. This example is dual to the Hopf group-coalgebra of [42, Subsection 1.2].

3. Sutured manifolds and Reidemeister torsion In this section we recall a few concepts from the theory of sutured manifolds [12, 15, 17] and twisted Reidemeister torsion [9, 39] as well as some considerations from our previous work [26]. In what follows, all 3-manifolds will be assumed to be compact and oriented. Homology will always be taken with integral coeﬃcients, unless speciﬁed otherwise.

3.1. Sutured manifolds. A sutured manifold is a (compact, oriented) 3-manifold-with- boundary M endowed with a collection γ of pairwise disjoint annuli in ∂M subject to the following properties: (1) Each annuli is the closed tubular neighborhood of an oriented simple closed curve called a suture. The collection of sutures is denoted by s(γ). (2) The surface R := ∂M \ int (γ) is oriented and each (oriented) component of ∂R is oriented-parallel to some suture.

We denote by R+(γ) (resp. R−(γ)) the union of the components of R whose orientation coincide (resp. is opposite) with the induced orientation on ∂M. We say that (M, γ) is balanced if M has no closed components, each component of ∂M has at least one suture and χ(R−(γ)) = χ(R+(γ)). Example 3.1. Let Y be a connected closed oriented 3-manifold. Let B3 ⊂ Y be an 3 embedded open 3-ball. Then M0 = Y \ B is a balanced sutured 3-manifold if we let γ0 be a single annulus in ∂M0. The subsurfaces R±(γ0) are both disks in this case. TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 11

Example 3.2. Let L be a link in a closed oriented 3-manifold Y . Let N(L) be an open tubular neighborhood of L in Y . Then ML = Y \ N(L) is a balanced sutured 3-manifold if we let γ = γL consists of a pair of oppositely oriented meridians, one pair for each component of ∂ML. For each component T ⊂ ∂ML, T \ (T ∩ γL) consists of two annuli, one in R−(γL) and the other in R+(γL). 3.2. Extended Heegaard diagrams. Let (M, γ) be a sutured 3-manifold. An (embedded) sutured Heegaard diagram is a tuple H = (Σ, α, β) consisting of the following data: (1) A compact oriented surface-with-boundary Σ embedded in M with ∂Σ = s(γ) as oriented 1-manifolds. (2) A set α (resp. β) of pairwise disjoint simple closed curves in Σ bounding disks to the negative (resp. positive) side of Σ.

We will usually identify the set α with the 1-submanifold ∪α∈αα ⊂ Σ, and similarly for β. We assume that α and β are transverse submanifolds of Σ. We require that the surface Σ[α](resp. Σ[β]) obtained by compressing Σ along the disks corresponding to α (resp. β) results in a surface isotopic to R−(γ) (resp. R+(γ)) relative to γ. In other words, Σ splits M into two handlebodies Uα and Uβ, where Uα is obtained by attaching one handles to R− × [−1, 1] along R− × {1} with belt circles the curves in α while Uβ is obtained by attaching one handles to R+ × [−1, 1] along R+ × {−1} with belt circles the curves in β. We say that H is balanced if |α| = |β| and every component of Σ \ α and Σ \ β contains a component of ∂Σ. From now on, we will refer to (balanced) embedded sutured Heegaard diagrams only as (balanced) Heegaard diagrams. An extended Heegaard diagram of (M, γ) is a Heegaard diagram (Σ, α, β) endowed with a set a of pairwise disjoint properly embedded arcs in Σ \ α that forms a cut system of the surface Σ[α], that is, for any component R′ of Σ[α], R′ \ R′ ∩ N(a) is homeomorphic to a disk, where N(a) is an open tubular neighborhood of a in Σ[α]. We denote αe = α ∪ a and if |α| = d, |a| = l, then we will denote α = {α1,...,αd} and a = {αd+1,...,αd+l}. As is well-known, any sutured manifold (M, γ) has a Heegaard diagram and (M, γ) is balanced if and only if any (and hence all) of its Heegaard diagrams is balanced [15]. The Reidemeister-Singer theorem extends to sutured manifolds: any two Heegaard diagrams of a given sutured manifold are related by Heegaard moves [15], see also [17, Proposition 2.36] for a precise version in the embedded case. A Heegaard diagram can always be extended as above and any two extended Heegaard diagrams of (M, γ) are related by extended Heegaard moves [26]: usual Heegaard moves of (Σ, α, β) with the condition that the α’s are isotoped or handleslided in the complement of the arcs in a, isotopies of arcs, and sliding an arc over a curve in α or over another arc. Note that the deﬁnition of extended Heegaard diagram of [26] includes also a cut system of Σ[β], we do not require this in the present paper.

3.3. A presentation of π1. Suppose now that R−(γ) is connected. We describe a pre- e sentation of π1(M) out of an extended Heegaard diagram H = (Σ, α , β) of (M, γ) (see also [8, Section 4.1]). We assume M has a basepoint p ∈ s(γ) (note that this is ﬁxed when doing Heegaard moves since our diagrams have ﬁxed boundary ∂Σ= s(γ)). Let αe = α∪a where α = {α1,...,αd} are the closed curves and a = {αd+1,...,αd+l} are the arcs. Write M = Uα ∪ Uβ where Uα,Uβ are as above and think of Uα as constructed from R− × [−1, 1] by attaching 3-dimensional 1-handles with belt circles the closed α curves. The cocores of these 1-handles are disks D1,...,Dd with ∂Di = αi for each i = 1,...,d. We also have disks Di = αi × [−1, 1] ⊂ R− × [−1, 1] for each i = d + 1,...,d + l. Since we supposed 12 DANIEL LOPEZ´ NEUMANN

a∗ p p b b

b β a a

α∗

α α Figure 1. Left: an oriented, β-based, extended Heegaard diagram of the (sutured) complement of the left trefoil knot K ⊂ S3. The square is assumed to be oriented towards the reader and opposite sides have to be identiﬁed. ∗ ∗ 3 Right: the corresponding dual curves α , a ∈ π1(S \ K,p) (drawn over Σ in blue).

R−(γ) is connected, the complement of a sufficiently small neighborhood of D1 ∪···∪ Dd+l in Uα is homeomorphic to a single 3-ball. This implies that for each i = 1,...,d + l there is ∗ a unique unoriented loop αi based at p contained in Uα that intersects Di in a single point e ∗ and is disjoint from Dj for j 6= i. If each curve and arc in α is oriented, then we orient αi ∗ ∗ so that αi · αi = +1 in Σ (when αi is homotoped to lie in Σ), see Figure 1 (right). From ∗ now on, we consider αi as an element of π1(M,p). Suppose H is oriented, that is, all the curves and arcs are oriented, and β-based, meaning that each curve in β has a basepoint in Σ \ αe. For each β ∈ β let β be the word in the ∗ ∗ ∗ mx generators α1,...,αd+l defined as follows: for each crossing x of β write (αx) where e αx ∈ α is the curve or arc passing through that point. Then multiply these generators from left to right as we follow the orientation of β starting from its basepoint. Then the elements α∗(α ∈ αe) and the words β(β ∈ β) define a presentation of the fundamental group of M: ∗ ∗ (3) π1(M,p)= hα1,...,αd+l | β1,..., βdi. 3 For instance, the diagram of Figure 1 (left) determines the presentation π1(S \ K,p) = hα, a | aαa−1α−1a−1αi where we denote α∗, a∗ just by α, a for simplicity.

3.4. Spinc structures and multipoints. Let (M, γ) be a balanced sutured 3-manifold. Fix a vector field v0 on ∂M that points into M (resp. out of M) along R− (resp. R+) and along γ it is equivalent to the gradient of some height function γ = s(γ) × [−1, 1] → [−1, 1]. Let v, w be two nowhere-vanishing vector fields on M such that v|∂M = v0 = w|∂M . We say that v and w are homologous if there is an embedded open 3-ball B ⊂ int (M) such that v, w are homotopic rel ∂M in M \ B through nowhere-vanishing vector fields. A Spinc-structure is an homology class of such vector fields. The set of Spinc structures on M is denoted c c by Spin (M, γ). The obstruction to homotope a Spin structure s1 onto another s2 lies in 2 2 c H (M,∂M), we denote it by s1 − s2 ∈ H (M,∂M). Hence Spin (M, γ) is an affine space over H2(M,∂M), see [15] for more details. TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 13

Now let (Σ, α, β) be a balanced Heegaard diagram of (M, γ) and let d = |α| = |β|. Then Spinc structures on (M, γ) can be encoded in a very simple way: there is a map c s : Tα ∩ Tβ → Spin (M, γ) where Tα ∩Tβ is the set of multipoints of (Σ, α, β), i.e. unordered d-tuples x = {x1,...,xd} with xi ∈ ασ(i) ∩ βi for each i where σ is some permutation in Sd. We refer the reader to [15] for the deﬁnition of this map (which is a sutured extension of the map of [29] in the closed case). The only thing we will need from this map is that (4) s(y) − s(x)= P D[ǫ(y, x)] where P D denotes Poincar´eduality and ǫ(y, x) is an homology class which is easily de- scribed through an extended Heegaard diagram of (M, γ) (assuming R−(γ) connected). To do this, we put basepoints on the β curves of our diagram as in [26].

Deﬁnition 3.3. Let H be an oriented Heegaard diagram and let x ∈ Tα∩Tβ be a multipoint in H, say x = {x1,...,xd} where xi ∈ ασ(i) ∩ βi for each i = 1,...,d and some σ ∈ Sd. For each i, we let qi = qi(x) ∈ βi be a basepoint deﬁned as follows: if the crossing xi is positive (resp. negative), then qi lies just before xi (resp. after xi) when following the orientation of βi, see Figure 2.

b qi bc ασ(i) bc ασ(i) b qi

βi βi

Figure 2. Basepoints on β coming from x ∈ Tα ∩ Tβ. The surface is oriented towards the reader, so that the left crossing is positive. The white dot represents xi ∈ ασ(i) ∩ βi.

So let H = (Σ, αe, β) be an oriented extended Heegaard diagram of (M, γ) and let x, y ∈ Tα ∩ Tβ. For each i = 1,...,d we let di ⊂ βi be the arc from qi(x) to qi(y). Then ∗ ±1 e there is an element di ∈ π1(M,p) obtained by multiplying the (α ) ’s (α ∈ α ) obtained from the crossings through di following its orientation. Then for any x, y ∈ Tα ∩ Tβ one has d (5) ǫ(y, x)= h(di) i Y=1 in H1(M) (in multiplicative notation) where h : π1(M) → H1(M) is the Hurewicz map [26, Lemma 4.4]. 3.5. Homology orientations. A homology orientation of a balanced sutured manifold (M, γ) is an orientation ω of the vector space H∗(M,R−(γ); R). If H = (Σ, α, β) is a balanced Heegaard diagram of (M, γ) and d = |α| = |β|, then it is shown in [8] that a homology orientation is equivalent to a sign-ordering of H, that is, an orientation of all the curves α∪β and an total order of each set α, β up to some equivalence. We refer the reader to [8] for this correspondence. Given an ordered (i.e. α and β are totally ordered), oriented 14 DANIEL LOPEZ´ NEUMANN

Heegaard diagram H we denote by o(H) the corresponding homology orientation. Note that if H∗(M,R−(γ); R) = 0 (e.g. when M is the complement of a link in a rational homology 3-sphere) there is a canonical sign-ordering of any Heegaard diagram, determined by the condition det(A) > 0 where A = (αi · βj)i,j=1,...,d is the intersection matrix of H. When Y is a closed oriented 3-manifold and (M0, γ0) is the associated sutured manifold (Example 3.1), there is also a canonical homology orientation of H∗(M0,R−(γ0)) = H2(Y ) ⊕ H1(Y ), determined by any basis of H1(Y ) followed by its Poincar´edual in H2(Y ) (see [41, Section 18]).

3.6. Twisted torsion of sutured manifolds. We now briefly introduce the twisted relative Reidemeister torsion of a sutured manifold, for more details see [9, 39]. Let X be a finite connected CW complex and Y a possibly empty subcomplex such that χ(X,Y ) = 0. Let ρ : π → GL(V ) be a representation, where V is a finite dimensional vector space over a field K and π = π1(X). The twisted Reidemeister torsion is a topological invariant τ ρ(X,Y ) ∈ K of (X, Y, ρ) defined up to multiplication by ± det ρ(g), g ∈ π. Briefly, this is defined as follows: consider the universal covering q : X → X of X and let Y ′ = q−1(Y ). ′ Then the cellular complex C∗(X,Y ) is a complex of free left Z[π]-modules, which we consider as right Z[π]-modules via c · g := g−1(c) for each ge∈ π and cell c of X. Now consider ρ ′ Z the complex C∗ (X,Y ) := C∗(X,Ye ) ⊗Z[π] V where V is a left [π]-module via ρ. If we choose a lift to X of each cell of X \ Y , then we get a preferred basis ofe this complex by ρ tensoring these lifts with an arbitrarye basis of V . If the complex C∗ (X,Y ) is not acyclic we set τ ρ(X,Y ) =e 0 and if it is acyclic, we define τ ρ(X,Y ) as the algebraic torsion of this based complex, see [39]. The ± det ρ(g) indeterminacy of the torsion comes from the choice of lifts as well as their order and orientation. If X has dimension 2 and all the 0-cells are ′ contained in Y (in particular Y is non-empty) then Ci(X,Y ) = 0 for i 6= 1, 2 and so the ρ ρ ρ torsion is just the determinant of the boundary map ∂2 : C2 (X,Y ) → C1 (X,Y ) which is computed via Fox calculus (see e.g. [39, Claim 16.6]). Thise implies that we can define the torsion provided K is only a commutative ring with unit and τ ρ(X,Y ) ∈ K (instead of the ring obtained by inverting the non-zero divisors of K). In particular, if h : π1(X) → H1(X) is the Hurewicz map and if we define ρ ⊗ h : π1(X) → GL(V ⊗K K[H1(X)]) by (6) ρ ⊗ h(δ)(v ⊗ f) := ρ(δ)(v) ⊗ (h(δ) · f)

ρ⊗h where δ ∈ π1(X), v ∈ V,f ∈ H1(X), then there is a torsion τ (X,Y ) ∈ K[H1(X)] under the above hypothesis for (X,Y ). When ρ ≡ 1 and dim(V ) = 1 this is the maximal abelian torsion of the pair (X,Y ).

Now suppose (M, γ) is a balanced sutured manifold with R−(γ) connected and let π = π1(M,p) where p ∈ s(γ) is a basepoint. We now explain how to compute the ρ e torsion τ (M,R−) from an extended Heegaard diagram H = (Σ, α , β). Suppose H is oriented, β-based and ordered, that is, each set α and β is linearly ordered and write α = {α1 ...,αd}, β = {β1,...,βd} in the corresponding orders. Let a = {αd+1,...,αd+l} (unordered). Consider the presentation of π1(M,p) given in (3). If F is the free group ∗ ∗ ∗ Z Z generated by α1,...,αd+l, recall that the Fox derivatives ∂/∂αi : [F ] → [F ] are charac- ∗ ∗ ∗ ∗ ∗ terized by ∂αj /∂αi = δij and ∂(uv)/∂αi = ∂u/∂αi + u · ∂v/∂αi for each u, v ∈ F . From ∗ Z now on, we will consider the Fox derivatives ∂βj/∂αi as elements of [π]. TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 15

Proposition 3.4. Given an ordered, oriented, based, extended Heegaard diagram (Σ, αe, β) of (M, γ), the twisted torsion of the pair (M,R−) is computed via Fox calculus by ∂β τ ρ(M,R ) = det ρ σ j − ∂α∗ i !!!i,j=1,...,d up to an indeterminacy of the form ± det(ρ(g)), g ∈ π where σ : Z[π] → Z[π] is deﬁned by σ(g)= g−1 for g ∈ π. The same formula is valid for τ ρ⊗h. Proof. This is the twisted version of [8, Lemma 3.6]. Think of (M, γ) as obtained from a single 0-handle by attaching l-handles (specifying R− × I), then d-handles corresponding to the α’s and then d-handles corresponding to the β’s. Now, let X be the CW complex obtained by collapsing each of these handles to its core and let Y be the subcomplex of X corresponding to R− × I. By [41, Corollary 8.5] collapsing a handle to its core does ρ ρ ρ ρ not changes the relative torsion, so τ (M,R−)= τ (X,Y ). But τ (X,Y ) = det(∂2 ) where ρ ρ ρ ∂2 : C2 (X,Y ) → C1 (X,Y ) is the boundary map, since X is a 2-complex whose 0-cells are ρ ρ all contained in Y . For an appropriate basis of C2 (X,Y ) and C1 (X,Y ), the boundary map ρ ∗ ∂2 is represented by the matrix ρ σ ∂βj/∂αi i,j=1,...,d [39, Claim 16.6], this implies the desired formula.

3.7. Refining τ using Spinc via multipoints. We now explain a simple Heegaard- diagrammatic way to remove the indeterminacies of the above formula, relying on the multipoint representation of Spinc. This is an idea from [26] which we restate in the case of torsion. Let H be an ordered, oriented and β-based extended Heegaard diagram of (M, γ) as above, then the determinant of the right hand side of Proposition 3.4 is a well-defined scalar of K (i.e., there is no indeterminacy). This depends on the extra structure on H (basepoints and orientations) up to ± det(ρ(g)) as follows. Let q ∈ β be the basepoint of a curve β ∈ β and let q′ ∈ β be another basepoint. Let c ⊂ β be the oriented arc from q to q′. Let β, β′ be the words in the α∗’s obtained from β but starting from q and q′ respectively. These are related by β = c · β′ · c−1 where c ∈ π is obtained by the multiplying the (α∗)±1 through c when following its orientation. But then the Fox derivatives satisfy ∂β ∂β′ = c · ∂α∗ ∂α∗ when thought as elements of Z[π], for each α ∈ α. Thus, changing the basepoint of a β-curve has the effect of multiplying the above determinant by det ρ(c), where c is the arc from the old basepoint to the new one. There is another source of a det ρ(g) indeterminacy, coming from the choice of orientations of the closed α curves. Indeed, reversing a curve αi ∗ ∗ has the effect of multiplying ∂βj/∂αi by −αi on the right for each j, hence the determinant n ∗ −1 of Proposition 3.4 is multiplied by (−1) det(ρ(αi ) ). The above det ρ(g) indeterminacies can be removed using the rule of Definition 3.3. Thus, pick a multipoint x = {x1,...,xd} of H, for simplicity we suppose xi ∈ αi ∩ βi for each i, and put basepoints on all β curves as in Definition 3.3. This immediately removes the det(ρ(g)) indeterminacy coming from the orientation of the α’s: if the orientation of an αi is reversed, then the basepoint of βi is crossed through αi, and this cancels the above ∗ det(ρ(αi )) factor (only leaving a sign indeterminacy). Now if we change the multipoint, say to y ∈ Tα ∩ Tβ, then the basepoints of all β curves change. More precisely, if dj is the 16 DANIEL LOPEZ´ NEUMANN arc of βj from the x-basepoint to the y-basepoint, then the whole determinant is multiplied d by j=1 det ◦ρ(dj ) by the above argument. But by (5), this is exactly det ◦ρ(ǫ(y, x)) (note that det ◦ρ descends to H (M)). Similarly, if we are given s ∈ Spinc(M, γ), then by (4) Q 1 the scalar det ◦ρ(P D[s − s(x)]) is multiplied by det ◦ρ(ǫ(x, y)) when going from x to y. It follows that the expression ∂β det ◦ρ(P D[s − s(x)]) · det ρ σ j ∂α∗ i i,j=1,...,d is independent of the multipoint chosen, where the β curves have the basepoints coming from x. This still has a sign indeterminacy, but this is easily corrected by further picking an orientation ω of H∗(M,R−(γ); R). Indeed, if δ is the sign defined by o(H)= δω where o(H) is the orientation of H∗(M,R−(γ); R) induced from the ordering and orientation of H [8] and if n = dim(V ), then ∂β δn · det ◦ρ(P D[s − s(x)]) · det ρ σ j ∂α∗ i i,j=1,...,d is independent of the extra structure we added to H (ordering, orientation and basepoints). One can then show directly that this is invariant under extended Heegaard moves, hence c it depends only on (M, γ), ρ : π1(M,p) → GL(V ), s ∈ Spin (M, γ) and the orientation ω. Thus, the latter expression can be considered as another normalization of the twisted ρ Reidemeister torsion τ (M,R−(γ)), cf. [8].

Definition 3.5. Let (M, γ) be a balanced sutured 3-manifold with connected R−(γ) and ρ let ρ, s,ω be as before. Then we denote by τ0 (M,γ, s,ω) the above refinement of the twisted ρ Reidemeister torsion τ (M,R−(γ)). 3.8. The disconnected case. Now let (M, γ) be a balanced sutured 3-manifold with disconnected R−(γ). Attach a 2-dimensional one-handle h0 to two circles of s(γ) corresponding to different components of R−(γ). Thickening h0 to a 3-dimensional one-handle h0 ×[−1, 1] ′ ′ ′ attached along s(γ) × [−1, 1] = γ results in a sutured manifold (M , γ ) in which R−(γ ) has one component less than R−(γ). Therefore, after sufficiently many such handle at- tachments, we can ensure that the resulting sutured manifold, still denoted (M ′, γ′), has ′ c connected R−(γ ). Let s ∈ Spin (M, γ) and suppose v is a vector field representing s. ′ ′ ′ One can extend v to a non-vanishing vector field v on any such M by letting v |M = v ′ and letting v be the vertical vector field over each one-handle h0 × [−1, 1] attached to M. The Spinc structure on M ′ defined by v′ is denoted i(s). This defines an affine map i : Spinc(M, γ) → Spinc(M ′, γ′), that is, it satisfies that ∗ j (i(s1) − i(s2)) = s1 − s2 c ∗ 2 ′ ′ 2 for any s1, s2 ∈ Spin (M, γ), where j : H (M , ∂M ) → H (M,∂M) is induced by inclusion, see [16, Proposition 5.4]. a homology orientation ω of (M, γ) induces a homology ′ ′ ′ R ∼ ′ ′ R orientation ω for (M , γ ) since H∗(M,R−; ) = H∗(M ,R−; ). Then we set ′ ρ ρ ′ ′ ′ τ0 (M,γ, s,ω) := τ0 (M , γ , i(s),ω ) ′ ′ ′ where (M , γ ) is any sutured manifold with connected R−(γ ) constructed as above and ′ ′ ′ ρ : π1(M ) → GL(V ) is an arbitrary extension of ρ (note that π1(M )= π1(M) ∗ F where F is a free group with as many generators as handles were attached to M to get M ′). It is also a consequence of [26] (via Theorem 1 and Theorem 2) that this is well-defined, i.e., TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 17 independent of the (M ′, γ′) chosen and it is a consequence of [8, Lemma 3.20] that this is ρ indeed a refinement of the torsion τ (M,R−(γ)). Note that if we take ρ ⊗ h, then a priori ′ ′ ρ ⊗h ′ ′ ′ K ′ K τ0 (M , γ , i(s),ω ) ∈ [H1(M )]. This actually belongs to [H1(M)] as a consequence of Lemma 4.7 below together with Theorem 2.

3.9. Twisted torsion of sutured link complements. Now let L be a link in a closed oriented 3-manifold Y , with components L1,...,Lm. Let (ML, γL) be the associated sutured manifold (see Example 3.2) and let ρ : π1(ML) → GL(V ) be a representation into a finite dimensional vector space V over a field K. For simplicity, we will assume Y is an ∼ ⊕m homology sphere, so H1(ML) = Z . Let h : π1(M) → H1(ML) be the projection, recall ′ that ρ can be combined with h to define a representation ρ ⊗ h over V = V ⊗K K[H1(ML)] ρ⊗h ′ as in (6). If ML is the universal covering of ML, then C∗ (ML) = C∗(ML) ⊗Z[π] V is a ′ complex of free Z[H1(ML)]-modules (V is a Z[π]-module via ρ⊗h). For each i ≥ 0, the i-th homology groupg of this complex is a finitely presented modulegover theg Noetherian UFD K[H1(ML)], so we can define its order, which is an element of K[H1(ML)] defined up to ρ multiplication by a unit. This is the i-th twisted Alexander polynomial of L, denoted ∆L,i, ρ the first twisted Alexander polynomial is denoted just by ∆L. If L is ordered and oriented and if ti ∈ H1(ML) is the class of a meridian around Li oriented so that lk(ti,Li) = +1 K ∼ K ±1 ±1 for each i = 1,...,m then [H1(ML)] = [t1 ,...,tm ] canonically, and so we can think K ±1 ±1 of the twisted Alexander polynomials of L as elements of [t1 ,...,tm ] defined up to multiplication by a unit. See [9, 39] for more details.

We now relate these polynomials to the torsion of (ML,R−). If R− = R1 ∪···∪ Rm ∗ where Ri is an annulus around Li, let yi ∈ Ri be a basepoint and ai be the generator ∗ of π1(Ri,yi) corresponding to ti. Then ai is deﬁned in π1(ML,p) up to conjugation, so ∗ det(tiρ(ai ) − In) is well-deﬁned and non-zero. Note that the 0-th Alexander polynomial ρ ∆L,0 is always non-zero by [9, Proposition 3.2, (1)].

Proposition 3.6. Let (ML, γL) be the sutured manifold associated to the complement of an m-component (ordered, oriented) link L ⊂ Y where Y is an homology sphere. Let ρ : π1(ML) → GL(V ) be a homomorphism, where V is an n-dimensional vector space over a ﬁeld K. Then m det(t ρ(a∗) − I ) τ ρ⊗h(M ,R (γ )) ˙= i=1 i i n · ∆ρ (t ,...,t ) ∈ K[t±1,...,t±1]. L − L ∆ρ L 1 m 1 m Q L,0 ρ⊗h m ∗ −1 Proof. Let M = ML, since τ (R−)= i=1 det(tiρ(ai ) − In) is non-zero, we have ρ⊗h ρ⊗h −1 ρ⊗h τ (M,R−)=Q τ (R−) τ (M) ρ ρ⊗h ρ⊗h by multiplicativity of the torsion. If ∆L = 0, then C∗ (M) is not acyclic and so τ (M)= ρ⊗h ρ 0. Thus τ (M,R−) is zero and the above equation holds. Suppose now that ∆L 6= 0, then ρ ρ ∆L,2 = 1 by [9, Proposition 3.2] (since M has non-empty boundary) and since ∆L,3 = 1 ρ ρ⊗h and ∆L,0 6= 0 always hold, it follows that C∗ (M) is acyclic. By [39, Theorem 4.7] we get m ∆ρ τ ρ⊗h(M,R ) ˙= det(t ρ(a∗) − I ) · L,1 − i i n ∆ρ i L,0 Y=1 as desired. 18 DANIEL LOPEZ´ NEUMANN

ρ Note that ∆L,0 = 1 whenever m> 1 (and ρ is arbitrary) or if m = 1 and ρ is irreducible with ρ|Ker (h) 6= 1, while if m = 1 and ρ ≡ 1 (with dim(V ) = 1), then ∆L,0 = t − 1 [9, Proposition 3.2].

4. Twisted Kuperberg invariants ρ In this section, we define the invariant IH (M,γ, s,ω) and study some of its properties. ρ We begin by defining a scalar ZH (H) by twisting Kuperberg’s formulas via Fox calculus using the homomorphism ρ, where H is an extended Heegaard diagram of a balanced sutured 3-manifold (M, γ) with connected R−(γ). This is not necessarily a topological invariant of (M, γ, ρ) as it may have some indeterminacies of the form ±rH (ρ(g)), g ∈ π1(M). In Subsection 4.2 we fix these indeterminacies using homology orientations and c ρ Spin structures, leading to IH (M,γ, s,ω) and we treat the case of disconnected R−(γ) as well. In Subsection 4.3 we explain how to extend such numerical invariants to polynomial invariants when H is N-graded. In Subsection 4.4, we prove Theorem 2. Finally, we discuss in Subsection 4.5 the constraints imposed by the involutory condition of Hopf superalgebras. Throughout all this section, we let (H,m, 1, ∆,ǫ,S) be a finite dimensional involutory Hopf superalgebra over a field K. We suppose H admits a two-sided cointegral c ∈ H and a two-sided integral µ ∈ H∗, and we assume µ(c) = 1. We also let (M, γ) be a connected balanced sutured 3-manifold and let ρ : π1(M,p) → Aut(H) be a group homomorphism, where p ∈ s(γ) is a basepoint. 4.1. Twisted tensors associated to Heegaard diagrams. We begin by supposing that e R−(γ) is connected. Let H = (Σ, α , β) be an extended Heegaard diagram of (M, γ) and e let α = α ∪ a with α = {α1,...,αd} the closed curves and a = {αd+1,...,αd+l} the arcs. From now on, we suppose H is ordered, that is, both sets α and β are totally ordered, and that H is oriented and β-based, so all the curves and arcs are oriented and each curve in ρ K β has a basepoint. With this additional structure, we will define a scalar ZH (H) ∈ by contracting some tensors associated to the closed circles and the crossings. These tensors are defined in the following way. First, to each closed α ∈ α we associate the tensor

. c ∆ . where the coproduct has as many outcoming legs as crossings through the curve α. More precisely, we suppose that α has a basepoint (in α \ β) and that from top to bottom, the legs correspond to the crossings of α encountered as one follows the orientation of α starting from this basepoint. Note that since we supposed c is two-sided, this tensor does not depends on the basepoint of α chosen. Now, to each crossing x ∈ α ∩ β we associate the tensor

Sǫx

ǫx where ǫx ∈ {0, 1} is defined by (−1) = mx and mx is the intersection sign. So far, these are the tensors from [21] and we haven’t used yet that H is an extended Heegaard diagram. The arcs in a will play a role in the definition of the β-tensors, through the TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 19 homomorphism ρ. To define these recall that, since R−(γ) is assumed to be connected, the extended Heegaard diagram together with the orientation of αe ∪ β and β-basepoints ∗ e determines a presentation of π1(M,p) as in 3. This has a generator α for each α ∈ α and each β ∈ β determines a relator β, which is obtained by writing (α∗)mx for each crossing e x ∈ α ∩ β with sign mx and multiplying these elements from left to right as one follows the orientation of β starting from its basepoint. Thus, for a given α, the appearances of α∗ in β are indexed by the intersection points of α ∩ β and so the Fox derivative ∂β/∂α∗can be written as ∂β = m β ∈ Z[F ], ∂α∗ x x x∈Xα∩β ∗ e for some word βx ∈ F , where F is the free group generated by α , α ∈ α . More precisely, this word is defined as follows: Notation 4.1. Suppose a curve β ∈ β is oriented and has a basepoint in β \ αe. For each ∗ mx ′ crossing x through β let αx be the α-curve containing x. Write β = w(αx) w where w (resp. w′) is the product of the α∗’s corresponding to the crossings of β that precede (resp. succeed) x, where the crossings are ordered following the orientation of β starting from its basepoint. Then

w if mx = 1 βx := ∗ −1 w(α ) if mx = −1. x

From now on, we think of βx as an element of π1(M,p) instead of the free group generated by the α∗’s.

Hence, each crossing x through a given curve β ∈ β has associated a Hopf automorphism ρ(βx) ∈ Aut(H). If x1,...,xk are the points of α ∩ β encountered as one follows the orientation of β starting from its basepoint, then we associate to β the tensor

ρ(βx1 )

ρ(βx2 ) m µ

ρ(βxk )

where, from top to bottom, the incoming legs of this tensor correspond to x1,...,xk. Now, take the tensor product of all the α-tensors according to the ordering of α. In other words, stack the above α-tensors one over another in such a way that, from top to bottom, they are ordered as in α. Do the same for the β-tensors, now following the ordering of β. Since each outcoming leg of an α-tensor corresponds to a unique crossing x, as well as to a unique incoming leg of a β-tensor, we can compose all the outcoming legs of the above ordered α-tensor with all the incoming legs of the β-tensors along the tensors corresponding to the crossings. The result is a scalar in the base ﬁeld of H. 20 DANIEL LOPEZ´ NEUMANN

Definition 4.2. Let H = (Σ, αe, β) be an ordered, oriented, β-based extended Heegaard diagram of (M, γ) and ρ : π1(M) → Aut(H) a group homomorphism. We denote by ρ K ZH (H) ∈ ρ the contraction of the tensors defined above. We will also denote by KH (H) the tensor H⊗d → H⊗d obtained by contracting the tensors above (here d = |α| = |β|), but without the cointegrals and integrals, so that ρ ⊗d ρ ⊗d ZH (H)= µ (KH (H)(c )). ρ Note that H needs to be α-based as well for KH (H) to be defined (as we need to know at which crossing of a given α we start applying the coproduct).

ρ The scalar ZH (H) is not always a topological invariant of (M, γ, ρ), but it is so provided Im (ρ) ⊂ Ker (rH ) and the cointegral of H has degree zero (as we will see later). If Im (ρ) ρ is not contained in Ker (rH ), then ZH (H) has a ±Im (rH ◦ ρ)-valued indeterminacy coming from the choice of basepoints of β as well as the orientations of the alpha curves. This is analogous to what happens to the torsion as we saw in Subsection 3.7. For instance, let q,q′ be two basepoints on a curve β ∈ β and let c ⊂ β the oriented arc from q to q′. If ′ ′ βx, βx denote the above Fox calculus elements starting from q and q respectively, then ′ βx = c · βx for each crossing x through β. Since µ ◦ ρ(c) = rH (ρ(c))µ, it follows that the tensor associated to β starting from q is rH (ρ(c)) times the tensor associated to β but starting ′ ρ ρ ′ ′ from q . Hence ZH (H)= rH (ρ(c))ZH (H ) where H is the ordered, oriented, based extended Heegaard diagram obtained by moving the basepoint q to q′. A similar indeterminacy appears when reversing the orientation of an α-curve. When the cointegral has degree one, ρ then ZH (H) has a sign indeterminacy coming from the choice of ordering and orientations of α ∪ β. Example 4.3. Let K ⊂ S3 be the left trefoil knot and let (M, γ) be the associated sutured 3-manifold, that is, M = S3\N(K) and γ consists of a pair of oppositely oriented meridians. Consider the (oriented, based) extended Heegaard diagram of (M, γ) of Figure 1 (left). For ∗ ∗ simplicity of notation, we denote α , a ∈ π1(M) just by α, a. If x1,x2,x3 are the points of β ∩ α, encountered as one follows the orientation of β starting from the given basepoint, −1 −1 −1 −1 −1 then βx1 = a, βx2 = aαa α and βx3 = aαa α a . If we suppose ρ abelian, then the above contraction of tensors is given as follows: ρ(a)

c ∆ S m µ ρ(a−1)

If H = K[X]/X2 is an exterior Hopf algebra, then ρ(a) ∈ Aut(H) has the form ρ(a)(X)= × j tX for some t ∈ K . Using that c = X and µ(X ) = δj,1 the above contraction of Hopf algebra tensors reduces to

µ(ρ(a)(X)+ S(X)+ ρ(a−1)(X)) = t − 1+ t−1 which is the Alexander polynomial of the left trefoil. TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 21

Remark 4.4. If H = (Σ, α, β) is a Heegaard diagram of a closed 3-manifold Y in the usual sense, so Σ is a closed surface, then deleting a small disk to Σ deﬁnes a sutured Heegaard 3 1 diagram H0 of the associated sutured 3-manifold (M0 = Y \ B , γ0 = S ⊂ ∂B), where B ⊂ Y is an embedded 3-ball. In this situation, the (involutory) formulas of Kuperberg [21] are the ρ ≡ 1 case of the formulas here, so we get ρ≡1 Kup ZH (H0)= IH (Y ).

Similarly, if HK is a Heegaard diagram of the sutured complement of a knot K in a closed ρ 3-manifold Y , then computing ZH (HK) at ρ ≡ 1 has the effect of forgetting the sutures. But doing this results in a Heegaard diagram of the underlying 3-manifold, so ρ≡1 Kup ZH (HK )= IH (Y ) as well. ρ c 4.2. Definition of IH (M,γ, s,ω). Fix a Spin structure s on (M, γ) and an orientation ω of the vector space H∗(M,R−(γ); R). We now explain how to remove the indeterminacies ρ ρ of ZH (H) (when R−(γ) is connected) and define the invariant IH (M,γ, s,ω) for arbitrary (M, γ, ρ, s,ω). That this is an invariant is a consequence of Theorem 1, but we emphasize that the arguments are a generalization of those of Subsection 3.7 for the torsion. Let H be an ordered, oriented, extended Heegaard diagram of (M, γ) (with connected R−(γ)). Let x ∈ Tα ∩ Tβ be a multipoint, and suppose H is β-based via x using the ρ ρ convention of Definition 3.3. We note by ZH (H, x) the tensor ZH (H) where H has the basepoints on β induced from x. The multipoint x defines a Spinc structure s(x) as in Subsection 3.4, and comparing this with our chosen s ∈ Spinc(M, γ) we get an homology class fs,x := P D[s(x) − s] ∈ H1(M). × The composition rH ◦ ρ : π1(M,p) → K descends to H1(M) and so can be evaluated over fs,x. Then the scalar ρ K rH ◦ ρ(fs,x) · ZH (H, x) ∈ turns out to be independent of the multipoint chosen. The argument is the same as in Subsection 3.7, only that det is replaced by rH . Moreover, this scalar depends on the orientations of the closed curves only up to a sign. To correct this sign indeterminacy (along with the sign indeterminacy coming from the ordering of the closed curves) we use the homology orientation in the same way as we did for the torsion: the ordering and orientation of H induces a homology orientation o(H) and so there is a sign δ defined by o(H)= δω. Then |c| ρ δ · rH ◦ ρ(fs,x) · ZH (H, x) is independent of the ordering, orientations and the multipoint x chosen. Then one has to show that this formula is invariant under extended Heegaard moves, so it defines a topological invariant of the tuple (M, γ, ρ, s,ω). This is a consequence of [26] via Theorem 1.

Deﬁnition 4.5. Let (M, γ) be a balanced sutured manifold with connected R−(γ). We ρ denote the above invariant by IH (M,γ, s,ω), that is, ρ |c| ρ K IH (M,γ, s,ω) := δ · rH ◦ ρ(fs,x) · ZH (H, x) ∈ 22 DANIEL LOPEZ´ NEUMANN where H is any ordered, oriented, extended Heegaard diagram of (M, γ) with basepoints coming from a multipoint x of H and fs,x ∈ H1(M) and δ ∈ {±1} are deﬁned as above.

Whenever (M, γ) has disconnected R−(γ) we proceed as in Subsection 3.8. Thus, we take any balanced sutured manifold (M ′, γ′) obtained by adding one-handles to M along γ ′ in such a way that R−(γ ) is connected and set ′ ρ ρ ′ ′ ′ IH (M,γ, s,ω) := IH (M , γ , i(s),ω ) ′ ′ ′ where ρ : π1(M ) → Aut(H) is any homomorphism such that ρ ◦ j∗ = ρ, where j∗ : ′ ′ π1(M) → π1(M ) is induced by inclusion and i(s),ω are as in Subsection 3.8. Again, this is well-defined as a consequence of our previous work via Theorem 1, see [26, Proposition 4.7]. ρ c Remark 4.6. The invariant IH depends on the Spin structure on (M, γ) as follows: for any h ∈ H2(M,∂M) ρ −1 ρ IH (M,γ, s + h, ω)= rH ◦ ρ(P D[h]) · IH (M,γ, s,ω) 2 where P D : H (M,∂M) → H1(M) is Poincaréduality. This follows from the definitions using that P D[s(x) − (s + h)] = P D[s(x) − s]P D[h]−1. 4.3. Twisted Kuperberg polynomials. Suppose now that H is N-graded (together with the previous hypothesis as well). We denote the N-degree of H by | · |0. We explain now ρ ρ⊗h K how to extend our invariant IH to IH ∈ [H1(M)] in a way that mimics the passage from the torsion τ ρ to τ ρ⊗h.

Let (M, γ) be a balanced sutured manifold and set HM := H ⊗K K[H1(M)]. Recall that for any α ∈ Aut(H) and f ∈ H1(M) there is a K[H1(M)]-linear Hopf automorphism α ⊗ f of HM deﬁned by α ⊗ f(h ⊗ f ′) := α(h) ⊗ (f |h|0 · f ′) ′ where h ∈ H is homogeneous and f ∈ K[H1(M)], see (1). Then, a homomorphism ρ : π1(M,p) → Aut(H) can be combined with the Hurewicz map h : π1(M) → H1(M; Z) to deﬁne a homomorphism

ρ ⊗ h : π1(M,p) → Aut(HM ) δ 7→ ρ(δ) ⊗ h(δ).

Note that though K[H1(M)] is not even necessarily a domain, HM has a two-sided cointegral and integral induced from that of H (given by cHM := c ⊗ 1 and µHM := µ ⊗ idK[H1(M)]). Thus, Theorem 1 holds for the K[H1(M)]-linear Hopf superalgebra HM , and we get an invariant ρ⊗h |c| ρ⊗h I (M,γ, s,ω) := δ · r ◦ ρ ⊗ h(f x) · Z (H, x) ∈ K[H (M)], HM HM s, HM 1 where fs,x = P D[s(x) − s] ∈ H1(M) as usual. For simplicity of notation, we denote this just by Iρ⊗h(M,γ, s,ω) and denote Zρ⊗h just by Zρ⊗h as well. Here Zρ⊗h(H) is obtained H HM H H by contracting the same tensors associated to the α’s and the crossings as before (since cHM = c ⊗ 1), only the β-tensors contribute H1(M)-coeﬃcients coming from ρ ⊗ h. Since we are only considering automorphisms of HM of the form α ⊗ f, by Lemma 2 we can write ρ⊗h IH more explicitly as ρ⊗h |c| |c|0 ρ⊗h IH (M,γ, s,ω)= δ · rH ◦ ρ(fs,x) · fs,x · ZH (H, x). TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 23

ρ⊗h Suppose now that (M, γ) has disconnected R = R−(γ). Then our deﬁnition of IH is ′ ′ ρ⊗h ρ ⊗h ′ ′ ′ IH (M,γ, s,ω) := IH (M , γ , i(s),ω ) where (M ′, γ′) is a sutured manifold obtained by adding one-handles to γ so that R′ = ′ ′ ′ ′ R−(γ ) is connected, ρ : π1(M ) → Aut(H) is such that ρ ◦ j∗ = ρ where j∗ : π1(M) → ′ ′ ′ ′ ′ π1(M ) is induced by inclusion, i(s),ω are as in Subsection 3.8 and h : π1(M ) → H1(M ) ′ ′ is the Hurewicz map of M . However, the right hand side belongs to K[H1(M )] and ′ m H1(M )= H1(M) ⊕ Z where m is the number of one-handles attached to γ.

Lemma 4.7. Let (M, γ) be a connected balanced sutured manifold with disconnected R−(γ) ′ ′ ′ ′ ′ ′ ′ ρ ⊗h ′ ′ ′ K and let (M , γ ), ρ , i(s),ω , h be as above. Then IH (M , γ , i(s),ω ) ∈ [H1(M)].

Proof. Let R0,...,Rm be the components of R. It suffices to prove the statement for a ′ ′ single (M , γ ) which we suppose is obtained by attaching a single one-handle between R0×I and Ri × I for each i = 1,...,m, where I = [−1, 1]. We will show the lemma by finding ′ ′ ′ an appropriate oriented, based Heegaard diagram of (M , γ ) for which h (βx) ∈ H1(M) for each β and each crossing x. To achieve this, let H = (Σ, αe, β) be an extended Heegaard e diagram of (M, γ) where as usual α = α ∪ a and α = {α1,...,αd}. For each i = 1,...,d let hi be the one-handle attached to R × I associated to αi, so hi has belt circle αi. Since M is connected, after sufficiently many handleslidings among the hi’s, we can suppose that hi has one feet in R0 × {1} and the other in Ri × {1} for each i = 1,...,m and that hi is attached to R0 × {1} for each i>m. Thus, Σ \ (α1 ∪···∪ αm) has m + 1 components Σ0,..., Σm, where for each i = 1,...,m, αi has one side on Σ0 and the other ∗ ∗ on Σi. Note that the αi with i>m together with the α , α ∈ a generate H1(M) while ∗ ∗ Zm ′ Zm α1,...,αm generate the factor in H1(M )= H1(M)⊕ . We will give αi the orientation induced from Σi for each i = 1,...,m and an arbitrary orientation for i>m. Let β ∈ β with an arbitrary orientation. By our choice of orientation of the α ∈ α, if a crossing x ∈ αi ∩ β is positive, then β is oriented from Σ0 to Σi near x and oppositely if the crossing is negative. Moreover, since Σ \ αi is disconnected and since Σi contains no α ∈ α, if β intersects αi positively at x, then the next intersection of β with α (following the orientation of β) is a negative intersection at αi. Now, suppose the basepoint of β lies in Σ0. Then, when following the orientation of β, the first intersection of β with α1 ∪···∪ αm is positive, and each positive intersection with the same set is followed by a negative one ∗ ∗ (possibly after several intersections with a). Hence, the appearances of α1,...,αm in each ′ βx cancel out in homology, so h (βx) ∈ H1(M) for each crossing x through β. In general, the basepoints of β come from a multipoint x = {x1,...,xd} of H, say with xi ∈ αi ∩ βi for each i, but with our choice of orientations of α1,...,αm the induced basepoints on β ′ indeed lie in Σ0. Thus h (β ) ∈ H1(M) for each β ∈ β and each crossing x. Therefore ′ ′ x ρ ⊗h ′ K ′ ′ ′ ZH (H , x) ∈ [H1(M)] where H is the Heegaard diagram of (M , γ ) obtained from H by gluing the corresponding 2-dimensional one-handles to Σ. On the other hand, the vertical extension map i : Spinc(M, γ) → Spinc(M ′, γ′) satisfies s′(x)= i(s(x)) where s′ is the map from multipoints to Spinc of M ′ so ′ P D[s (x) − i(s)] = j∗(P D[s(x) − s]) ′ lies in the image of the map j∗ : H1(M) → H1(M ) induced by inclusion. It follows that ′ ′ ρ ⊗h ′ ′ ′ K IH (M , γ , i(s),ω ) ∈ [H1(M)] as desired. 24 DANIEL LOPEZ´ NEUMANN

Deﬁnition 4.8. Let (M, γ) be an arbitrary balanced sutured 3-manifold and let ρ : c π1(M,p) → Aut(H), s ∈ Spin (M, γ) and ω be as usual and h : π1(M) → H1(M) be ρ⊗h K the Hurewicz map. We call IH (M,γ, s,ω) ∈ [H1(M)] the twisted Kuperberg polynomial of (M, γ) with respect to ρ. Example 4.9. Consider the left trefoil complement as in Example 4.3 and let ρ ≡ 1. Let t be a generator of H1(M) =∼ Z. Then, for the diagram of Figure 1, the twisted Kuperberg polynomial is given by

1⊗h |c(1)|0 −|c(3)|0 ZH (H)= µ(t c(1) · S(c(2)) · t c(3)) where this time we have used Sweedler’s notation for the coproduct, that is, we write ∆(x)= x(1) ⊗ x(2), (∆ ⊗ idH )∆(x)= x(1) ⊗ x(2) ⊗ x(3), etc. for each x ∈ H. As before, this recovers the Alexander polynomial of the trefoil knot if H is set to be an exterior algebra on one generator of degree one.

Recall that the augmentation map aug : K[H1(M)] → K is the K-linear map deﬁned by aug (f) = 1 for all f ∈ H1(M). Then we have the following. Proposition 4.10. The twisted Kuperberg polynomial satisﬁes ρ⊗h ρ aug (IH (M,γ, s,ω)) = IH (M,γ, s,ω).

Proof. Set aug H := idH ⊗ aug : HM → H. It is easy to see that aug ◦ µHM = µ ◦ aug H and aug H ◦ [(ρ ⊗ h)(δ)] ◦ jH = ρ(δ) for any δ ∈ π1(M), where jH : H → HM ,x 7→ x ⊗ 1 is the inclusion. Since the cointegral of HM is jH (c), where c is the cointegral of H, it follows that aug (Zρ⊗h(H)) = Zρ (H) where H is a Heegaard diagram of (M, γ). From this the HM H result follows. Now let L be an ordered oriented m-component link in an homology sphere Y and let (ML, γL) be the sutured manifold complement. The ordering and the orientation of L allows K K ±1 ±1 to identify [H1(ML)] = [t1 ,...,tm ] canonically. Corollary 4.11. Let L be an ordered oriented m-component link in an homology sphere ρ⊗h Y . Then IH (ML, γL, s,ω) is a multivariable polynomial invariant of L belonging to K ±1 ±1 [t1 ,...,tm ]. This satisfies that ρ⊗h ρ aug (IH (ML, γL, s,ω)) = IH (ML, γL, s,ω). In particular, for ρ ≡ 1 we have h Kup aug (IH (ML, γL, s,ω)) = IH (Y ). ρ⊗h K ±1 ±1 Proof. That IH (ML, γL, s,ω) belongs to [t1 ,...,tm ] is a consequence of Lemma 4.7. The rest follows from the above proposition together with Remark 4.4. 4.4. Reidemeister torsion as a Kuperberg invariant. We now show Theorem 2. This follows essentially from Proposition 4.12 below. So let (M, γ) be a balanced sutured 3- manifold with connected R−(γ), p ∈ s(γ) and let ρ : π1(M,p) → GL(V ) be a homomorphism, where V is a finite dimensional vector space over a field K. Let H = (Σ, αe, β) be an extended Heegaard diagram of (M, γ) which is ordered, oriented and based. This e determines a presentation of π1(M,p) as in (3). As usual, let α = α ∪ a, d = |α| = |β| and α = {α1,...,αd}. TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 25

Proposition 4.12. If Λ(V ) is the exterior Hopf algebra over V , then

∂β Zρ (H) = det ρ i ∈ K. Λ(V ) ∂α∗ j !!i,j=1,...,d

The same formula is valid for ρ ⊗ h as well.

This follows from the following simple observations on exterior algebras: (1) Λ(V ) is N-graded and super-commutative, (2) Λ(V ) is a universal enveloping algebra, that is, algebra maps Λ(V ) → A into some superalgebra A over K are in one-to-one correspondence (via restriction) with Lie superalgebra maps V → A (V has the trivial Lie algebra structure), (3) If T : V → V is a linear map, then the induced algebra morphism Λ(T ) : Λ(V ) → Λ(V ) satisﬁes

det(T )= µ(Λ(T )(c))

where µ and c denote the integral and cointegral of Λ(V ) normalized by µ(c) = 1.

ρ Proof of Proposition 4.12. Recall that ZΛ(V )(H) is deﬁned as

ρ ⊗d ρ ⊗d ZΛ(V )(H)= µ (KΛ(V )(H)(c ))

ρ ⊗d ⊗d ⊗d where KΛ(V )(H) : Λ(V ) → Λ(V ) is the twisted Kuperberg tensor. Note that Λ(V ) is naturally isomorphic to Λ(V ⊕d) and that c⊗d, µ⊗d are respectively a cointegral and ⊕d ρ N integral for Λ(V ). It is easy to see that KΛ(V ) preserves the -grading of Λ(V ), since ρ it is a composition of degree-preserving maps. Now, by deﬁnition, KΛ(V ) is a composition of coproducts, antipodes, automorphisms of Λ(V ) and the multiplication map. Since Λ(V ) ρ is commutative, all these are algebra maps, hence KΛ(V ) is also an algebra map. By the ρ ⊕d universal enveloping algebra property, KΛ(V ) is determined over V , and since it preserves N ⊕d ⊕d ⊕d ρ the -grading, the restriction to V is a linear map T : V → V so KΛ(V )(H) = Λ(T ) and

ρ ZΛ(V )(H) = det(T )

ρ by property (3) above. Now, by deﬁnition of KΛ(V ), one easily sees that the map T is ∗ represented by the matrix (ρ(∂βi/∂αj )), which implies the proposition. The assertion for ρ ⊗ h follows from its deﬁnition and the fact that V is concentrated in degree one.

Proof of Theorem 2. Suppose first that R−(γ) is connected. Note that the inverse-transpose satisfies ρ−T (σ(x)) = ρ(x)T for any x ∈ Z[π], where recall that σ : Z[π] → Z[π] is the map −1 defined by σ(g)= g for g ∈ π. Recall also that fs,x := P D[s(x)−s], that |c| = n (mod 2) 26 DANIEL LOPEZ´ NEUMANN if dim(V )= n and c is the cointegral of Λ(V ), and that rΛ(V ) = det. Thus, we get ρ |c| ρ x IΛ(V )(M,γ, s,ω) := δ · rΛ(V ) ◦ ρ(fs, ) · ZΛ(V )(H, x) n ∗ = δ · det ◦ρ(fs,x) · det(ρ(∂βi/∂αj )) n ∗ T = δ · det ◦ρ(fs,x) · det(ρ(∂βj/∂αi ) ) n −T ∗ = δ · det ◦ρ(fs,x) · det(ρ (σ(∂βj/∂αi ))) n −T −T ∗ = δ · det ◦ρ (P D[s − s(x)]) · det(ρ (σ(∂βj/∂αi ))) ρ−T = τ0 (M,γ, s,ω) where we use Proposition 4.12 in the second equality and the last equality is by definition of τ0 (which is itself based on Proposition 3.4). The case when R−(γ) is disconnected follows from the connected case by the definition of τ0, see Subsection 3.8. The assertion − ρ⊗h (ρ⊗h) T ρ−T ⊗h−1 ρ−T ⊗h for IΛ(V ) follows from the above since clearly τ0 = τ0 = σ(τ0 ) for any (M,γ, s,ω). Note that if we had used the convention that (g · c) ⊗ v = c ⊗ (ρ(g)t(v)) for the tensor ρ ρ product C∗(M)⊗Z[π] V (as in [30]), then IΛ(V ) would be exactly τ0 . Note also that Corollary 1 follows directly from the above theorem together with Corollary 3.6. f 4.5. Going beyond Reidemeister torsion? We now address the question of whether invariants different than Reidemeister torsion can be found within the present framework. Recall that we suppose H is an involutory Hopf superalgebra over a field K whose cointegral and integral are two-sided. We will further assume that char (K) = 0. It turns out that the involutory condition imposes serious restrictions. Suppose first that we seek for involutory Hopf algebras (i.e. without mod 2 grading). Then we are constrained by the following theorem of Larson-Radford [23, 24]: a finite dimensional (ungraded) Hopf algebra over a field K of characteristic zero is involutory if and only if it is semisimple and cosemisimple. As far as the author knows, all finite dimensional semisimple Hopf algebras over a field of characteristic zero are built from group algebras in some way. Therefore, one may expect that such Hopf algebras only lead to invariants related to a count of homomorphisms π1(M) → G. Fortunately, Larson-Radford’s theorem is no longer true for Hopf superalgebras, indeed, the opposite holds [2, Corollary 3.1.2]: if H = H0 ⊕ H1 is a finite dimensional involutory Hopf superalgebra over a field of characteristic zero, then H1 6= 0 if and only if H is non-semisimple. So, suppose we seek for involutory Hopf superalgebras (with H1 6= 0 and char (K) = 0). Suppose we add the stronger condition of cocommutativity (which immediately implies involutority). Then, we are constrained by the classical theorem of Cartier-Kostant-Milnor-Moore (see e.g. [2, Theorem 2.3.4]), which says that H is isomorphic, as a Hopf superalgebra, to a semidirect product: H =∼ K[G(H)] ⋉ U(P (H)). Here G(H) is the group of group-likes of H, that is, the elements that satisfy ∆(g)= g ⊗ g and ǫ(g) = 1, P (H) is the Lie algebra of primitive elements, i.e. those h ∈ H satisfying ∆(h) = 1 ⊗ h + h ⊗ 1 and U(P (H)) is the universal enveloping algebra of P (H). Now, if we wish U(P (H)) to be finite dimensional, then P (H) is forced to be an abelian Lie TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 27 algebra concentrated in degree one, so U(P (H)) is just an exterior algebra. Therefore, a finite dimensional cocommutative Hopf superalgebra (over a characteristic zero field) is necessarily of the form K[G] ⋉ Λ(V ) for some finite subgroup G ⊂ GL(V ). Thus, we get nothing beyond torsion with such Hopf superalgebras. In view of the above, if one wishes to find an example of a finite dimensional involutory Hopf superalgebra considerably different from group algebras, exterior algebras and their semidirect products, then one has to look for non-commutative and non-cocommutative Hopf superalgebras. The quantum group Uq(gl(1|1)) at a root of unity (see e.g. [38]) is such an example, for another one see [18, Example 5.6]. However, the cointegrals and integrals of such are not two-sided, so they don’t fit in the present framework. Still, these Hopf superalgebras seem to be obtained from exterior algebras somehow, for instance, Uq(gl(1|1)) (at a root of unity of order n) is the Drinfeld double of a semidirect product K[Z/nZ] ⋉ Λ(K). Therefore, it seems difficult to expect something beyond torsion with such examples. It has to be noted that more interesting involutory examples can be found in positive characteristic. For instance, for every restricted Lie algebra g there is an associated restricted universal enveloping algebra U res(g) which is cocommutative (hence involutory) and is of finite dimension if g is. Whether such Hopf algebras lead to interesting invariants will be the subject of future work.

5. Fox calculus via semidirect products ρ In this section we explain where the Fox calculus formula for IH (M,γ, s,ω) comes from. We begin in Subsection 5.1 by briefly recalling the setting of [26]. In Subsection 5.2 we show that a semidirect product K[Aut(H)] ⋉ H fits into this setting and we prove Theorem 1. In Subsection 5.3 we give an explanation of our invariant in terms of the more common notion of a Hopf group-algebra, therefore relating it to work of Virelizier [44]. Furthermore, we discuss why we don’t use more general Hopf group-algebras. Finally, in Subsection 5.4, we discuss the relation between both approaches. 5.1. Invariants from relative integrals. We begin by briefly recalling the setting of [26]. There, we also built a Kuperberg-style invariant of balanced sutured 3-manifolds, but with a more involved and general algebraic input. This consisted of a (possibly infinite- dimensional) involutory Hopf superalgebra H endowed with relative versions of the Hopf algebra cointegral and integral, satisfying certain conditions. More precisely, we defined a right relative cointegral as a tuple (A, πA, ιr) wheree A is a Hopf subalgebra of H, πA : H → A is a cocentral Hopf morphism with πA|A = idA (where cocentral means (πA ⊗ idHe )∆He = op (π ⊗ id e )∆ ) and ι : A → H is a left A-comodule map (where H is a lefte A-comodulee A H He r via πA) that satisfies the following relative version of the right cointegral equation: e e mA ιr ιr m e H = . πA

Moreover, H is endowed with a right relative integral, which is the notion dual to the above one, relative to a central Hopf subalgebra B ⊂ H. Here we will only consider the e e 28 DANIEL LOPEZ´ NEUMANN case B = K, so a relative integral is just an ordinary integral µ : H → K. We further ∗ required the existence of two distinguished group-likes a ∈ Hom alg(A, K) and b ∈ G(B), K ∗ so b = 1 if we assume B = . The group-like a must satisfy e e |ι| op ∆a∗ ◦ ιr = (−1) S e ◦ ιr ◦ SA and µ ◦ m = µ ◦ m e ◦ (id e ⊗ ∆a∗ ), H He H H ∗ where ∆ ∗ : H → H is given by ∆ ∗ := (a π ⊗ id e ) ◦ ∆ e . Since b = 1, the dual condition a a A H e H e becomes e e |µe| op µ ◦ S e = (−1) µ and ∆ ◦ ιr =∆ e ◦ ι. H He H

In addition, we assume µ(ιr(1A)) = 1. e e ρ Using such structures we deﬁned a topological invariant I e (M,γ, s,ω) where (M, γ), s,ω e H are as in the present paper and ρ is a homomorphism from H1(M) into the group of group- likes G(A ⊗ B∗) (= G(A) since here we assume B = K).e The construction also holds for arbitrary homomorphisms ρ : π1(M) → G(A), the necessary extra details will be carried out in Appendix A (we will do this only for the semidirect product case). When R−(γ) is connected, this invariant was built as follows: pick an ordered, oriented, based extended Heegaard diagram H = (Σ, αe, β) of (M, γ). Now we require that H is both α-based and β-based. Then to each α ∈ αe we associate the tensor

∗ ∆ e . ∗ ∆ e . ρ(α ) ιr H . or ρ(α ) iA H .

depending on whether α is a closed curve or an arc. Here iA : A → H is the inclusion and the outcoming legs of the iterated coproduct correspond to the crossings through α, starting from its basepoint and following its orientation. Since we considere B = K, to each β ∈ β we associate the tensor

. m e µ . H

e where the incoming legs correspond to the crossings through β as usual. Finally, to each e ǫx ǫx crossing x ∈ α ∩β we associate S (where, as before, ǫx ∈ {0, 1} is defined by (−1) = mx He and mx is the intersection sign at x). The contraction of all these tensors is denoted by Zρ (H) ∈ K or Zρ (H, x) if the basepoints of α and β come from a multipoint x. Here He He the rule of Definition 3.3 is extended in an obvious way to α: if the crossing xi is positive (resp.e negative),e we put a basepoint on αi just before (resp. after) xi when following the ρ orientation of αi. The invariant I (M,γ, s,ω) of [26] is then defined by He Iρ (M,γ, s,ω)= δ|ι| · ha∗, ρ(P D[s(x) − s])i · Zρ (H, x) He e He ∗ K where δ is the signe characterized by o(H) = δω as before andea ∈ Hom alg(A, ) is the above distinguished group-like. This definition extends to the case of disconnected R−(γ) by using (M ′, γ′), ρ′, i(s),ω′ as in Subsection 3.8. TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 29

5.2. Proof of Theorem 1. We begin by showing that semidirect products fit into the preceding framework, hence they define invariants of balanced sutured 3-manifolds. Lemma 5.1. Let H be a finite dimensional Hopf superalgebra with a two-sided cointegral c and a two-sided integral µ, normalized by µ(c) = 1, and let A = K[Aut(H)]. Then H = A ⋉ H has a right relative cointegral over A given by πA = idA ⊗ ǫH and ιr(α) := −1 ∗ c · α = rH (α) α ⊗ c for each α ∈ Aut(H). The distinguished group-like a ∈ G(A) is given ∗ bye a (α) := rH (α) for each α ∈ Aut(H). The integral is given by µ(α ⊗ h)= δα,idH µ(h) for any α ∈ Aut(H) and h ∈ H. e Proof. That ιr is a relative cointegral and µ is an integral is easy to check. That these satisfy the conditions stated in Subsection 5.1 follows from the fact that the cointegral and integral of H are assumed to be two-sided. Fore instance, we have ∗ ∆a∗ ◦ ιr(α)= a (πA(c(1) · α))c(2) · α = rH (α)ǫ(c(1))c(2) · α = rH (α)c · α while

|ιr| |c| −1 |c| (−1) SHe ιSA(α) = (−1) SHe (c · α ) = (−1) α · S(c)= α · c = rH (α)c · α where we used that S(c) = (−1)|c|c, which is equivalent to say that c is two-sided.

Proof of Theorem 1. Let H = K[Aut(H)] ⋉ H. We begin by showing that Zρ (H)= Zρ (H) e He H for any ordered, oriented, based extended Heegaard diagram of a balanced sutured manifold e (M, γ) (with connected R−(γ)), where the right hand side is the Fox calculus tensor of Subsection 4.1. This follows by writing the contraction of the tensors of H above in terms of the tensors of H. Indeed, for the semidirect product Hopf algebra H, the H-tensor associated to a closed α is e e e iH mHe . . ιr ∆ e . = c . . α H . ∆H . α . iH mHe . α

∗ by deﬁnition of ιr, where we write α ∈ Aut(H) instead of ρ(α ) for simplicity and iH : H → H is the inclusion. The tensor corresponding to an arc only puts the automorphism associated to the dual of the arc on the crossings through that arc. It follows that contracting alle the H-tensors from αe ∪ β is the same as contracting the H-tensors associated to the closed α’s (of Subsection 4.1) with the following H-tensors for each β: e e iH αi1

iH mHe µ αi2

iH e αik 30 DANIEL LOPEZ´ NEUMANN

where we write β = αi1 ...αik and, for the moment, we assume that all crossings through β are positive. Now, using the deﬁnition of the multiplication of H, and that µ ◦ iH = µ, the last tensor is the same as e e

αi1 mH µ

αi1 αi2 ...αik−1 which is the Fox calculus tensor of Subsection 4.1. This shows that Zρ (H) = Zρ (H) He H assuming all crossings are positive. If there is any negative crossings through β, it is easy −1 e e to see that the tensor SH contributes the extra αij at the tail of αi1 αi2 ...αij−1 for each negative x ∈ αij ∩ β (see Notation 4.1). This follows from −1 −1 −1 SHe (h · α)= α · SH (h)= α (SH (h)) · α . This shows that Zρ (H) = Zρ (H) in all cases. That Iρ = Iρ follows by the definitions He H He H of the refined invariants, since the distinguished group-like of H with the given relative e ∗ e cointegral structure is a = rH : Aut(H) → K. e 5.3. Hopf G-algebras and Fox calculus. The Fox calculus formula of Subsection 4.1 is also quite transparent when thought in terms of Hopf G-algebras. Indeed, we show that Virelizier’s extension of Kuperberg’s invariant through Hopf G-algebras [44], restricted to the Hopf G-algebra of Example 2.5, leads directly to our Fox calculus formula. We begin by briefly recalling the (dual) construction of [44], using an arbitrary involutory finite type Hopf G-algebra H = {Hα}α∈G. Let c = {cα}α∈G be a cointegral of H and µ : H1 → K be an integral of H1 such that µ(c1) = 1. Let Y be a closed oriented 3- manifold endowed with a representation ρ : π1(Y ) → G. Let H = (Σ, α, β) be an ordered, oriented, β-based Heegaard diagram of Y . For simplicity, we note Hα instead of Hρ(α∗) and cα instead of cρ(α∗) for each α ∈ α. Then, to each α ∈ α we associate the tensor

c ∆ . α α . where the are as many outcoming legs as crossings through α and to each crossing x we associate the tensor

ǫx . Sα

Now, suppose a curve β ∈ β has associated the word β = αm1 ...αmk i1 ik when starting from its basepoint and following its orientation (as usual, we denote α∗ ∈ m m π1(M,p) just by α). Since β =1 in π1(M,p), the iterated multiplication mα 1 ,...,α k of H i1 ik TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 31 has target H1, which can then be composed with the integral. Thus, we associate to each β the tensor

m m . mα 1 ,...,α k µ . i1 ik where there are as many incoming legs as crossings through β. These tensors can be ρ K contracted as usual, leading to a scalar ZH (H) ∈ . The main theorem of [44] is that this is an invariant of (Y, ρ) when H is involutory with dim(H1) 6= 0 in K, we denote it by ρ IH (Y ).

Proposition 5.2. Let H be the Hopf Aut(H)-algebra associated to the semidirect product K[Aut(H)] ⋉ H where H is a ﬁnite dimensional involutory Hopf algebra with dim(H) 6= 0 in K. Let Y be a closed 3-manifold and let (M0, γ0) be the associated sutured manifold, that 3 is, M0 = Y \ B where B is an embedded 3-ball in Y and γ0 is a single suture in ∂M0. Then ρ ρ IH (Y )= IH (M0, γ0) for any ρ : π1(Y ) → Aut(H). Proof. Let H = (Σ, α, β) be a (ordered, oriented, β-based) Heegaard diagram of Y and let H0 be the resulting diagram for (M0, γ0) obtained by deleting a small disk to Σ. For each α ∈ Aut(H), identify Hα =∼ H via the coalgebra isomorphism h · α 7→ h (recall Hα = {h·α | h ∈ H}⊂ K[Aut(H)]⋉H). The result follows by writing the above H-tensors associated to H in terms of the structure tensors of H and its automorphisms. Indeed, since our identiﬁcation is a coalgebra isomorphism, the above α-tensors immediately correspond to those of Subsection 4.1. Now, the above tensor associated to a crossing corresponds to

Sǫx α−ǫx .

This is not the same as the corresponding tensor of Subsection 4.1, but it will be after composition with the β-tensors. Indeed, under the above identiﬁcation the structure map mα1,α2 becomes

m , α1

m m hence, the iterated multiplication mα 1 ,...,α k is given by i1 ik

αm1 i1 m . . m − . αm1 ...α k 1 i1 ik−1 32 DANIEL LOPEZ´ NEUMANN

Thus, whenever a crossing x is negative, the α−1 coming from the antipode tensor provides −1 the α in the tail of βx as in the β-tensors of Subsection 4.1. This immediately implies that ρ ρ ZH (H)= ZH (H0) where the left hand side is the above tensor and the right hand side is ours, hence also ρ ρ IH (Y )= IH (M0, γ0).

Remark 5.3. The conditions on H in the above proposition imply semisimplicity of H and H∗ [32, Theorem 10.4.3]. However, in Theorem 1, H is not necessarily semisimple, for instance if H is a Hopf superalgebra with H1 6= 0 [2, Corollary 3.1.2] or if char (K)= p> 0 and p | dim(H) (also by [32, Theorem 10.4.3]). In this case, the Hopf Aut(H)-algebra H coming from K[Aut(H)] ⋉ H is non-semisimple, in particular, it may be non-unimodular so the construction of [44] does not applies. It turns out that there are two sorts of non- unimodularity of H: on the one hand, the graded-cointegral of such H is not two-sided if rH 6≡ 1 (see Example 2.5). On the other hand, even if we assume that the cointegral of the neutral component (which is H) is two-sided, H is not unimodular if the cointegral has mod 2 degree one (see Remark 2.2). In our setting, the Spinc structure takes care of the former failure of unimodularity while the orientation ω takes care of the latter. In particular, Spinc only appears in the non-unimodular G-graded setting, this is why this additional structure is not present in other works on Kuperberg invariants. The reason we do not deﬁne sutured manifold invariants from more general (involutory) Hopf G-algebras is the following. Suppose we want to extend the construction of [44] sketched above to an extended Heegaard diagram of a sutured 3-manifold. Then to each arc α ∈ a we need to associate a tensor

i ∆ . α α .

for some special element iα ∈ Hα (recall that we denote Hα instead of Hρ(α∗) for simplicity). By the cointegral property, the scalar obtained by contracting all the tensors associated to curves and arcs is invariant under arc-curve slidings. However, if we want this scalar to be ρ invariant under arc-arc slidings, so as to get an invariant IH (M, γ) of (M, γ) together with ρ : π1(M) → G, then it is natural to require iα to be a group-like element of Hα and that mα1,α2 (iα1 ⊗ iα2 ) = iα1α2 for each α1, α2 ∈ G. But if there exists (iα)α∈G satisfying these conditions, then G acts on H1 by conjugation by iα: if we set φ(α)(h)= mα,1,α−1 (iα ⊗ h ⊗ iα−1 ) for all α ∈ G, h ∈ H1, then φ : G → Aut(H1) is a group homomorphism. Moreover, the map Hα → (H1)φ(α), h 7→ φ(α) · (iα−1 h) is a Hopf morphism from the original Hopf G-algebra to the Hopf Aut(H1)-algebra built from K[Aut(H1)]⋉H1 and this map preserves the cointegrals. It follows that the scalar invariant obtained from H and (iα)α∈G can be obtained through the latter semidirect product, that is Iρ (M, γ)= Iφ◦ρ(M, γ), thus we do H H1 not gain anything new. 5.4. Relative integrals and Hopf G-(co)algebras. We now explain the relation between the relative integral setting of [26] and Hopf group-(co)algebras. Suppose H is equipped with a relative right cointegral ιr : A → H and a relative right integral µ : H → B, e e e e TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 33 where A ⊂ H is a cocentral subalgebra and B ⊂ H is a central subalgebra. Let’s assume a while that A = K. Then it is well-known that the subalgebra B induces a Hopf group-coalgebrae with group G := Hom alg(B, K) as follows:e for each β ∈ G, let Iβ be the two-sided ideal of H generated by elements of the form b − β(b)1He with b ∈ B and set Hβ := H/Iβ, so each Hβ is an algebra. It is easy to see that the coproduct ∆He descends e − to ∆β1β2 : Hβ1β2 → Hβ1 ⊗ Hβ2 and similarly the antipode descends to Sβ : Hβ → Hβ 1 . e e e Hence, {Hβ}β∈G is a Hopf group-coalgebra with the structure maps induced from H. In these terms,e a relativee integrale µ : H → B as in [26] induces a Hopf group-coalgebrae e inte- K gral {µβ}eβ∈G as in [43]. Indeed, it is easy to see that for each β ∈ Hom alg(B, ) thee map K K K β ◦ µ : H → descends to a mape eµβ : Hβ → . In other words, whenever A = , the relative integral approach of [26] is an explicit form of the Hopf group-coalgebra approach of [44e ], extendede to sutured manifolds. Dually,e the cocentral subalgebra A ⊂ H induces a Hopf G(A)-algebra structure and the relative cointegral is equivalent to a Hopf group- algebra cointegral (this generalizes Example 2.5). Whenever both A, B are non-trivial,e we are naturally led to an algebra-graded and coalgebra-cograded object, which is not a particular case of [44], but we do not have interesting concrete examples.

Appendix A. The non-abelian case In [26] we supposed ρ is an abelian representation, though this is not strictly necessary. Everything works for non-abelian ρ, but we have to take care of the basepoint p when we isotope the arcs of an extended Heegaard diagram. We now work out the necessary extra details. e Let (M, γ) be a balanced sutured 3-manifold with connected R−(γ) and let H = (Σ, α , β) e be an oriented, extended Heegaard diagram of it. Let α = α∪a where α = {α1,...,αd} are ∗ the closed curves and a = {αd+1,...,αd+l} are the arcs. Then the dual curves α ∈ π1(M,p) are invariant under isotopy of αe except for an overall conjugation if an arc α ∈ a is isotoped along ∂Σ past the basepoint p. More precisely, suppose the arc α is oriented so that α·δ = +1 where δ is the oriented boundary component of ∂Σ containing p. Suppose further that the arc α is isotoped past p in the opposite direction of δ, and denote by α′ the new arc and by H′ the oriented extended Heegaard diagram obtained by replacing α with α′. If ′ ′ we denote by αi the curves αi in H , then ′ ∗ ∗ ∗ ∗ −1 (αi) = α αi (α ) in π1(M,p) for all i = 1,...,d + l, see Figure 3. ρ Thus, to guarantee that IH is a topological invariant for non-abelian ρ we need to show that its deﬁning formula is invariant under conjugation of ρ by an element of Aut(H). This is done in the next lemma.

Lemma A.1. Given φ ∈ Aut(H), let ρφ : π1(M,p) → Aut(H) be the homomorphism −1 ρφ ρ deﬁned by ρφ(x) := φ ◦ ρ(x) ◦ φ for any x ∈ π1(M,p). Then ZH (H) = ZH (H). Hence, ρ IH is a topological invariant for non-abelian ρ and depends on ρ only up to conjugation. Proof. Since φ is a Hopf automorphism, we have

ρφ ⊗d ρ −1 ⊗d KH (H)= φ ◦ KH (H) ◦ (φ ) . 34 DANIEL LOPEZ´ NEUMANN ′ αi α p αi p α′ b b bbbb δ bbbb δ

Figure 3. An arc α is slided past the basepoint p ∈ δ ⊂ ∂Σ. On the left we ∗ −1 e draw αi as cAic , where Ai is a circle in int (Σ) intersecting α once at αi e and c is an arc disjoint from α from the basepoint to a point in Ai. On the right we see that the arc c has to be slided over α∗ to avoid an intersection ′ ∗ ∗ ∗ ∗ −1 with α. This shows that (αi) = α ai (α ) .

Now, since φ(c)= rH (φ)c and µ ◦ φ = rH (φ)µ we get ρφ d −1 d ρ ρ ZH (H)= rH (φ) rH (φ ) ZH (H)= ZH (H).

The second assertion follows from this and from the fact that rH ◦ ρ = rH ◦ ρφ, which holds since the target of this map is an abelian group.

ρ This lemma also implies that IH is independent of the basepoint p chosen. More precisely, let p1,p2 ∈ s(γ) be basepoints and let δ : [0, 1] → M be a path from p1 to p2. Denote by C[δ] the isomorphism π1(M,p1) → π1(M,p2) deﬁned by C[δ](α) = [δ]α[δ] for α ∈ π1(M,p1).

Corollary A.2. Let p1,p2 ∈ s(γ) be two basepoints and let ρi : π1(M,pi) → Aut(H) for i = 1, 2 be group homomorphisms related by ρ1 = ρ2 ◦ C[δ] for some path δ from p1 to p2. ρ1 ρ2 Then IH (M,γ, s,ω)= IH (M,γ, s,ω).

ρ1 ρ2 Proof. It suffices to show that ZH (H)= ZH (H). Note that changing the path δ by another path changes C[δ] by an inner automorphism of π1(M,p2). Therefore, by Proposition A.1, ∗ it suffices to prove the corollary for a specific path δ. Let αpi ∈ π1(M,pi) be the dual curves of the α’s coming from the diagram H with basepoint pi, i = 1, 2. If we just let δ be a path e from p1 to p2 contained in Σ \ α , which is connected since we assume R−(γ) connected, ∗ ∗ ∗ ∗ ∗ then αp2 = [δ]αp1 [δ]. Using ρ1 = ρ2 ◦ C[δ] we get ρ2(αp2 )= ρ2(δαp1 δ)= ρ1(αp1 ) ∈ Aut(H). It follows that Zρ2 (H)= Zρ1 (H). ρ When R−(γ) is disconnected, the above assertions also hold by our definition of IH in this case.

References 1. J. W. Alexander, Topological invariants of knots and links, Trans. Amer. Math. Soc. 30 (1928), no. 2, 275–306. 2. N. Andruskiewitsch, P. Etingof, and S. Gelaki, Triangular Hopf algebras with the Chevalley property, Michigan Math. J. 49 (2001), no. 2, 277–298. 3. C. Blanchet, F. Costantino, N. Geer, and B. Patureau-Mirand, Non-semi-simple TQFTs, Reidemeister torsion and Kashaev’s invariants, Adv. Math. 301 (2016), 1–78. TWISTING KUPERBERG INVARIANTS VIA FOX CALCULUS 35

4. L. Chang and S. X. Cui, On two invariants of three manifolds from Hopf algebras, Adv. Math. 351 (2019), 621–652. 5. Q. Chen, S. Kuppum, and P. Srinivasan, On the relation between the WRT invariant and the Hennings invariant, Math. Proc. Cambridge Philos. Soc. 146 (2009), no. 1, 151–163. 6. F. Costantino, N. Geer, B. Patureau-Mirand, and V. Turaev, Kuperberg and Turaev-Viro invariants in unimodular categories, arXiv:1809.07991 (2018). 7. P. Etingof, S. Gelaki, D. Nikshych, and V. Ostrik, Tensor categories, Mathematical Surveys and Mono- graphs, vol. 205, American Mathematical Society, Providence, RI, 2015. 8. S. Friedl, A. Juhász, and J. Rasmussen, The decategorification of sutured Floer homology, J. Topol. 4 (2011), no. 2, 431–478. 9. S. Friedl and S. Vidussi, A survey of twisted Alexander polynomials, The mathematics of knots, Contrib. Math. Comput. Sci., vol. 1, Springer, Heidelberg, 2011, pp. 45–94. 10. , Twisted Alexander polynomials detect fibered 3-manifolds, Ann. of Math. (2) 173 (2011), no. 3, 1587–1643. 11. , The Thurston norm and twisted Alexander polynomials, J. Reine Angew. Math. 707 (2015), 87–102. 12. D. Gabai, Foliations and the topology of 3-manifolds, J. Differential Geom. 18 (1983), no. 3, 445–503. 13. M. Hennings, Invariants of links and 3-manifolds obtained from Hopf algebras, J. London Math. Soc. (2) 54 (1996), no. 3, 594–624. 14. V. F. R. Jones, A polynomial invariant for knots via von Neumann algebras, Bull. Amer. Math. Soc. (N.S.) 12 (1985), no. 1, 103–111. 15. A. Juhász, Holomorphic discs and sutured manifolds, Algebr. Geom. Topol. 6 (2006), 1429–1457. 16. , The sutured Floer homology polytope, Geom. Topol. 14 (2010), no. 3, 1303–1354. 17. A. Juhász, D. P. Thurston, and I. Zemke, Naturality and mapping class groups in Heegaard Floer homology, arXiv:1210.4996 (2012). 18. R. Kashaev and A. Virelizier, Generalized Kuperberg invariants of 3-manifolds, Algebr. Geom. Topol. 19 (2019), no. 5, 2575–2624. 19. P. Kirk and C. Livingston, Twisted Alexander invariants, Reidemeister torsion, and Casson-Gordon invariants, Topology 38 (1999), no. 3, 635–661. 20. , Twisted knot polynomials: inversion, mutation and concordance, Topology 38 (1999), no. 3, 663–671. 21. G. Kuperberg, Involutory Hopf algebras and 3-manifold invariants, Internat. J. Math. 2 (1991), no. 1, 41–66. 22. , Non-involutory Hopf algebras and 3-manifold invariants, Duke Math. J. 84 (1996), no. 1, 83– 129. 23. R. G. Larson and D. E. Radford, Finite-dimensional cosemisimple Hopf algebras in characteristic 0 are semisimple, J. Algebra 117 (1988), no. 2, 267–289. 24. , Semisimple cosemisimple Hopf algebras, Amer. J. Math. 110 (1988), no. 1, 187–195. 25. X. S. Lin, Representations of knot groups and twisted Alexander polynomials, Acta Math. Sin. (Engl. Ser.) 17 (2001), no. 3, 361–380. 26. D. López-Neumann, Kuperberg invariants for balanced sutured 3-manifolds, arXiv:1904.05786 (2019). 27. C. McPhail-Snyder, Holonomy invariants of links and nonabelian Reidemeister torsion, arXiv:2005.01133 (2020). 28. J. Murakami, The multi-variable Alexander polynomial and a one-parameter family of representations 2 of Uq (sl(2, C)) at q = −1, Quantum groups (Leningrad, 1990), Lecture Notes in Math., vol. 1510, Springer, Berlin, 1992, pp. 350–353. 29. P. Ozsváth and Z. Szabó, Holomorphic disks and topological invariants for closed three-manifolds, Ann. of Math. (2) 159 (2004), no. 3, 1027–1158. 30. J. Porti, Reidemeister torsion, hyperbolic three-manifolds, and character varieties, Handbook of group actions. Vol. IV, Adv. Lect. Math. (ALM), vol. 41, Int. Press, Somerville, MA, 2018, pp. 447–507. 31. D. E. Radford, The group of automorphisms of a semisimple Hopf algebra over a field of characteristic 0 is finite, Amer. J. Math. 112 (1990), no. 2, 331–357. 32. , Hopf algebras, Series on Knots and Everything, vol. 49, World Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 2012. 36 DANIEL LOPEZ´ NEUMANN

33. K. Reidemeister, Homotopieringe und Linsenräume, Abh. Math. Sem. Univ. Hamburg 11 (1935), no. 1, 102–109. 34. N. Reshetikhin, Quantum supergroups, Quantum field theory, statistical mechanics, quantum groups and topology (Coral Gables, FL, 1991), World Sci. Publ., River Edge, NJ, 1992, pp. 264–282. 35. N. Reshetikhin and V. G. Turaev, Ribbon graphs and their invariants derived from quantum groups, Comm. Math. Phys. 127 (1990), no. 1, 1–26. 36. , Invariants of 3-manifolds via link polynomials and quantum groups, Invent. Math. 103 (1991), no. 3, 547–597. 37. L. Rozansky and H. Saleur, Quantum field theory for the multi-variable Alexander-Conway polynomial, Nuclear Phys. B 376 (1992), no. 3, 461–509. 38. A. Sartori, The Alexander polynomial as quantum invariant of links, Ark. Mat. 53 (2015), no. 1, 177–202. 39. V. G. Turaev, Quantum invariants of knots and 3-manifolds, De Gruyter Studies in Mathematics, vol. 18, Walter de Gruyter & Co., Berlin, 1994. 40. , Homotopy field theory in dimension 3 and crossed group-categories, arXiv:math/0005291 (2000). 41. , Introduction to combinatorial torsions, Lectures in Mathematics ETH Zürich, Birkhäuser Ver- lag, Basel, 2001, Notes taken by Felix Schlenk. 42. , Homotopy quantum field theory, EMS Tracts in Mathematics, vol. 10, European Mathematical Society (EMS), Zürich, 2010, Appendix 5 by Michael Müger and Appendices 6 and 7 by Alexis Virelizier. 43. A. Virelizier, Hopf group-coalgebras, J. Pure Appl. Algebra 171 (2002), no. 1, 75–122. 44. , Involutory Hopf group-coalgebras and flat bundles over 3-manifolds, Fund. Math. 188 (2005), 241–270. 45. M. Wada, Twisted Alexander polynomial for finitely presentable groups, Topology 33 (1994), no. 2, 241–256. 46. E. Witten, Quantum field theory and the Jones polynomial, Comm. Math. Phys. 121 (1989), no. 3, 351–399.

Institut de Mathematiques´ de Jussieu - Paris Rive Gauche, Universite´ Paris Diderot, F- 75013 Paris, France Email address: [email protected]