32 27 PDF

Theory and Applications of Categories, Vol. 32, No. 32, 2017, pp. 889–969.
BIEXTENSIONS, BIMONOIDAL FUNCTORS, MULTILINEAR

FUNCTOR CALCULUS, AND CATEGORICAL RINGS
ETTORE ALDROVANDI
Abstract. We associate to a bimonoidal functor, i.e. a bifunctor which is monoidal in

each variable, a nonabelian version of a biextension. We show that such a biextension
satisfies additional triviality conditions which make it a bilinear analog of the kind of
spans known as butterflies and, conversely, these data determine a bimonoidal functor.
We extend this result to n-variables, and prove that, in a manner analogous to that
of butterflies, these multi-extensions can be composed. This is phrased in terms of a
multilinear functor calculus in a bicategory. As an application, we study a bimonoidal
category or stack, treating the multiplicative structure as a bimonoidal functor with
respect to the additive one. In the context of the multilinear functor calculus, we view
the bimonoidal structure as an instance of the general notion of pseudo-monoid. We
show that when the structure is ring-like, i.e. the pseudo-monoid is a stack whose fibers
are categorical rings, we can recover the classification by the third Mac Lane cohomology
of a ring with values in a bimodule.
Contents
1 Introduction 890
I 897
2 Preliminaries 897
3 Biadditive morphisms 903
4 Biextensions by braided crossed modules 904
5 Biadditive morphisms and biextensions 908
6 Symmetry properties of biextensions 911
7 Butterflies 913
8 Biadditive morphisms and butterflies 916
9 Commutative structures 920
II 922
10 Multiextensions and compositions 922
Received by the editors 2015-08-05 and, in final form, 2017-07-28.

Transmitted by Larry Breen. Published on 2017-08-02.
2010 Mathematics Subject Classification: 18D10, 13D03, 18G55, 55P43, 14A20.
Key words and phrases: Categorical ring, biextension, bimonoidal, ring-like stack, butterfly, multi-
extension, multi-category, multi-functor, Mac Lane cohomology.
c Ettore Aldrovandi, 2017. Permission to copy for private use granted.
889
890 ETTORE ALDROVANDI
11 Compositions of n-butterflies 924

12 The bi-multicategories of group-like stacks and crossed modules 929
III 931
13 Bimonoidal structures and weakly categorical rings 931
14 The presentation of a categorical ring 934
15 Decomposition of categorical rings and the cohomology of rings 937
IV 950
A Multi-variable compositions 950
B (Unbiased) Monoids and monoidal structures 953
C Pentagons 956
D Some lemmas on symmetric braidings 960
E Hypercohomology computations 961
References 966
1. Introduction
Let H and G be monoidal stacks in a topos T and F a monoidal functor F : H → G . It
is well known that F can be represented by a special kind of span. More precisely, if the
monoidal laws are group-like, or if we restrict to the invertible objects (restricting to the
invertible objects does not affect the characteristic class), we can find presentations H•
for H and G• for G by crossed modules of T so that F is represented by a diagram of
group-objects of the form
H1 G1
! ~
E
}
H0 G0
where the salient feature is that the sequence G1 → E → H0 of objects of T is an
extension of H0 by G1 → G0 : an exact sequence in which the conjugation action of E on
G1 is compatible with that of G0 [AN09]. Such an extension is conveniently described in
geometric terms as a G1 -bitorsor E over H0 , with the property that one of the actions
is obtained by way of the crossed module structure of G• [Bre90]. A converse of this
correspondence is also available, establishing an equivalence between the groupoid of
monoidal functors from H to G and that of diagrams like the one above (a morphism
between two diagrams with the same wings H• and G• is a homomorphism of E → E 0
compatible with all the maps).
One of the main results of this paper is an analogous result for bi monoidal functors
F : H × K → G , that is, bifunctors which are monoidal in each variable.
BIEXTENSIONS, CATEGORICAL RINGS 891
The concept of bimonoidal functor requires G to be braided. In addition, F must
satisfy a compatibility condition requiring that applying the monoidal condition on both
of its variables in the two possible orders lead to the same result. (We can think of F
as an obvious generalization of a bilinear map, in which case this condition is trivially
satisfied.) This condition is formally equal to the compatibility one between the two partial
multiplication laws of a biextension [Mum69, Gro72]. Recall that for abelian groups A, B,
C of T, a biextension of B and C by A consists of an A-torsor E over B × C equipped
with two partial commutative and associative product laws, making E, for each generalized
point b ∈ B, an extension of C by A, and, similarly, for each c ∈ C, an extension of B
by A. The coincidence between these conditions is not purely formal: the bilinear point
of view is to regard a biextension as providing a bifunctor ϕE : B × C → Tors(A), which
is then monoidal (i.e. a homomorphism) in each variable. Notable examples are certain
duality pairings where, in particular, C = B ∨ and A = Gm (see, e.g. [Boy10, Dat10]).
Our result begins from a generalization of these ideas, starting with that of biextension.
Thus, for two groups H and K of T, and a braided crossed module (G1 , G0 ), we define
a biextension of H, K by (G1 , G0 ) as a G1 -bitorsor (or, more appropriately, using the
terminology of [Bre90], a (G1 , G0 )-torsor) E over H × K such that, for each point x ∈ H
(resp. y ∈ K), E is an extension of K (resp. of H) by the crossed module G1 → G0 . The
braiding is required by the compatibility between the two partial product laws of E. The
relative diagram is formally the same as in the classical situation, with some differences
due to the fact that none of the groups involved is assumed abelian—although the crossed
module G1 → G0 and its associated stack are braided.
Now, let F be a bimonoidal functor F : H × K → G , where each of K , H , and G has
a presentation by a crossed module. We show that F determines a biextension of H0 , K0
by the crossed module G1 → G0 , equipped with a pair of compatible trivializations for the
two pullbacks along the maps ∂ × id : H1 × K0 → H0 × K0 and id ×∂ : H0 × K1 → H0 × K0 .
This is what we call, in this context, a butterfly —the bilinear version of the notion
introduced in [AN09] (earlier, over a point, in [Noo05]). Viceversa, given a biextension E
of (H0 , K0 ) by G1 → G0 , we obtain a bimonoidal functor
ϕE : H0 × K0 −→ Tors(G1 , G0 ),
where the right hand side denotes the stack of (G1 , G0 )-torsors equivalent to G , and where
H0 and K0 , are interpreted as discrete monoidal stacks. The additional data provided by
the two compatible trivializations allow to conclude that ϕE is compatible with descent
along the presentations H0 n H1 // H0 → H and K0 n K1 // K0 → K by the groupoids
determined by their respective crossed modules, so it determines a bimonoidal functor
FE : H × K → G . With the obvious notion of morphism, we obtain:
1.1. Theorem. [Theorem 8.1 in the main text] There is an equivalence of (pointed)
groupoids
∼
u : Biext(H• , K• ; G• ) −→ Hom(H , K ; G ),
where the right hand side denotes the groupoid of bimonoidal functors, and left hand side
that of biextensions equipped with the aforementioned trivializations.
This result is actually valid over a variable object S of T, hence we have a similar
statement where the left and right side above are replaced by the corresponding stacks,
which one obtains by letting S → Biext(H• |S , K• |S ; G• |S ), and similarly for the right
hand side.
Bimonoidal categories or stacks provide examples of bimonoidal morphisms. If R is
bimonoidal, it has two monoidal structures, say and , satisfying an appropriate set of
axioms [Lap72a, Lap72b]; the distributivity one, in particular, says that is a bimonoidal
functor : R × R → R with respect to the other structure, . Now, assuming to be
group-like (see [BDRR13] for passing from a merely additive monoidal to a full group-like
one, i.e. from “rig” to “ring”) and R to have a presentation of the form R0 n R1 // R0 → R
as above, the bifunctor can be described by a biextension E of (R0 , R0 ) by R1 → R0 .
We are interested in extracting information about R, especially of a cohomological nature
such as the characteristic class, from the biextension E . Information of this kind, which
reduces to (usually complicated) cocycle calculations, ordinarily come from the coherence
diagrams of , among others. These diagrams involve morphisms like ◦ ( × id) and
◦ (id ×), which are monoidal in each of the three variables, and higher iterations of
compositions involving and id. Ideally, each of those multi-functors corresponds to some
kind of iteration of E . So we need to generalize the representation of bimonoidal functors
by biextensions to an arbitrary number of variables.
Extending the concept of bimonoidal functor to n variables is immediate, the only
difference being the same compatibility condition we have in the n = 2 case must hold
for each pair of variables. Therefore, for the multivariable analog of the right hand side
in the above statement, the only substantial difference is in the bookkeeping aspect, and
we immediately see that monoidal functors in n variables can be composed—provided
we restrict ourselves to considering braided objects. While this is easy to verify, it has
the far-reaching consequence that braided monoidal categories, or more generally stacks,
comprise the 2-categorical analog of a multicategory [CS10, Shu10] we denote MBGrSt,
where BGrSt is the 2-category they form relative to ordinary unary monoidal functors.
As for the left hand side of the equivalence in the theorem, we must extend the concept
of biextension to n variables and investigate whether such objects admit a composition law.
We define a multi-extension, or an n-extension, by G1 → G0 as a (G1 , G0 )-torsor (same as
for biextensions) over an n-fold product, this time equipped with n partial product laws
which are required to be pairwise compatible in the same manner as those in a biextension.
(The generalization of a biextension to n variables, in the fully abelian context, is outlined
in a remark in [Gro72].) Given n crossed modules Hi,• = (∂ : Hi,1 → Hi,0 ), i = 1, . . . , n,
we are interested in the n-extensions of H1,0 × · · · × Hn,0 by G1 → G0 which are also
equipped with n compatible trivializations of their pullbacks along each of the n morphisms
∂i = id × · · · × ∂ × · · · × id. We call them n-butterflies. Now, a direct extension of the
above theorem (cf. Theorem 10.1 below) provides an equivalence
∼
MExt(H1,• , . . . , Hn,• ; G• ) −→ Hom(H1 , . . . , Hn ; G )
of pointed groupoids, where each Hi,• is a presentation of the corresponding Hi , and G•

of G , the left hand side is the groupoid of n-extensions, and the right hand side denotes
the groupoid of n-monoidal functors H1 × · · · × Hn → G .
These multi-butterflies can be composed in a way reminiscent of the unary case discussed
in [AN09, Noo05], that is by “wing juxtaposition,” which turns out to be associative up
to coherent isomorphism. Wing juxtaposition (we informally use this term thanks to
the shape of the composition diagrams in the unary case) is an operation which allows
to associate to butterflies F1 , . . . , Fn , E, where E is an n-butterfly, a new one, denoted
E(F1 , . . . , Fn ), whose underlying bitorsor is a quotient of the pullback of E to the product
F1 × · · · × Fn —each Fi is equipped with an equivariant section with values in the ith factor
in the base of E; the pullback is along the product of these maps. The upshot is that
braided crossed modules of T now form a bi-multicategory (i.e. bi categorical analog of a
multi-category) of their own if we use these multi-butterflies as morphisms. In addition,
the composition is compatible with the previous equivalence. More precisely, we have:
1.2. Theorem. [Theorem 11.1, Propositions 11.3 and 11.4, and Theorem 12.3] Braided
crossed modules of T, equipped with the groupoids MExt(H1,• , . . . , Hn,• ; G• ) as Hom-
categories, form a bi-multicategory MBXMod. Further, there is an equivalence of bi-
multicategories
∼
MBXMod −→ MBGrSt,
induced by the associated stack functor, carrying each MExt(H1,• , . . . , Hn,• ; G• ) to
Hom(H1 , . . . , Hn ; G ).
This is the multi-categorical analog of the unary case treated in [AN09]1 , which states
the bicategory BXMod of braided crossed modules, equipped with groupoids of spans as
categories of morphisms, is equivalent (as a bicategory) to BGrSt. The proof, in particular
that of Theorem 11.1, is geometrical, unlike that of corresponding result in [AN09, Noo05].
It requires showing first that the pullback of the bitorsor E to the product F1 × · · · × Fn
is equivariant under the left and right actions of H1,1 × · · · × Hn,1 , and that it descends
to the base of F1 × · · · × Fn ; second, that this descended bitorsor has the structure of a
multi-butterfly.
We can now discuss bimonoidal categories from the broader and more convenient
perspective afforded by the idea of a bi-multicategory. If MC is a bi-multicategory, we
informally look into monoids with respect to the multi-composition structure of MC.
Formally, we say that an object X of MC is a (pseudo-)monoid of C, the underlying
bicategory of MC, if it is a pseudo-algebra over the club N, where N, after [Kel72], denotes
the natural numbers equipped with the category structure described in section A; a club is
a multicategory with one object.2 The monoidal structure carried by X is of the “unbiased”
kind [Lei04], namely we have an n-ary operation mn : X × · · · × X → X for each object
n ∈ N, as opposed to a privileged binary one, with the various coherence conditions being
taken care of by the pseudo-algebra structure.
1
In fact the unary version proved therein holds without any commutativity assumption.
2
For the sake of simplicity we are ignoring the “extraordinary” structure given by the action of
permutations.
We apply the previous observation to C = BGrSt or C = BXMod. We say that

R is weakly ring-like (for want of a better name) if it is a pseudo-monoid of BGrSt.
The “weakly” adverb refers to the fact that the underlying categorical group structure
of R is only braided commutative, whereas we usually define categorical rings as having
underlying braided symmetric categorical groups (see, e.g. [JP07]). Thus “weak” does not
mean “not strict,” but it signals the commutative law of the additive structure of R is
weaker than usually required.
We make a similar definition for a pseudo monoid in BXMod: this is a novel object
comprised of a braided crossed module R• and multi-extensions En of (R0 , . . . , R0 ) (n
factors) by R1 → R0 for each n ∈ N, subject to the conditions dictated by the pseudo-
algebra structure. We show that even in the realm of a genuine bicategory such as BXMod,
we obtain coherence conditions resembling a weak version of Mac Lane’s pentagon (see
sections B and C, in particular diagram (C.2)).
Since pseudo-monoids are transported across equivalences, pseudo-monoids in BGrSt
are presented by those in BXMod. We have:
1.3. Proposition. [Proposition 13.5, Proposition 14.1, and Corollary 14.2] Let R be
(weakly) ring-like. Let R0 n R1 // R0 → R be a presentation by a braided crossed module.
Then R• is equipped with biextensions En , n ∈ N as above so that (R• , E• ) is a pseudo-
monoid in BXMod. Furthermore, if A = π0 (R) ' coker ∂ and M = π1 (R) ' ker ∂, then
A is a (possibly non-unital) ring of T and M is an A-bimodule. The converse also holds.
We may ask whether the weakly ring-like structure is strong enough to force the
underlying braiding to become symmetric. The answer is affirmative, at least in the more
interesting (to us) case where the multiplicative structure has a unit object. As we shall
see, this is tied to the seemingly unrelated problem of computing a characteristic class for
weakly categorical rings.
Our procedure is a bit nonstandard in two aspects. The first is that to compute the
class corresponding to R we only choose local data for the underlying braided group-like
structure of R, or which is the same thing, for the presentation. We compute the rest of
the cocycle from the pseudo-monoidal structure carried by the presentation, by examining
the biextension’s behavior with respect to the chosen local data.
The second aspect is that we choose to initially dispense with the symmetry condition,
therefore we start with a choice of local data for the underlying braided group-like structure
of R. This yields an invariant in the Eilenberg Mac Lane group H 4 (K(A, 2), M ), lifting to
H 5 (K(A, 3), M ) if braiding is symmetric [JS93, Bre99].3 In fact the initial choice for the
local data steers the computation into being based on the Eilenberg Mac Lane iterated bar
construction [EML53, EML54], in place of the more customary Mac Lane’s Q-construction.
(In the stack context, the iterated bar construction emerges as part of the decomposition
of R as a gerbe over A, together with its additional braiding structure [Bre94a, Bre99].)
Over a point, a standard categorical ring gives rise to an element in the third Mac Lane
3
In a general topos we should interpret them as hypercohomology groups. See the discussion in [Bre69,
§2], [Bre94a, §7], [Bre99, §6].
cohomology group HML3 (A, M ) [JP07] (a gap was filled in [Qua13]; see also [Lod98, chap.
13] for a general definition), so the question is whether our procedure yields anything
standard, and in particular the expected invariant in HML3 (A, M ).
That it does, at least in the unital case, is due to the fact that, by [ML58, §11], the third
Mac Lane cohomology can also be computed by employing the infinite bar construction
B ∞ (A) := B(A, 1) ⊆ B(A, 2) ⊆ · · · instead of the cubical complex Q(A).4 One needs a
product structure on B ∞ (A), explicitly up to degrees corresponding to the subcomplex
B(A, 2), to use in the multiplicative bar construction. It follows that we can plug one of
B(A, 2), B(A, 3), or B ∞ (A) to calculate a cohomology of A with values in the bimodule
M . In the unital case, all these choices give rise to isomorphic cohomologies. The product
structure on B(A, 2), which is unusual, is explicitly given in Mac Lane’s work. We recover
that product, and hence the rest of the structure, directly from the pseudo-monoidal
structure attached to the presentation of R, namely the biextension plus the various
compatibility conditions it satisfies.
We do not obtain the representing cocycle for the class of R in the form corresponding
to class in HML3 (A, M ) right away. Instead, we find that the class of R is represented
by a twisted cocycle in the multiplicative bar resolution for B(A, 2), but the twisting
disappears in the unital case, which yields the desired result. In particular this implies
that the braiding is necessarily symmetric. More precisely, we have:
1.4. Theorem. [Theorem 15.1] There is a bijective correspondence between equivalence
classes of weak ring-like stacks R, with A = π0 (R) and M = π1 (R), and twisted classes
in He 23 (A, M ) defined below (cf. Definition 15.9). In the unital case, i.e. when the external
monoidal structure of R has a unit element, A is likewise unital and the underlying braiding
of R is necessarily symmetric. Hence the weak ring-like structure of R is fully ring-like,
and [R] ∈ H e 3 (A, M ) ' HML3 (A, M ).
2
Hence, unital categorical rings in the sense of the present paper are categorical rings
tout-court and, by [ML58], we re-obtain their classification in terms of HML3 (A, M ). Had
we chosen to work with braided symmetric objects from the beginning there is little doubt
the same procedure—i.e. a decomposition of the underlying group-like stack followed by
an analysis of the attached biextension—would have yielded a Mac Lane cocycle in the
right group right from the start. However, part of the interest was to test to what extent
our framework successfully captures the standard theory.
Organization of the material. We have collected various preliminary items in
section 2, including a quick review of standard biextensions. The recapitulation of the
relation between group-like stacks and their presentations by crossed modules, though
standard, involved a contravariance issue which becomes relevant, therefore we provided a
brief outline.
4
Recall that classically MacL̃ane cohomology is defined as the Hochschild cohomology of the complex
Q(A). The latter computes the stable homology of the Eilenberg-Mac Lane spaces, that is, Hq (Q(A)) '
Hn+q (K(A, n)), n > q.
Aside from the preliminary section, we can divide the rest in roughly three parts, plus
another containing some appendices.
We develop the generalization of biextensions and their symmetry properties in sec-
tions 3 to 6, and the butterfly special kind, including the representation of bimonoidal
functors, from section 7 to 9.
The extension to an arbitrary number of variables takes sections 10 to 12. In particular,
the composition of butterflies is discussed in section 11. There, Theorem 11.1 establishes the
existence of the wing juxtaposition operation. Its proof is technically involved bookkeeping,
but it ultimately is straightforward; it is reproduced in full because, while the juxtaposition
product has features similar to the unary case, the proof itself is not an immediate
generalization of the corresponding one in [AN09, Noo05]. Note also that the constructions
in the proof are used in the cocycle computations of sect. 15.13.
The bi-multicategories comprised by braided group-like stacks on one side, and braided
crossed modules of T on the other, are discussed in section 12.
The idea of categorical ring, or ring-like stack, and its presentation as a pseudo-monoids
in their respective bicategories is expounded in sections 12 and 14, where we also discuss
some specific facts about the presentations. In section 15 we discuss the cohomology of
rings and the computation of the characteristic class. We do this in slightly simplified form
by omitting the additional computations which arise because the choices for the various
trivializations required for the cocycle computations are necessarily local. The complete
computations are deferred to Appendix E.
To avoid unnecessary and possibly long detours, parts of the material have been placed
in a number of appendices. Some of this material is necessary for self-consistency but
known to the experts. Thus, bi-multicategories are discussed in appendices A and B, the
analog of the pentagon in a bicategory in C, and appendix D contains just some technical
lemmas pertaining to section 6. Finally, the resolution of a simplicial object by hypercovers
and the complete hyper-cohomology computations necessary for a full calculation of the
characteristic class are contained in Appendix E.
What to read. One possibility is to only read Part I followed by Part III, if the reader
is willing to only skim through Part II. For the latter, it is possible to just read the
statements, in particular for the n-fold composition of multi-extensions theorem proved in
sect. 11, only referring to the proof of 11.1 when needed in sect. 15.13. Alternatively, one
can read Part III, in particular section 15, only skimming through the previous two, and
omitting most of section 15. Parts II and III depend on the multivariable functor calculus
in a bicategory, an account of which is contained in the first two appendices, which may
only be referred to when needed. Most their content should be known to the experts.
An account of what the pentagonal axiom would look like in a bicategory is contained
in Appendix C, and in general a reader will need only equation (C.1) and diagram (C.2).
While in section 15.13, the reader can refer to Appendix E for the full hyper-cohomology
arguments.
Notations and conventions. The convention we use is to equate “additive” with
“monoidal,” therefore the term “biadditive” means “bimonoidal” in the sense used above in
the introduction. By extension, “n-additive” means “n-monoidal,” i.e. monoidal in each
of the n-variables.
A notation of the form G• or G denotes a crossed module (G1 , G0 , ∂), where ∂ : G1 → G0
is the homomorphism, and the (right) action is denoted element-wise by (x, g) 7→ g x , but
is not otherwise labeled. Very often we will simply write (G1 , G0 ).
For ease of notation we will often use the convention: (G1 , G0 ) ∼= (G, Π) and, later in
∼
the paper, (R1 , R0 ) = (R, Λ) when we discuss ring-like structures.
Up to section 13, all monoidal structures are notated multiplicatively, with I denoting
the identity object. We switch to an additive notation for one of the two monoidal
structures in a bimonoidal situation. In that case, the identity object is 0. If this is the
structure for which we construct a presentation by a crossed module, then we use an
additive notation for the groups in the crossed module, even though they are by no means
assumed commutative.
The notation GS denotes the base-change of G to S, that is, S × G → S. If G is a group
object, then GS is group over S, with group law (s, g)(s, g 0 ) = (s, gg 0 ), in set-theoretic
notation.
Acknowledgements. I would like to thank Sandra Mantovani, Giuseppe Metere, Cris-
tiana Bertolin, and Enrico Vitale for interesting discussions and the kind hospitality
extended to me by their institutions in Milan, Turin, and Louvain-la-Neuve, where parts
of this work were presented.
Special thanks are also due to the anonymous referee, whose pointed criticisms and
inquiries led to a much improved version of this work.
Part I
2. Preliminaries
In the following, let T be a topos which is assumed to be Sh(C) for a site C. For the
reader’s convenience we recall some well know facts on monoidal stacks and their relations
with crossed modules of T.
2.1. Group-like stacks, crossed modules, and bitorsors. We follow [Bre90,
AN09], to which we refer for further details. A crossed module of T consists of a group
homomorphism ∂ : G → Π of T and a right action of Π on G such that: (1) ∂ is a morphism
of right Π-objects, with Π acting on itself by conjugation; (2) the action of G on itself
induced by the Π-action via ∂ coincides with conjugation. If g, h are (generalized) points
of G and x of Π we have the familiar conditions:
∂(g x ) = x−1 ∂(g) x
g ∂(h) = h−1 g h,
where the exponent notation denotes the action.

The crossed module G gives rise to a groupoid Γ : Π n G // Π , and hence, via the nerve
construction, to a simplicial group N G whose object in degree p is (· · · (Π n G) n · · · n G)
(p factors). The groups π1 = ker ∂ and π0 = coker ∂ are the homotopy (sheaves of) groups
of G• are in fact the only two nonzero homotopy sheaves of N G. It is easily verified that
π1 is abelian and central in G and that π0 is the sheaf of connected components of the
groupoid corresponding to G (and hence of N G).
The group laws of G and Π n G equip the groupoid Γ with a structure of strict
categorical group, that is, a morphism
m : Γ × Γ −→ Γ
of groupoids satisfying the usual axioms of a group-like, strict monoidal category (see also
[JS93]).
The associated stack construction performed on Γ yields a stack G of T, which inherits
the—now lax—group-like structure. (“The” associate stack is only unique up to equivalence.
The specific model we consider is the one whose fibered categories consist of descent data
for Γ and their morphisms—see [LMB00, Lemme 3.2]. It is easy to see that in the case at
hand these descent data are cocycles with values in Γ—see[AN09, §3.3-4].) Thus, G is a
monoidal stack, in the sense that there exists a stack morphism
m : G × G −→ G ,
and an associator µ : m ◦ (m × 1) ⇒ m ◦ (1 × m) satisfying the axioms of a monoidal
category. We specify the rest of the group-like structure by requiring the existence of a
unit object I, which we can identify with a morphism I : e → G (where e is the terminal
object of T), and that left and right-multiplications by a fixed object x induce functorial
equivalences
m(x, −) : G −→ G m(−, x) : G −→ G .
The group-like stack obtained in this way from a crossed module will be denoted [∂ : G →
Π]∼ .
There is a compelling geometric picture for the associated stack obtained in this way
from a crossed module. Recall the stack Tors(G, Π) is the stack of right G-torsors
P whose structure group extension P ∧G Π is isomorphic to the trivial torsor [Del79].
Equivalently, an object of this stack is a pair (P, s) where P is a G-torsor and s : P → Π
a G-right-equivariant morphism, where G acts on Π on the right via multiplication.
Objects of Tors(G, Π) are called (G, Π)-torsors. The notion makes sense for each group
homomorphism, but if ∂ : G → Π is a crossed module, then Tors(G, Π) becomes group-like.
Indeed, each torsor is also equipped with a left G-action defined by
gu = ug s(u) ,
where u ∈ P and g ∈ G, which makes it into a G-bi torsor. The monoidal structure is
given by the contracted product
m((P, s), (Q, t)) = (P ∧G Q, st),
and it is easily proved to be group-like. Finally, the stack Tors(G, Π) is equivalent, as a
group-like stack, to [∂ : G → Π]∼ , obtained via the associated stack construction described
above.
Conversely, for any group-like stack G there exists a presentation by a crossed module,
namely there exists a crossed module G = (G, Π, ∂) and an equivalence G ' [∂ : G → Π]∼ .
For any group-like stack G we can define π1 = AutG (I), and π0 as the sheaf associated
to the presheaf of connected components. Upon choosing a presentation, its homotopy
sheaves are isomorphic to the ones of G .
In view of the previous geometric interpretation, G is also equivalent to Tors(G, Π).
In addition, the sequence
(2.1) G
∂ /Π π /G ,
where each object is regarded as a stack, is exact in a homotopical sense. If G is identified
with Tors(G, Π), the projection Π → G sends x ∈ Π to the bitorsor (G, x), that is the
trivial right G-torsor whose left action on itself is given by g · u = g x u, with u, g ∈ G.
(This follows from the fact that x is identified with the equivariant section that assigns x
to the unit section eG .) In particular, (G, eΠ ) can be identified with the unit object of G .
2.2. Remark. Since a morphism ϕ : (P, s) → (Q, t) in Tors(G, Π) has the form
ϕ
P /Q
s t
Π
it is immediately seen that the equivalence
[∂ : G → Π]∼ −→ Tors(G, Π)
is contravariant. In particular, if g : x → x0 is a morphism in the strict categorical group
Γ, that is, x0 = x ∂g, the corresponding morphism of (G, Π)-torsors is (G, x0 ) → (G, x).5
A morphism of group-like stacks is a stack morphism F : H → G preserving the
monoidal structure. A morphism of crossed modules determines one between the associated
group-like stacks in the obvious way. In the converse direction, a morphism F : H → G
only determines a morphism in the homotopy category between corresponding crossed
modules. This morphism can be represented by a butterfly, namely a diagram of group
objects of T of the form:
H G
σ ı
~
∂ E 
∂
π
~
Σ Π
5
There appears to be no good way to get around the issue. A “fix” is to replace the crossed module
with a left one. This restores the expected direction of the arrows, at the cost of turning one of the
monoidal laws into the opposite one (see e.g. [Bre90, after Théorème 4.5]).
(where the vertical arrows are crossed modules) from which a fraction representing the
morphism F : H → G can be obtained (see loc. cit. and below for more details).
2.3. Braidings. A braided crossed module, (see [JS93, AN09]), is a crossed module
G = (G, Π, ∂) equipped with a bracket
{−, −} : Π × Π −→ G
such that ∂{x, y} = y −1 x−1 yx, for all points x, y of Π. The bracket satisfies:
{x, yz} = {x, y}z {x, z}, {x, ∂h} = h−1 hx ,
(2.2)
{xy, z} = {y, z}{x, z}y , {∂g, y} = g −y g,
for all x, y, z ∈ Π and g, h ∈ G. These properties all arise from the observation that the
bracket {−, −} corresponds to a braiding in the usual sense for the categorical group
determined by G• . Thus cx,y = {x, y} gives a family of functorial isomorphisms
cx,y : xy −→ yx
for each pair (x, y) of objects in Γ, so that c : m ◦ τ ⇒ m : Γ × Γ → Γ, where τ is
the interchange functor. The properties above can be derived from functoriality and
Mac Lane’s hexagon diagrams. Conversely, if Γ is a braided strict categorical group,
setting {x, y} = y −1 x−1 cx,y : e → y −1 x−1 yx defines a braiding on the corresponding crossed
module.
The braiding is symmetric if it has the property that {y, x} = {x, y}−1 for all x, y ∈ Π.
A symmetric braiding is Picard if in addition it satisfies the condition {x, x} = e. These
conditions match the corresponding ones for the categorical group Γ. As observed in
[Bre94a, §1], for a Picard crossed module the bracket is a full lift of the commutator map.
The braided, symmetric, and Picard structures translate in the expected manner to
the associated group-like stack G ' [∂ : G → Π]∼ , and conversely, if G is a braided (resp.
symmetric, Picard) stack with presentation given by [∂ : G → Π], then the latter acquires a
braided (resp. symmetric, Picard) structure as above. Let us observe here that a braiding
on G = Tors(G, Π) induces one on the crossed module by way of the butterfly representing
the morphism m : G × G → G . In the Picard case this leads to two possible presentations:
by a braided crossed module satisfying the Picard condition, or, according to Deligne
[Del73], by a length-one complex of abelian sheaves. We refer the reader to [AN09, §7]
for full details on this correspondence. Here we limit ourselves to observe that if (P, s)
and (Q, t) are two (G, Π)-torsors, the choice of two sections u ∈ P and v ∈ Q allows us to
write the braiding morphism cP,Q : P ∧G Q → Q ∧G P as cP,Q (u ∧ v) = (v ∧ u) χu,v , where
χ : P ∧G Q → G, is a coordinate representation of cP,Q subject to certain equivariance
conditions dictated by the requirements:
cP,Q (gu ∧ v) = g cP,Q (u ∧ v),
cP,Q (ug ∧ v) = cP,Q (u ∧ g v),
cP,Q (u ∧ vg) = cP,Q (u ∧ v) g.
2.4. Lemma. The coordinate χ : P ∧G Q → G has the expression:
(2.3) χu,v = {s(u), t(v)}−1 .
Proof Proof (Sketch). The relation (2.3) is immediate for (P, s) = (G, x) and
(Q, t) = (G, y), where x, y ∈ Π, using Remark 2.2. The general case follows by descent by
exploiting the equivariance conditions for χ and for {−, −} recalled above.
2.5. Biextensions. Biextensions were introduced in Mumford [Mum69], and later reex-
amined by Grothendieck [Gro72]. We refer to the latter and the text by Breen [Bre83]
for details on the standard concept of biextensions. We briefly recall the basic definitions,
and later extend them to introduce crossed modules as coefficients.
Let H, K, G be groups of T, with G assumed to be abelian. A biextension of H,
K by G is a GH×K -torsor E on H × K equipped with two partial composition laws ×1
(resp. ×2 ) making E a central extension of HK (resp. KH ) by GK (resp. by GH ). This
means that ×1 (resp. ×2 ) is only defined on E ×K E (resp. E ×H E), where E ×K E (resp.
E ×H E) means fiber product over K (resp. H) with respect to the obvious projection.
These composition laws are required to be compatible with one another. Usually one
also requires ×1 and ×2 to be commutative, which make sense whenever H and K are
abelian. Our definition, in which ×1 and ×2 are not required to be commutative is then
referred to as a weak biextension. Even when H and K are commutative, this notion
of biextension is weaker than that of loc. cit. in which the partial multiplication laws are
commutative.
Analogously to extensions, the composition laws can be represented by morphisms of
G-torsors. Let B H and B K be the standard classifying simplicial objects of T.6 The
bisimplicial object B H × B K has face maps dh and dv (see e.g. [GJ99]). In particular, for
i = 0, 1, 2 we consider dhi : H × H × K → H × K and dvi : H × K × K → H × K. Then
×1 and ×2 correspond to morphisms of G-torsors
∗ ∗ ∗
(2.4a) γ 1 : dh2 E ∧G dh0 E −→ dh1 E
and
(2.4b) γ 2 : dv2 ∗ E ∧G dv0 ∗ E −→ dv1 ∗ E.
The associativity and commutativity diagrams for γ 1 and γ 2 are the obvious ones. More
interesting is the compatibility condition of the two structures, which can be expressed as
follows.
Let Eh,k be the fiber of E over a generalized point (h, k) of H × K. Then the
6
That is, B H = N H[1], where H[1] is the groupoid with a single object e and Aut(e) = H.
compatibility condition reads
Eh,k Eh0 ,k Eh,k0 Eh0 ,k0 / Eh,k Eh,k0 Eh0 ,k Eh0 ,k0
1
γh,h 1 2 2
0 ;k γh,h0 ;k0 γh;k,k 0 γh0 ;k,k0
x &
(2.5) Ehh0 ,k Ehh0 ,k0 Eh,kk0 Eh0 ,kk0
2
γhh 1
γh,h
0 ;k,k0 0 ;kk0
* t
Ehh0 ,kk0
where we have suppressed the torsor contraction symbol. The horizontal arrow is the
canonical symmetry map swapping the factors in Tors(G), which exists whenever G is
abelian. A more intrinsic way to express the same thing is to observe that the above
compatibility condition amounts to the equality
∗ ∗
(2.6) γd2h ∗ E ◦ (dv2 ∗ γ 1 dv0 ∗ γ 1 ) = (1 × c × 1) ◦ γd1v1 ∗ E ◦ (dh2 γ 2 dh0 γ 2 )
1
as morphisms from (dh0 dv2 )∗ E ∧ (dh2 dv0 )∗ E ∧ (dh0 dv0 )∗ E ∧ (dh1 dv1 )∗ E to (dh1 dv1 )∗ E over H × H ×
K × K. c is the symmetry morphism which swaps the two terms in the middle.
With the straightforward notion of morphism, when H, K are also abelian, biextensions
of T form a Picard category Biext(H, K; G). The relative version of it, where we
consider the various Biext(HS , KS ; GS ) over a variable base S, is a Picard stack, denoted
Biext(H, K; G). As observed in Grothendieck [Gro72] and Breen [Bre83], Biext(H, K; G)
is biadditive in all three variables.
2.6. Schreier-Grothendieck-Breen theory of extensions. It is well known
that an extension E of K with a possibly nonabelian kernel G determines a morphism
 : E → Aut(G) from the action of E on G by conjugation (see, e.g. [ML95]). A refinement
of this situation is when G is part of a crossed module ∂ : G → Π. Since part of the crossed
module data is precisely a homomorphism Π → Aut(G), an extension of K by that crossed
module, or by (G, Π) for short, is a commutative diagram of group objects of the form
[Bre90, §8]
p
1 /G ı / E / K / 1
∂ 

Π
where the row is exact and ı(g)e = eı(g (e) ), for points g of G and e of E.
An equivalent characterization is that E is a (GK , ΠK )-torsor over K, equipped with
(G, Π)-torsor isomorphisms
∼
γk,k0 : Ek ∧G Ek0 −→ Ekk0
satisfying the standard associativity condition [Gro72, Bre90, AN10], such that the
equivariant structural morphism to Π is a homomorphism. By loc. cit. and [AN10], such
an extension corresponds to a group-like stack morphism F : K → [G → Π]∼ , where K
is identified with the (discrete) group-like stack. In this sense the above diagram is a
butterfly representing this morphism.
In all these characterizations the morphism  arises as the structural equivariant section
of the (G, Π)-torsor, whereas ı is the identification G ' E1 , where the (G, Π)-torsor E1 ,
isomorphic to the unit one, is the fiber over the unit section of K. This identification is
explicitly given as ı(g) = g e0 = e0 g, where e0 is the central section of E1 corresponding
to the unit section of G. Since  is a homomorphism, we obtain the commutativity of the
“wing” in the diagram above. The conjugation property satisfied by ı and  reflects the
structure of (G, Π)-torsor of E. In particular, we have that ı(g)e (resp. eı(g)), product in
E, agrees with the left (resp. right) G1 -action on E.
3. Biadditive morphisms
If H, K, G are groups of T, a biadditive (or bimultiplicative) morphism is a map f : H ×
K → G which is a homomorphism in each variable. For abelian groups this is none other
than a Z-bilinear morphism.
More generally, if H, K, and G are categories, which we assume to be group-like,
we say that a bifunctor is biadditive (or bimultiplicative) if it is monoidal in both
variables, in a compatible way.7 For this, we must also assume G be endowed with a
braiding c. More precisely, we have:
3.1. Definition. A bifunctor F : H × K → G is biadditive if:
1. it has the structure of additive functor with respect to each variable, namely there
exist functorial (iso)morphisms
λ1h,h0 ;k : F (h, k)F (h0 , k) −→ F (hh0 , k) and λ2h;k,k0 : F (h, k)F (h, k 0 ) −→ F (h, kk 0 )
satisfying the standard associativity conditions and compatibility with the braiding of
G;
2. for all objects h, h0 of H and k, k 0 of K we have a functorial commutative diagram
(3.1)
F (h, k)F (h0 , k) F (h, k 0 )F (h0 , k 0 )
ĉ / F (h, k)F (h, k 0 ) F (h0 , k)F (h0 , k 0 )

λ1h,h0 ;k λ1h,h0 ;k0 λ2h;k,k0 λ2h0 ;k,k0

F (hh0 , k)F (hh0 , k 0 ) / F (hh0 , kk 0 ) o F (h, kk 0 )F (h0 , kk 0 )
λ2hh0 ;k,k0 λ1h,h0 ;kk0
where the upper horizontal arrow is the “commuto-associativity” morphism

(x y) (z w) −→ (x z) (y w)
7
A better choice would be to use “bi-monoidal” in place of biadditive. The latter is motivated
by continuity with the naming convention in [AN09], where morphisms between gr-stacks are called
additive—a convention derived from the particular but significant Picard case.
arising from the braiding of G (an explicit definition can be found in [JP07]);
3. the two morphisms F (IH , IK ) → IG that can be deduced from the first condition
coincide.
The second condition ensures that the two possible ways to compute F (hh0 , kk 0 ) agree.
The obvious similarity with Diagram (2.5) will be exploited below.
The definition can be extended to the case where all three categories are braided. In
this case, we add to the biadditivity the condition that F be braided in each variable,
namely that the following diagrams commute:
λ1 / λ2 /
F (h, k)F (h0 , k) F (hh0 , k) F (h, k)F (h, k 0 ) F (h, kk 0 )
cF,F F (c,k) and cF,F F (h,c)

F (h0 , k)F (h, k) / F (h0 h, k) F (h, k 0 )F (h, k) / F (h, k 0 k)
λ1 λ2
A natural transformation ϕ : F ⇒ F 0 between two biadditive morphisms is a natural

transformation of bifunctors such that in each variable the conditions to be a natural
transformation of additive functors are satisfied. We denote by Hom(H, K; G) the
resulting groupoid of biadditive morphisms.
It is clear that the same definitions can be stated in the case of group-like stacks as op-
posed to categories. Similarly to the pointwise case, we obtain a groupoid Hom(H , K ; G )
of biadditive morphisms. For the latter, by way of a standard process of restricting to
variable base S, we obtain a stack Hom(H , K ; G ) arising from the various fiber categories
S → Hom(H |S , K |S ; G |S ).
4. Biextensions by braided crossed modules

We extend the notion of biextension by adapting it to include braided crossed modules as
coefficients.
Let G = (∂ : G → Π, {·, ·}) be a braided crossed module of T, and let H, K be groups
of T.
4.1. Definition. A biextension of H, K by (G, Π) is a (G, Π)H×K -torsor E over H × K
equipped with two partial composition laws ×1 (resp. ×2 ) making E into an extension of
HK (resp. KH ) by (G, Π)K (resp. by (G, Π)H )—cf. sect. 2.6. We require the composition
laws to be compatible in the manner described by a diagram formally identical to (2.5) in
sect. 2.5.
To be precise, in the last condition of the above definition we must specify that the
horizontal arrow in diagram (2.5), adapted to the present case,
(4.1) ĉ : (Eh,k ∧G Eh0 ,k ) ∧G (Eh,k0 ∧G Eh0 ,k0 ) −→ (Eh,k ∧G Eh,k0 ) ∧G (Eh0 ,k ∧G Eh0 ,k0 )
is given by the braiding of the stack G = [∂ : G → Π]∼ in a manner analogous to the
horizontal arrow of diagram (3.1) in Definition 3.1. A global version of the compatibility
condition is given by equation (2.6).
The two partial composition laws give rise to morphisms of (G, Π)-torsors
1 G ∼
γh,h 0 ;k : Eh,k ∧ Eh0 ,k −→ Ehh0 ,k
and
2 G ∼
γh;k,k 0 : Eh,k ∧ Eh,k0 −→ Eh,kk0
each satisfying the obvious associativity condition, relative to the relevant variables, which
can be obtained, mutatis mutandis, from refs. [Gro72, Bre83]. Again, a global version is
given by (2.4).
The morphism (4.1) can be explicitly computed through the braiding of G as follows.
4.2. Proposition. Let u, u0 , v, v 0 be points of Eh,k , Eh0 ,k , Eh,k0 , and Eh0 ,k0 , respectively.
Then we have
0
(4.2) (u ×1 u0 ) ×2 (v ×1 v 0 ) = (u ×2 v) ×1 (u0 ×2 v 0 ) {f (u0 ), f (v)}−f (v ) ,
where f is the equivariant section into the trivial Π-torsor.

Proof. Use Lemma 2.4 to express the morphism Eh0 ,k ∧G Eh,k0 → Eh,k ∧G Eh0 ,k . The right
action by f (v 0 ) arises from the relation between the left and right G-torsor structures of
E.
4.3. Example. A standard biextension E of H, K by G (all groups assumed commutative)
is a biextension in the sense specified above with coefficients in the complex G[1] : [G → 0].
The interchange law, when expressed in terms of points, takes the standard form
(u ×1 u0 ) ×2 (v ×1 v 0 ) = (u ×2 v) ×1 (u0 ×2 v 0 ).
Dropping the commutativity of the partial multiplication laws, E becomes a biextension

of the sort mentioned in the “variants” in [Gro72]. For such a biextension each extension
determined by the partial multiplication laws is a central extension.
4.4. Remark. We do not require the partial multiplication laws to be commutative, even
if the groups H and K are. If we do require the multiplication laws to be commutative,
then G must be abelian and the composite of ∂ : G → Π with Π → Aut(G) arising from
the crossed module structure, factors through zero. This follows from the fact that the
strict commutativity forces the braiding to be identically equal to the identity morphisms,
and hence, by Lemma 2.4, we have χ = 1. By evaluating on torsors of the form (G, x)
we obtain {·, ·} = 1 identically on Π × Π. Then from the properties (2.2) it follows at
once that G and Π must be abelian. As a consequence, the action of ∂g, for any g ∈ G,
is trivial, therefore the biextension reduces to one of the type mentioned in Example 4.3
above. See however below for a more appropriate notion of commutativity.
4.5. Definition. A morphism of biextensions of H, K by (G, Π) is a morphism of the

underlying (G, Π)-torsors compatible with the partial composition laws.
The compatibility with the composition laws means that there are commutative
diagrams of (G, Π)-torsors
γ1 γ2
Eh,k ∧G Eh0 ,k / Ehh0 ,k Eh,k ∧G Eh,k0 / Eh,kk0
ϕ∧ϕ ϕ and ϕ∧ϕ ϕ

0
Eh,k ∧G Eh0 0 ,k / 0
Ehh0 ,k
0
Eh,k 0
∧G Eh,k 0
/ 0
Eh,kk 0.
γ10 γ20
The trivial biextension is I = (GH×K , eΠ ). More generally, a biextension will be

considered trivial if it is isomorphic, in the sense specified above, to the trivial one. Such
an isomorphism amounts to the existence of a “central” section of the underlying (G, Π)-
torsor, namely a global s : H × K → E commuting with the left and right actions of G on
E, obtained as the image by ϕ of the unit section eG of I. Then the two above diagrams
show that s is a homomorphism in each variable.
With the above notion of morphism and trivial object, biextensions of H, K by G
form a pointed groupoid Biext(H, K; G). A standard argument based on descent shows
that over a variable base S, the collection of all Biext(HS , KS ; GS ) forms a stack, equally
pointed, denoted Biext(H, K; G). The question of whether either of Biext(H, K; G)
and Biext(H, K; G) possess a monoidal structure is more delicate, and it will require
additional commutativity properties from G, as discussed below.
The groupoid Biext(H, K; G) is evidently additive with respect to the last variable,
namely
Biext(H, K; G1 × G2 ) ' Biext(H, K; G1 ) × Biext(H, K; G2 ),
and the same holds for Biext(H, K; G). Additivity on the other two variables relies on
the commutativity of the partial multiplications for the same reason as [Bre83, §1.2], hence
it will not hold in our case.
Biextensions by (G, Π) have certain structural properties analogous to those of exten-
sions reviewed in sect. 2.6, more precisely, we have:
4.6. Lemma. If E → H × K is a biextension of H, K by (G, Π), then E is equipped
with a map  : E → Π which is a homomorphism for each partial composition law. For all
points e ∈ Eh,1 , or e ∈ E1,k , and g ∈ G we have the relation
(4.3) ı(g)e = eı(g (e) ).
and the diagram

p
G ı / E / H ×K
(4.4) ∂ 

Π
∼
commutes, where ı is the identification (G, eΠ ) → E1,1 as (G, Π)-torsors.
The identification of the fiber E1,1 above the unit (eH , eK ) of H × K with the unit
torsor I = (G, eΠ ) amounts to the definition of exactness of the row in (4.4).
Proof. The arguments are virtually the same as those of [Gro72], except for the part
concerning the morphism . Following loc. cit., we can use the facts about extensions
recalled in sect. 2.6 by working over the terminal object K (resp. H) when we view E as
an extension of HK (resp. KH ) by (G, Π)H (resp. (G, Π)K ).
With this in mind, let f = (, p) : E → ΠH×K , with  : E → Π, be the equivariant
morphism which is part of the (G, Π)H×K -torsor structure. As an extension of HK by
(G, Π)K , E has a structure of (G, Π)K ×K HK -torsor over HK , hence f : E → ΠK ×K HK '
Π × H × K is written as f = (K , p), where K : E → ΠK . We have K = (, pK ), where
pK (resp. pH ) is the composite of p with the projection to K. A similar picture holds
by exchanging H with K. Since K and H are homomorphisms for the corresponding
product structure,  is a homomorphism for both.
Let eH/K = (eH , idK ) be the unit section of HK and eK/H = (idH , eK ) that of KH .
Since the fibers E1,k = e∗H/K E and Eh,1 = e∗K/H E are respectively identified with GK and
GH as (G, Π)-torsors, we have the unit section eE/K (resp. eE/H ) in E1,k (resp. Eh,1 ). Now
let E1,1 = (eH , eK )∗ E be the fiber over the unit section (eH , eK ) of H × K. This is the
common pullback of both E1,k and Eh,1 . The same argument as in [Gro72, §2.2] shows
that the restriction eE/K and eE/H to (eH , eK ) agree, so let eE ∈ E1,1 be their common
value. From [Gro72, §2.1] eE is central, since both eE/K and eE/H are, and we define ı by
ı(g) = g eE = eE g. Note, ı is the common restriction of ıK and ıH respectively defined by
eE/K and eE/H .
Analogously to the case of extensions, the commutativity of the “wing” in (4.4) follows
from the equivariance of f , and hence of , having observed that (eE ) must be equal
to the unit of Π. Finally, the relation (4.3) follows from the definition of the bitorsor
structure underlying that of a (G, Π)-bitorsor (cf. the end of sect. 2.1).
4.7. Remark. In the converse direction to Lemma 4.6 a biextension of H, K by (G, Π)
can be recovered from the datum (E, ×1 , ×2 , ı, ), where p : E → H × K is an object of
T/H×K equipped with two partial composition laws satisfying the interchange law, and ı
and  satisfy the conditions of the lemma.
If E 0 is a second biextension, with maps ı0 : G → E and 0 : E → Π satisfying the
conditions of Lemma 4.6, a biextension morphism ϕ : E → E 0 makes the following diagram
ı0
ϕ
G
ı / E / E0
∂ 

Πo 0
commutative.
For a biextension which is globally of the form (GH×K , x), where x : H × K → Π

characterizes the equivariant structural morphism, we obtain a decomposition in terms of
a pair of nonabelian cocycles, which we briefly describe.
The partial multiplication laws are described by
0
(4.5a) (h, k, a)(h0 , k, a0 ) = (hh0 , k, g1 (h, h0 ; k)ax(h ,k) a0 ),
0
(4.5b) (h, k, b)(h, k 0 , b0 ) = (h, kk 0 , g2 (h; k, k 0 )bx(h,k ) b0 ).
The actions stem from the change from right to left action,
0
(h, k, a)(h0 , k, a0 ) = ((h, k, eG )a)(h0 , k, a0 ) = (h, k, eG )(a(h0 , k, a0 )) = (h, k, eG )(h0 , k, ax(h ,k) a0 ),
and similarly for the other one. From the associativity property and the fact that the
equivariant section GH×K → Π must be a homomorphism for both laws, we find that
(g1 , g2 , x) must satisfy a pair of nonabelian cocycle conditions:
00
(4.6a) g1 (hh0 , h00 ; k)g1 (h, h0 ; k)x(h ,k) = g(h, h0 h00 ; k)g1 (h0 , h00 ; k)
(4.6b) x(h, k)x(h0 , k) = x(hh0 , k) ∂g1 (h, h0 ; k)
and
00
(4.6c) g2 (h; kk 0 , k 00 )g2 (h; k, k 0 )x(h,k ) = g2 (h; k, k 0 k 00 )g2 (h; k 0 , k 00 )
(4.6d) x(h, k)x(h, k 0 ) = x(h, kk 0 ) ∂g2 (h; k, k 0 )
The cocycles (g1 , x) and (g2 , x) are not independent: from the compatibility between the
partial multiplication laws, using Proposition 4.2, we find:
0 0
(4.7) g2 (hh0 ; k, k 0 )g1 (h, h0 ; k)x(hh ,k ) g1 (h, h0 ; k 0 ) =
0 0 0 0
g1 (h, h0 ; kk 0 )g2 (h, k, k 0 )x(h ,kk ) g2 (h0 ; k, k 0 ){x(h0 , k), x(h, k 0 )}−x(h ,k ) .
These cocycles are a coboundary if there exists u : H × K → G such that:
(4.8a) u(hh0 , k) g1 (h, h0 ; k) = u(h, k)u(h0 , k)

(4.8b) u(h, kk 0 ) g2 (h; k, k 0 ) = u(h, k)u(h, k 0 )
(4.8c) x(h, k) = ∂ u(h, k).
5. Biadditive morphisms and biextensions

Let E be a biextension of H, K by (G, Π). Then E defines a biadditive morphism
FE : H × K −→ G ,
in the sense of Definition 3.1, where G is the group-like associated stack, by assigning to a
point (h, k) ∈ H × K the (G, Π)-torsor Eh,k . Indeed, it is easily seen that the isomorphisms
γ 1 and γ 2 plus the compatibility of the two composition laws expressed by (2.5) satisfy the
required conditions. It is also easily verified that a morphism of biextensions ϕ : E → E 0
induces a natural transformation (denoted with the same symbol)
ϕ : FE ⇒ FE 0 : H × K −→ G .
This establishes a functor
u : Biext(H, K; G) −→ Hom(H, K; G ),
and in fact one between stacks
u : Biext(H, K; G) −→ Hom(H, K; G ),
These functors are evidently fully faithful. In the converse direction, we have:
5.1. Proposition. Let F : H × K → G be a biadditive morphism, and let G have a
presentation by way of the braided crossed module G = (G, Π, ∂, {·, ·}). Then the pullback
of the sequence (2.1) along F determines a biextension E = EF of H, K by G.
Proof. Let E = (H × K) ×G Π be the pullback. Set p : E → H × K (resp.  : E → Π)
equal to the first (resp. second) projection. We claim that E is a biextension of H, K by G.
The sequence (2.1) can be seen as the universal extension by (G, Π). In particular,
Π can be identified with the universal G, Π)-torsor (the identity idΠ is tautologically the
structural equivariant section). This makes apparent that the pullback to H × K is a
(G, Π)-torsor with structural map . As for the partial multiplication laws, observe that,
with the same notation as sect. 2.5, the biadditivity of F amounts to a pair of natural
transformations
(F ◦ dh2 ) (F ◦ dh0 ) =⇒ (F ◦ dh1 ) : H × H × K −→ G ,

(F ◦ dh2 ) (F ◦ dh0 ) =⇒ (F ◦ dh1 ) : H × K × K −→ G ,
and the equality of two transformations (see eqns. (2.6) and (3.1))
(F dh0 dv2 ) (F dh2 dv0 ) (F dh0 dv0 ) (F dh1 dv1 ) =⇒ (F dh1 dv1 ) : H × H × K × K −→ G .
This translates into the required properties for E once we observe that the universality of Π
(as a (G, Π)-torsor over G ) implies that for any pair f, g : S → G we have an isomorphism
f ∗ Π ∧G g ∗ Π ' (f g)∗ Π of (G, Π)-torsors. Last, the morphism ı is the image, under a further
pullback along the unit section (eH , eK ) ∈ H × K, of ∂ : G → Π. It is clear ı and  have
the properties stated in Lemma 4.6.
In more down-to-earth terms, everything can be explicitly checked by writing out explicit
expressions for the points of the pullback. Thus, a point e ∈ E is a triple e = ((h, k), a, x),
where (h, k) ∈ H × K, x ∈ Π, and a : F (h, k) → (G, x). The G-action becomes evident by
looking at two points e, e0 in the same fiber. We must have a commutative diagram of
(G, Π)-torsors
a0 / (G, x0 )
F (h, k)
a g
%
(G, x)
where the vertical arrow, as a map of (G, Π)-torsors whose underlying G-torsor is trivial,
is completely determined by a point g ∈ G (the image of the unit section eG ) such that
x0 = x ∂g.
The situation for the left action is completely analogous, with x0 = ∂g x.

When (h, k) = (eH , eK ), the point eE = ((eH , eK ), l, eΠ ), where l is the unique morphism
coming from the third condition in Definition 3.1, is the central unit section of E. This
gives the morphism ı : G → E. The partial multiplications are obtained as follows. Let
e = ((h, k), a, x) and e = ((h0 , k), a0 , x0 ). Then ee0 = ((hh0 , k), a00 , xx0 ), where the morphism
a00 is obtained from the diagram
aa0 /
F (h, k)F (h0 , k) (G,
7
xx0 )
λ1hh0 ,k
a00
F (hh0 , k)
The second multiplication is defined in the same way. The interchange law immediately
follows from the above and (3.1).
Since it is clear that if ϕ : F ⇒ F 0 is a natural transformations of biadditive morphisms
by pullback we obtain a corresponding isomorphism EF → EF 0 of biextensions, the
construction of Proposition 5.1 provides functors
v : Hom(H, K; G ) −→ Biext(H, K; G)
and
v : Hom(H, K; G ) −→ Biext(H, K; G).
5.2. Proposition. The functors u and v are quasi-inverses.
Proof. We need to show that the functor u is essentially surjective.
For each F biadditive, we construct a morphism F → F 0 = FE , where E = (H×K)×G Π.
The biadditive morphism determined by E assigns to each (h, k) ∈ H ×K the (G, Π)-torsor
Eh,k consisting of pairs (f, x) such that
∼
f : F (h, k) −→ (G, x).
Having observed this, the proof proceeds in a manner identical to that of Proposition 4.4.2
in [AN09] (in particular, cf. 4.4.2.1 and 4.4.2.2).
To summarize the previous discussion, Hom(H, K; G ) and Biext(H, K; G) are equiva-
lent, pointed groupoids. The distinguished point is the trivial biextension, corresponding
to the trivial biadditive morphism sending (h, k) ∈ H × K to the trivial (G, Π)-torsor.
The same holds for Hom(H, K; G ) and Biext(H, K; G).
6. Symmetry properties of biextensions

It is well known (and see the discussion at the end of [AN10]) that the groupoid of
extensions of K by (G, Π) admits a monoidal structure if and only if the crossed module
∂ : G → Π (or equivalently the associated stack G ) is braided. This monoidal structure is
itself equipped with a commutativity isomorphism if and only if the braiding carried by G
is symmetric.
For biextensions the situation is more rigid. This is already apparent from the fact
that in the very definition of a biextension a braided structure is required in order to be
able to formulate the compatibility condition between the two multiplication laws.
If E and F are biextensions, we can use the braiding of G to construct the partial
multiplication laws in the expected manner:
(6.1)
ĉ γ 1 ∧µ1
(Eh,k ∧G Fh,k ) ∧G (Eh0 ,k ∧G Fh0 ,k ) →
− (Eh,k ∧G Eh0 ,k ) ∧G (Fh,k ∧G Fh0 ,k ) −−−→ Ehh0 ,k ∧ Fhh0 ,k ,
and similarly for the second partial multiplication.

6.1. Proposition. For any pair of biextensions E and F of (H, K) by (G, Π), the partial
multiplication laws (6.1) comprise a biextension structure on the contracted product (as
(G, Π)-torsors) E ∧G F if the braiding on the crossed module G = (G, Π, ∂, {−, −}) (or
equivalently G ) is symmetric.
Proof. With reference to (6.1), let us simply denote the first multiplication morphism
so obtained by γ 1 µ1 , and let γ 2 µ2 denote the second multiplication structure obtained in
exactly the same way.
For the compatibility condition between the two multiplications we must write dia-
gram (2.5) for γ 1 µ1 and γ 2 µ2 . To simplify notations and readability, let us suppress the
symbol ∧G and the associativity morphisms. Set: a = Eh,k , x = Fh,k , y = Eh0 ,k , z = Fh0 ,k ,
u = Eh,k0 , v = Fh,k0 , w = Eh0 ,k0 , b = Fh0 ,k0 . Also, let us use parentheses to denote the
various codomains of γ, µ, namely (ay) = Ehh0 ,k , (xz) = Fhh0 ,k , etc. Consider the following
diagram:
axyzuvwb / axuvyzwb

ayxzuwvb auxvywzb
" |
ayuwxzvb / auywxvzb
~
(ay)(xz)(uw)(vb) / (ay)(uw)(xz)(vb) (au)(yw)(xv)(zb) o (au)(xv)(yw)(zb)
, $ z r
Ehh0 ,kk0 Fhh0 ,kk0
The outer pentagon corresponds to the sought-after compatibility for the partial multipli-
cations of E ∧G F , whereas the little inner one is obtained by juxtaposing the compatibility
conditions for E and F . Thus the inner pentagon commutes. The triangles commute
by definition, since they just hold the definitions of γ 1 µ1 and γ 2 µ2 . The quadrangles are
commutative by inspection. This leaves the big hexagon in the middle. The position of
the external variables is unchanged throughout, hence the analysis of the diagram reduces
to the one in Lemma D.2. This proves the proposition.
Once the coefficient crossed module carries a symmetric braiding, we are provided
with a product of biextensions. An easy argument, based on the definition of the partial
multiplication laws in the proof of Proposition 6.1 and Lemma D.1, or alternatively [AN10,
§8.2], shows that the braiding provides a morphism
c : E ∧G F −→ F ∧G E
of biextensions, which is evidently symmetric. We summarize the previous discussion as

6.2. Proposition. For any H, K and G = (G, Π, ∂, {−, −}), Biext(H, K; G) (respec-
tively, Biext(H, K; G)) is a braided symmetric group-like groupoid (respectively, stack) if
and only if G carries a symmetric braiding.
By equivalence, the same statement holds for Hom(H, K; G ) and Hom(H, K; G ),
respectively; the resulting symmetric monoidal structure is given by the pointwise product
of functors (with respect to the monoidal structure of G ).
Let G be symmetric. If F1 , F2 : H × K → G correspond, via Proposition 5.2, to the
biextensions E1 and E2 , then it is not difficult to see that E1 ∧G E2 is the biextension
corresponding to F1 F2 : H × K → G , where F1 F2 is computed pointwise, i.e. for all h ∈ H
and k ∈ K we set (F1 F2 )(h, k) := F1 (h, k) F2 (h, k), where the juxtaposition on the right
indicates the monoidal structure in G . Thus, Proposition 5.2 is upgraded to an equivalence
of symmetric group-like groupoids or stacks. More precisely, we get:
6.3. Proposition. The assignment defined by u in Proposition 5.2 gives an equivalence
∼
u : Biext(H, K; G) −→ Hom(H, K; G )
(resp.
∼
u : Biext(H, K; G) −→ Hom(H, K; G ))
of symmetric group-like categories (resp. stacks).
7. Butterflies
∂ ∂
Let H• : H1 → H0 and K• : K1 → K0 be a pair of crossed modules, and let ∂ : G1 → G0
be a crossed module equipped with a braiding {−, −}. We denote by H , K , and G the
corresponding associated stacks.
7.1. Definition. A butterfly from H• ×K• to G• is a biextension E ∈ Biext(H0 , K0 ; G• )
equipped with trivializations of its pullbacks (∂, id)∗ E and (id, ∂)∗ E, i.e. maps s1 : H1 ×K0 →
E and s2 : H0 × K1 → E subject to the following conditions:
1. (Restriction) s1 and s2 agree on H1 × K1 : (id, ∂)∗ s1 = (∂, id)∗ s2 .
2. (Compatibility) For all (h, z) ∈ H1 × K0 and (y, k) ∈ H0 × K1 , and e ∈ Ey,z , s1 and

s2 satisfy:
s1 (h, z) ×1 e = e ×1 s1 (hy , z) ,
(7.1)
s2 (y, k) ×2 e = e ×2 s2 (y, k z ) .
A morphism ϕ : (E, s1 , s2 ) → (E 0 , s01 , s02 ) is a morphism of the underlying biextensions pre-

serving the trivializations. Butterflies form a pointed groupoid, denoted Biext(H• , K• ; G• ).
The distinguished object is the trivial biextension.
7.2. Remark. As noted, a trivialization of a biextension is given by a (central) global
section. In the previous definition the pullbacks are (∂, id)∗ E = (H1 × K0 ) ×H0 ×K0 E and
(id, ∂)∗ E = (H0 × K1 ) ×H0 ×K0 E, so, for example, a global section ŝ1 : H1 × K0 → (∂, id)∗ E
is equivalently given by a global map s1 : H1 × K0 → E. Same for s2 .
If we identify the sections corresponding to the trivializations with s1 : H1 × K0 → E
and s2 : H0 × K1 → E, condition 1 in the Definition implies that
s1 (h, ∂k) = s2 (∂h, k).
Let s : H1 × K1 → E be the resulting morphism. The next lemmas characterize these

objects in terms of explicit diagrams (see (7.2) and (7.5) below). Their shape justifies the
name.
7.3. Lemma. The trivializations s1 and s2 , as maps s1 : H1 × K0 → E and s2 : H0 × K1 →

E, render the following diagrams commutative
H1 × K0 G1 H0 × K1 G1
s1 ı s2 ı
% y % y
(7.2) (∂,id) E ∂ (id,∂) E ∂
p  p 
y % y %
H0 × K0 G0 H0 × K0 G0
with  ◦ s1 and  ◦ s2 equal to the trivial map (identically equal to the unit e0 of G0 ). In
addition, s1 and s2 have the following properties:8
1. (Multiplicative) s1 and s2 are multiplicative in each variable, namely
s1 (h, z) ×1 s1 (h0 , z) = s1 (hh0 , z), s1 (h, z) ×2 s1 (h, z 0 ) = s1 (h, zz 0 ),
(7.3)
s2 (y, k) ×1 s2 (y 0 , k) = s2 (yy 0 , k), s2 (y, k) ×2 s2 (y, k 0 ) = s2 (y, kk 0 );
2. (Central) For all g ∈ G1 , and whenever ∂h (resp. ∂k) is equal to the unit section of
H0 (resp. K0 ), we have:
s1 (h, z)ı(g) = ı(g)s1 (h, z),
(7.4)
s2 (y, k)ı(g) = ı(g)s2 (y, k).
Proof. The relations (7.3), as well as the fact that the compositions  ◦ s1 and  ◦ s2 must
be equal to the unit element, follow from the condition that the pullbacks (∂, id)∗ E and
(id, ∂)∗ E split as biextensions. Indeed, as it was previously described, the trivialization of,
say, (∂, id)∗ E amounts to an isomorphism of biextensions GH1 ×K0 ' (∂, id)∗ E. The image
of the unit central section of GH1 ×K0 defines the trivializing one for (∂, id)∗ E, and, hence,
the map s1 . In practice, the isomorphism is written (h, z, g) 7→ s1 (h, z)g, and the central
condition on the trivializing section means that
s1 (h, z) g = g s1 (h, z) ,
where the juxtaposition indicates the G1 -action. Of course, similar considerations hold for
s2 , as well.
Then  ◦ s1 and  ◦ s2 are trivial because, under the isomorphisms determined by s1
and s2 , they must indeed be equal to the equivariant section for the trivial bitorsor which
simply sends the unit e1 ∈ G1 to e0 ∈ G1 .
The relations (7.3) follow from the fact that the isomorphisms determined by s1 and
s2 are compatible with the partial multiplication laws. For example:
(s1 (h, z) g) ×1 (s1 (h0 , z) g 0 ) = s1 (h, z) ×1 (g s1 (h0 , z) g 0 ) = s1 (h, z) ×1 s1 (h0 , z) gg 0 ,
8
In some of the following formulas we explicitly write the product symbols ×1 and ×2 to avoid
ambiguities.
on the other hand in the trivial bitorsor (h, z, g) ×1 (h0 , z, g 0 ) = (hh0 , z, gg 0 ) is mapped to
s1 (hh0 , z) gg 0 , which proves the first of (7.3). The others are similar.
The relations (7.4) follow from the centrality and (4.3) of Lemma 4.6.
7.4. Remark. An alternative way to characterize a butterfly is to say that it consists
of a biextension E of (H0 , K0 ) by G• such that s1 and s2 provide it with the structure
of butterflies in the ordinary sense of [AN09] from (H• )K0 (resp. (K• )H0 ) to G• in a
compatible way.
It is clear that properties similar to those pertaining to s1 and s2 stated in Lemma 7.3
hold for their common restriction s : H1 × K1 → E. In particular, the following diagram
H1 × K1 G1
s ı
% y
(7.5) (∂,∂) E ∂
p 
y %
H0 × K0 G0
is commutative, with  ◦ s equal to the trivial map (identically equal to the unit e0 of G0 ).
In addition, s is multiplicative in each variable:
s(h, k) ×1 s(h0 , k) = s(hh0 , k)

(7.6)
s(h, k) ×2 s(h, k 0 ) = s(h, kk 0 ).
Relations (7.4) (collapsed into one) and (7.1) also hold. As an easy consequence we have
the following
7.5. Lemma. Let (E, s1 , s2 ) ∈ Biext(H• , K• ; G• ). The pullback (∂, ∂)∗ E is isomorphic
(via s) to the trivial biextension in Biext(H1 , K1 ; G• ).
7.6. Remark. The correspondence (E, s1 , s2 ) 7→ (E, s) determines a morphism
(∂,∂)∗
/ Biext(H1 , K1 ; G• )

Biext(H• , K• ; G• ) −→ H Ker Biext(H0 , K0 ; G• )
where H Ker denotes the homotopy kernel (recall both groupoids are pointed) which is not
an equivalence, in general. Requesting that the biextension E become trivial when pulled
back to H1 × K1 is a weaker condition for it does not provide for the two other pullbacks
to be trivializable.
7.7. Remark. A butterfly has a description in terms of cocycles if the underlying (G1 , G0 )-
torsor E is globally trivial as a right G1 -torsor. Therefore, as seen at the end of sect. 4, if E
has the form (H0 × K0 × G1 , x), with x : H0 × K0 → G0 , then a cocyclic description (4.5),
(4.6), (4.7) (with h, h0 ∈ H0 and k, k 0 ∈ K0 ) is available. A butterfly will described by a
cocycle consisting of that set plus additional relations for the trivializations of the two
pullbacks to H1 × K0 and H0 × K1 . (The required triviality conditions are the same as

those for a nonabelian 1-cocycles of the sort that appears in the theory of extensions.)
Since E = H0 × K0 × G1 , there exist u1 : H1 × K0 → G1 and u2 : H0 × K1 → G1 , such
that s1 and s2 are written as s1 (h, z) = (∂h, z, u1 (h, z)−1 ) and s2 (y, k) = (y, ∂k, u2 (y, k)−1 ).
The conditions in Lemma 7.3 then express the triviality of the cocycle pair (g1 , g2 ) when
pulled back to H1 × K0 and H0 × K1 , namely we get:
(7.7a) u1 (hh0 , z) g1 (∂h, ∂h0 ; z) = u1 (h, z) u1 (h0 , z)

(7.7b) u1 (h, zz 0 ) g2 (∂h; z, z 0 ) = u1 (h, z) u1 (h, z 0 )
(7.7c) x(∂h, z) = ∂ u1 (h, z)
and
(7.7d) u2 (yy 0 , k) g1 (y, y 0 ; ∂k) = u2 (y, k) u2 (y 0 , k)

(7.7e) u2 (y, kk 0 ) g2 (y; ∂k, ∂k 0 ) = u2 (y, k) u2 (y, k 0 )
(7.7f) x(y, ∂k) = ∂ u2 (h, z)
for all pairs (h, z), (h0 , z) ∈ H1 × K0 and (y, k), (y, k 0 ) ∈ H0 × K1 . Moreover, since the two
trivializations must agree when pulled back to H1 ×K1 , it follows that u1 (h, ∂k) = u2 (∂h, k).
Denoting this common restriction by u : H1 × K1 → G1 , the two previous sets coalesce
into
(7.8a) u(hh0 , k) g1 (∂h, ∂h0 ; ∂k) = u(h, k)u(h0 , k)

(7.8b) u(h, kk 0 ) g2 (∂h; ∂k, ∂k 0 ) = u(h, k)u(h, k 0 )
(7.8c) x(∂h, ∂k) = ∂ u(h, k).
for all h, h0 ∈ H1 and k, k 0 ∈ K1 .
8. Biadditive morphisms and butterflies

We now consider biadditive morphisms
F : H × K −→ G
with G braided (cf. sect. 3). As before, we assume we have presentations by crossed
modules H ' [H1 → H0 ]∼ , K ' [K1 → K0 ]∼ , and G ' [G1 → G0 ]∼ , the latter equipped
with a braiding structure {−, −}. Our purpose is to prove the following
8.1. Theorem. There is an equivalence of (pointed) groupoids
∼
u : Biext(H• , K• ; G• ) −→ Hom(H , K ; G ).
In essence, the theorem states that any biadditive functor F : H × K → G can be

represented by a butterfly involving the presentations. This equivalence is compatible with
restriction and base-change so that, via the usual mechanism of considering the above
equivalence relative to a variable object S of T, we obtain a corresponding equivalence
∼
u : Biext(H• , K• ; G• ) −→ Hom(H , K ; G ).
Following [AN09] and sect. 5, we exhibit a pair of quasi-inverse functors. We will essentially
confine ourselves to just exhibit the relevant definitions, for the methods are quite similar
to those in [AN09, §4], with the exception of biadditivity. The latter is discussed explicitly,
where the need arises.
Let F̄ be the composition of the projection π : H0 × K0 → H × K with F . It is
evidently biadditive. Consider the pullback E = (H0 × K0 ) ×F̄ ,G ,π G0 . By Proposition 5.1
it is a biextension.
8.2. Lemma. E defined above satisfies the conditions in the statement of Lemma 7.3.
Proof. We must verify the pullbacks of E to H1 ×K0 and H0 ×K1 are trivializable. This is
a direct consequence of the fact that H1 / H0 / H and K1 / K0 / K are (homotopically)
exact. Furthermore, the decomposition (∂, ∂) = (id, ∂) ◦ (∂, id) = (∂, id) ◦ (id, ∂) and
condition 3 of definition 3.1 ensure that the restriction condition in definition 7.1 is satisfied.
In explicit terms, a point of the pullback (∂, id)∗ E is given by a triple ((h, z), f, x)
where (h, z) ∈ H1 × K0 , x ∈ G0 and f is an isomorphism
f
F (π(∂h), π(z)) / (G1 , x).
We have π(∂h) = (H1 , ∂h), so there must be an isomorphism IH = (H1 , eh0 ) −→ (H1 , ∂h).
Thus, there is a chain of isomorphisms
f
F (IH , π(z)) / F (π(∂h), π(z)) / (G1 , x)

IG
so that we must have x = ∂g, g ∈ G1 . This provides an explicit trivializing section s1 for
(∂, id)∗ E. Similarly for the other one, s2 , and their common restriction s.
A computation based on the biadditivity of F and the same technique at the end of
the proof of 5.1 shows that both s1 and s2 are multiplicative, i.e. each satisfies condition 1
of lemma 7.3, with respect to both variables. Condition 2 also follows from a direct
calculation, as in [AN09, §4.3.6].
8.3. Proposition. There exists a functor v : Hom(H , K ; G ) → Biext(H• , K• ; G• )
defined by assigning to an object F the butterfly whose underlying bitorsor is E = (H0 ×
K0 ) ×F̄ ,G ,π G0 .
Proof. The pullback biextension construction is functorial (cf. the end of section 5). In
particular, if ϕ : F ⇒ F 0 is a morphism of biadditive functors, the resulting morphism of
biextensions is compatible with the trivializations. This observation proves Proposition 8.3.
To get a biadditive functor from a butterfly in functorial way, we proceed as in ref.

[AN09, §§4.3.2–4.3.4]. Specifically, an object of H × K will be represented by a pair of
torsors, as ((Y, y), (Z, z)) where (Y, y) is an (H1 , H0 )-torsor and (Z, z) an (K1 , K0 )-torsor.
Let (E, s) be an object of Biext(H• , K• ; G• ). We define a FE,s1 ,s2 : H × K → G by
assigning to the above pair the (G1 , G0 )-torsor (X, x) where:
X := HomH1 ,K1 (Y, Z; E)(y,z) , x : X −→ G0
e 7−→  ◦ e.
X consists of separately H1 or K1 -equivariant local lifts of (y, z) : Y × Z → H0 × K0 to
E. By this we mean that the lift of (y ∂h, z) is related to that of (y, z) by e(y ∂h, z) =
e(y, z) s1 (h, z). Similarly, e(y, z ∂k) = e(y, z) s2 (y, k). The notion is consistent thanks to
Conditions 1 and the multiplicativity of s1 and s2 . This construction is obviously functorial
with respect to each of its arguments. It gives the defining part of the
8.4. Proposition. There exists a functor
u : Biext(H• , K• ; G• ) −→ Hom(H , K ; G ),
whose value at (E, s1 , s2 ) is the bifunctor FE,s1 ,s2 defined above.
∼
Proof. We must verify the biadditivity property, namely that F (Y1 , Z) ∧G1 F (Y2 , Z) −→
F (Y1 ∧H1 Y2 , Z), which follows from the diagram:
γ1
/
4 E ×K0 E E
(e1 ,e2 )
(p,p) p
∆Z

Y1 × Y2 × Z / Y1 × Z × Y2 × Z / (H0 × K0 ) ×K0 (H0 × K0 ) / H0 × K0
It is easy (and left to the reader) to check that the diagram is invariant under replacing
(e1 g, e2 ) by (e1 , ge2 ) and (y1 h, y2 ) by (y1 , hy2 ). (The first arrow to the left is the diagonal
of Z followed by the swap of the two inner factors.)
8.5. Remark. A coordinate version of the construction of the functor u above is as
follows. According to the beginning of section 5, if E is the underlying biextension of an
object in Biext(H• , K• ; G• ), we obtain a biadditive morphism H0 × K0 → G by sending
the pair (y, z) to the (G1 , G0 )-torsor Ey,z . Since E is part of a butterfly, this construction
is compatible with morphisms in H and K (in fact, in the prestacks defined by the
presentations) because, if say y 0 = y ∂h, we have
∼
Ey0 ,z ←− Ey,z ∧G1 E∂h,z ←− Ey,z ,
since from the definition the existence of s1 implies E∂h,z is a trivial (G1 , G0 )-torsor.
(Similarly for Ey,∂k .) We have a similar calculation whenever z 0 = z ∂k, and the properties
of the butterfly (plus the compatibility of ×1 and ×2 ) ensure we obtain a unique morphism
Ey,z −→ Ey0 ,z0 ,
which allows us to define u(E) on more general objects by descent. The connection with
the global version above is of course that HomH1 ,K1 (Y, Z; E)(y,z) reduces to Ey,z when
(Y, y) = (H1 , y) and (Z, z) = (K1 , z).
Proof of Theorem 8.1. We prove that u and v are quasi-inverses. To this end, recall
that for a G1 -torsor P we have the isomorphism
∼
(8.1) HomG1 (G1 , P ) → P, m 7→ m(eG1 ),
[Gir71, Proposition III 1.2.7]. This extends to (G1 , G0 )-torsors by assigning to m : G1 → P

the element s(m(eG1 )) ∈ G0 (see [AN09, n. 4.4.2]). We apply this observation to both
v(u(E)) and u(v(F )), where (E, s1 , s2 ) is a butterfly with underlying biextension E and
F : H × K → G is biadditive. In the first case, a point of the biextension v(u(E)) is
given by a tuple ((y, z), f, x) with (y, z) ∈ H0 × K0 and x ∈ G0 , such that
∼
f : Ey,z −→ (G1 , x).
Considering the fiber over (y, z) we have the following chain of morphisms of (G1 , G0 )-
torsors:
∼ ∼
(8.2) v(u(E))y,z −→ HomG1 (G1 , Ey,z ) −→ Ey,z ,
where the first arrow sends (f, x) to f −1 , and the projection (f, x) → x, namely the
equivariant section, to  ◦ f −1 . Therefore we have obtained an isomorphism of biextensions
∼
v(u(E)) → E, by virtue of the result quoted at the beginning. This is (tautologically) an
∼
isomorphism of butterflies. For this, consider the composite isomorphism IG = (G1 , e) →
∼ ∼ ∼
E∂h,z → (G1 , x) (resp. IG = (G1 , e) → Ey,∂k → (G1 , x)) which provides the trivialization
of the pullback v(u(E)) to H1 × K0 (resp. H0 × K1 ), as in the proof of Lemma 8.2. The
resulting identifications of the pair (f, x) with an element g ∈ G1 is clearly compatible via
∼
the chain (8.2), with the trivialization (G1 , e) → Ey,z .
In the second case, v(F ) = (H0 × K0 ) ×G G0 and u(v(F )) is the biadditive morphism
that assigns to (y, z) ∈ H0 × K0 the (G1 , G0 )-torsor
f ∼
v(F )y,z = {(f, x)|F (π(y), π(z)) −→ (G1 , x)} −→ F (π(y), π(z)),
where the isomorphism on the right is by way of (8.1) and loc. cit. The pullback v(F )
has the required trivializations by Lemma 8.2, which are evidently compatible with
∼ ∼
F (π(∂h), π(z)) → (G1 , e) and F (π(y), π(∂k)) → (G1 , e), showing that the above iso-
morphism holds for general objects and is functorial, proving there is an isomorphism
u(v(F )) ' F .
8.6. Theorem. [Theorem 8.1, symmetric version] Let the monoidal structure of G be
symmetric. Then the equivalence in Theorem 8.1 extends to one of group-like groupoids or
stacks.
Proof Proof (Sketch). We need only check the trivializations. Suppose F , F 0 are
two biadditive morphisms and E, E 0 are the corresponding butterflies. Let us also use the
letters E, E 0 to denote the underlying torsors, and let s1 , s2 and s01 , s02 be the trivialization
morphisms of E and E 0 , respectively. Let us denote by F ∧ F 0 the biadditive morphism
H × K → G , (y, z) 7→ F (y, z)F 0 (y, z). (The unnamed juxtaposition denotes the product
structure in G .) Consider the bitorsor E ∧G E 0 , equipped with the biextension structure
of Proposition 6.1.
We claim that the pairs (s1 , s01 ) and (s2 , s02 ) give the two trivialization morphisms
turning the biextension E ∧G E 0 into a butterfly.
Indeed, the restriction condition 1 in Definition 7.1 is immediate. The compatibility
condition 2 can be verified by a simple computation, using the definition of the partial
multiplications from the proof of Proposition 6.1 and Lemma 2.4.
9. Commutative structures
With the equivalence between u : Biext(H• , K• ; G• ) → Hom(H , K ; G ) at our disposal,
we can discuss commutativity conditions for biextensions (cf. Remark 4.4 above).
9.1. Definition. Let H , K , and G all be equipped with a braiding structure. Then we
say that (E, s1 , s2 ) ∈ Biext(H• , K• ; G• ) (or simply E by abuse of language) is braided if
the biadditive morphism u(E) is braided in the sense of sect. 3.
In order to turn the definition into actual diagrams, we use the explicit variance of the
biextension E, Remark 8.5. Here the point of Remark 2.2 becomes relevant. Whereas the
morphism u(E) : H × K → G is covariant in each variable, the biextension E itself is a
(G0 , G1 )-torsor over H0 × K0 , and Tors(G1 , G0 ) is anti-equivalent to G .
Using the trivialization s1 : H1 × K0 → E and s2 : H0 × K1 → E, and the braidings
{−, −}H and {−, −}K for the presentations, we have morphisms η 1 and η 2 :
1 2
ηy,y 0 ;z : Ey 0 y,z −→ Eyy 0 ,z ηy;z,z 0 : Ey,z 0 z −→ Ey,zz 0
e 7−→ e ×1 s1 ({y 0 , y}H , z) e0 7−→ e0 ×2 s2 (y, {z 0 , z}K ).
Thus, for y, y 0 ∈ H0 and z, z 0 ∈ K0 the following diagrams must commute:

1
γy,y 2
γy;z,z
0 ;z 0
Ey,z ∧ G1
Ey0 ,z / Eyy0 ,z Ey,z ∧ G1
Ey,z0 / Ey,zz0
O O
cy,y0 ;z 1 cy;z,z0 2
(9.1) ηy,y 0 ;z and ηy;z,z 0

Ey0 ,z ∧G1 Ey,z / Ey0 y,z Ey,z0 ∧G1 Ey,z / Ey,z0 z
γy10 ,y;z 2
γy;z 0 ,z
In both diagrams the vertical arrow to the left comes from the braiding in G . The two
vertical arrows on the right express the functoriality with respect to the braiding structures
of H• and K• .
Alternatively, the directions in the right vertical arrows of both diagrams in (9.1) can
be restored if we interpret them as morphism in Tors(G1 , G0 ) arising from the morphisms
(H0 , yy 0 ) −→ (H0 , y 0 y) and (K0 , zz 0 ) −→ (K0 , z 0 z)
in Tors(H1 , H0 ) and Tors(K1 , K0 ), respectively. This requires writing η 1 and η 2 (going

in the opposite direction) in terms of
s1 ({y 0 , y}−1
H , z) and s2 (y, {z 0 , z}−1
K ).
Using Lemma 2.4 to express the braiding morphisms cy,y0 ;z and cy;z,z0 we arrive at the
expressions, valid for e ∈ Ey,z , e0 ∈ Ey0 ,z :
e0 ×1 e = e ×1 e0 ×1 s1 ({y 0 , y}−1 {(e), (e0 )}G ;

(9.2a) H , z)
and for e ∈ Ey,z , e0 ∈ Ey,z0 :
e0 ×2 e = e ×2 e0 ×2 s2 (y, {z 0 , z}−1 {(e), (e0 )}G .

(9.2b) K )
It follows that the (ordinary) butterfly corresponding to each variable must be braided in
the sense of [AN09, §7.4.1].
Theorems 8.1 and 8.6 specialize to this situation. We will use superscript ( )b to denote
the groupoids (or stacks) of braided biadditive morphisms and butterflies.
9.2. Theorem. [Theorems 8.1, 8.6 fully commutative case] Let K , H , and G be all
braided and have presentations by braided crossed modules. The equivalence u of Theo-
rem 8.1 restricts to an equivalence
∼
u : Biextb (H• , K• ; G• ) −→ Homb (H , K ; G )
of groupoids. Furthermore, if G is braided symmetric, u (from Theorem 8.6) becomes

an equivalence of group-like groupoids. Similar statements hold for Biextb (−, −; −) and
Homb (−, −; −).
As an application of these additional assumptions, we discuss whether Biext(H• , K• ; G• ),
Hom(H , K ; G ), etc. are biadditive in their variables. This means asking whether, for
example, we have an equivalence:
Homb (H1 × H2 , K ; G ) ' Homb (H1 , K ; G ) × Homb (H2 , K ; G ) ,
and a similar one relative to the variable K , as well as similar statements for Biext(H , K ; G ),
Biext(H• , K• ; G• ), and Hom(H , K ; G ).
We have observed that regardless the commutativity assumptions (but keeping G
braided symmetric) they always are biadditive in the third variable. Biadditivity in the
first and second variables only holds under the additional commutativity hypotheses. More
precisely, we have
9.3. Proposition. The groupoid Homb (−, −; −) is biadditive in all variables for symmet-
ric group-like stacks. (The same conclusion holds for Biextb (−, −; −), Biextb (−, −; −)
and Homb (−, −; −).)
Proof. We can use the same argument as [Bre83, §1.2]. Specifically, consider the
three (additive) functors di : H × H → H (d1 is additive since H is braided). Let
F : H × K → G be a biadditive morphism, and let E = v(F ) be the corresponding
butterfly. Then
λ1 : d∗2 F d∗0 F −→ d∗1 F
is a morphism in Homb (H × H , K ; G ). As in loc. cit. this follows from the compatibility
between λ1 and λ2 and the fact that λ1 is an additive morphism thanks to the (symmetric)
braiding. In view of the equivalence Homb (−, −; −) ' Biextb (−, −; −), the above
morphism corresponds to a morphism of biextensions
γ 1 : d∗2 E ∧G1 d∗0 E −→ d∗1 E
which is in fact a morphism of butterflies in Biextb (H• × H• , K• ; G• ), after one checks the
trivializations. From Proposition 6.1 we see that the braiding must be symmetric. Now,
for additive morphisms R, S : H 0 → H , we obtain a morphism
(R × I)∗ F (S × I)∗ F −→ ((RS) × I)∗ F
in Homb (H 0 , K ; G ), showing biadditivity in the first variable. The second variable is

treated analogously.
9.4. Remark. In the proof, if ER and ES are the butterflies corresponding to R, S : H 0 →
H , we obtain the morphism of butterflies in Biext(H•0 , K• ; G• ) is
(ER × I) ×H (ES × I) ×H ES ) × I ×H
G1 H1

H0 E ∧ H0 E −→ (ER ∧ H0 E
1 1 1
expressing the biadditivity in the first variable at the level of biextensions. Here I is
a shorthand for the diagram corresponding to the identity morphism. (This kind of
compositions is systematically studied in the next Part II.)
Part II
10. Multiextensions and compositions
The generalizations of the previous notions of biextension and biadditive morphism to the
case of n-variables is straightforward, see [Gro72, §2.10.2], [Bre99, §7]. Let (G, Π, ∂, {−, −})
be a braided crossed module. A multiextension, or n-extension, of (H1 , . . . Hn ) by
(G, Π) is a (G, Π)H1 ×···×Hn -torsor E over H1 ×· · ·×Hn equipped with n partial multiplication
laws ×1 , . . . , ×n , plus a compatibility relation of the type (2.5) for each pair (×i , ×j ).
Lemma 4.6 remains valid, as well as the analogs of Propositions 5.1 and 5.2. In fact,
the notion of butterfly can be extended to this case. Let us consider H1 , . . . , Hn and G ,
the latter equipped with a braiding as usual. There is an evident notion of n-additive
functor F : H1 × · · · × Hn → G , which can be defined by an appropriate generalization
of Definition 3.1. If we suppose that each Hi has a presentation Hi ' [Hi,1 → Hi,0 ]∼ ,
we can define a butterfly (or, more precisely, an n-butterfly) from H1,• × · · · × Hn,• to
G• as an n-extension E of (H1,0 × · · · × Hn,0 ) by G• equipped with n trivializations
si : H1,0 × · · · × Hi,1 × · · · × Hn,0 → E, each satisfying the conditions of Definition 7.1,
with the obvious modifications. Theorem 8.1 and its symmetric variant 8.6 generalize to
this case. Let us state this independently for future reference. Denoting by MExt the
groupoid of n-butterflies, we have
10.1. Theorem. There is an equivalence of (pointed) groupoids:
∼
u : MExt(H1,• , . . . , Hn,• ; G• ) −→ Hom(H1 , . . . , Hn ; G ),
sending a multi-extension E, to the n-additive functor u(E) that to the object (y1 , . . . , yn )
of H1 × · · · × Hn , where each yi is in the essential image of Hi,0 , assigns the (G, Π)-torsor
Ey1 ,...,yn . It is an equivalence of group-like groupoids whenever G is symmetric.
Once again, the axioms of Definition 7.1 ensure this is compatible with the descent,
ensuring u(E) is a well defined morphism. Just like for biextensions, the equivalence
is compatible with localization, with the stacks MExt and Hom in place of the global
groupoids.
Multi-additive functors can be composed in the following way. Let G , H1 , . . . , Hn ,
Ki,1 , . . . , Ki,mi , i = 1, . . . , n, be group-like stacks, with G and H1 , . . . , Hn braided. Let
F ∈ Hom(H1 , . . . , Hn ; G ), and for i = 1, . . . , n let Gi ∈ Hom(Ki,1 , . . . , Ki,mi ; Hi ). Then
if x1,1 , . . . , xn,mn , collectively denoted x1 , . . . , xm , are objects of K1,1 , . . . , Kn,mn , define, as
usual, F (G1 , . . . , Gn ) by
(10.1) F (G1 , . . . , Gn )(x1 . . . , xm ) := F (G1 (x1 , . . . ), . . . , Gn (. . . , xm )).
10.2. Proposition. The composition defined in (10.1) assigns to the tuple (F, G1 , . . . , Gn )
a well defined object F (G1 , . . . , Gn ) of Hom(K1,1 , . . . , Kn,mn ; G ). This composition is
associative.
An identical statement holds with Hom replaced by Hom.
Proof. The only thing to check is that the composition F (G1 , . . . , Gn ) satisfy the con-
ditions (3.1) for each pair (i, j) of indices within the list {1, . . . , m1 + m2 + · · · + mn },
where mk is the arity of Gk . There are two cases depending on whether the variables
corresponding to the pair (i, j) belongs to the same “slot,” say relative to Gk , or when i
6 l. We now indicate the
and j fall into two different slots, relative to Gk and Gl , with k =
main points of the verification, leaving the easy task of writing the complete diagrams to
the reader.
In the former case, the mechanics of the verification are completely captured by
considering the values n = 1, m1 = 2. First we write the pentagonal diagram corresponding
to (3.1) for F (G). In it, we use the functoriality of F to reduce the arrow
(F (G(x, y)) F (G(x0 , y))) (F (G(x, y 0 )) F (G(x0 , y 0 )))

(F (G(x, y)) F (G(x, y 0 ))) (F (G(x0 , y)) F (G(x0 , y 0 )))
to the arrow
F ((G(x, y) G(x0 , y)) (G(x, y 0 ) G(x0 , y 0 ))) / F ((G(x, y) G(x, y 0 )) (G(x0 , y) G(x0 , y 0 ))),
and then use the fact that G itself satisfies (3.1). For the latter case, the general situation
is captured by considering n = 2, m1 = m2 = 1, so we need to write the diagram (3.1) for
F (G1 , G2 ). For this, the interchange law for F gives us a unique morphism from
(F (G1 (x), G2 (y)) F (G1 (x0 ), G2 (y))) (F (G1 (x), G2 (y 0 )) F (G1 (x0 ), G2 (y 0 )))
to
F (G1 (x) G1 (x0 ), G2 (y) G2 (y 0 )).
Now the functoriality of F gives the morphism
F (G1 (x) G1 (x0 ), G2 (y) G2 (y 0 )) −→ F (G1 (xx0 ), G2 (yy 0 ))
by way of the square
F (G1 (x) G1 (x0 ), G2 (y) G2 (y 0 )) / F (G1 (x) G1 (x0 ), G2 (yy 0 ))

F (G1 (xx0 ), G2 (y) G2 (y 0 )) / F (G1 (xx0 ), G2 (yy 0 ))
The statement about associativity is immediate.
11. Compositions of n-butterflies

From Theorem 10.1, the multilinear functors can be expressed in terms of multi-extensions,
and from Proposition 10.2 multilinear functors can be composed. Hence we expect an
analogous composition exists for multi-extensions. In this section we provide a construction
of this composition, generalizing the composition for single butterflies found in [AN09,
§5.1].
For i = 1, . . . , n, and integers ji = 1, . . . , mi , let G• , H1,• , . . . Hn,• and Ki,1,• , . . . , Ki,mi ,•
crossed modules, of which G• , H1,• , . . . Hn,• are assumed to be braided. Let E be a braided
n-butterfly from H1,• × · · · × Hn,• to G• , and for i = 1, . . . , n let Fi be an mi -butterfly
from Ki,1,• × · · · × Ki,mi ,• to Hi,• . In the following we let
P = (F1 × · · · × Fn ) ×
H1,0 ×···×Hn,0
E,
be the (H1,1 × · · · × Hn,1 )-equivariant (G1 , G0 )-torsor over F1 × · · · × Fn .

11.1. Theorem. The (G1 , G0 )-torsor P has a well defined quotient by H1,1 × · · · × Hn,1 ,
denoted
H1,1 ×···×Hn,1
(F1 × · · · × Fn ) ×
H1,0 ×···×Hn,0
E,
which carries a structure of m1 + · · · + mn -extension (=butterfly) from K1,1,• × · · · × Kn,mn ,•

to G• .
11.2. Definition. The composition E(F1 , . . . , Fn ) of E with F1 , . . . , Fn is
H1,1 ×···×Hn,1
E(F1 , . . . , Fn ) := (F1 × · · · × Fn ) ×
H1,0 ×···×Hn,0
E.
We call this composition the juxtaposition product, since it entails placing the butterflies
wing-by-wing next to one another.
Proof of Theorem 11.1–Construction of the juxtaposition product. We will use
element notation throughout. The procedure consists of several steps. The first is to form
the fiber product exactly as in loc. cit.
P = (F1 × · · · × Fn ) ×
H1,0 ×···×Hn,0
E,
which as noted is an (H1,1 × · · · × Hn,1 )-equivariant (G1 , G0 )-torsor over F1 × · · · × Fn ,

and to mod out the left and right actions of H1,1 × · · · × Hn,1 . Let us use the notation
(y1 , . . . , yn ) := (1 (v1 ), . . . n (vn )) = p(u)
for the tuple in H1,0 × · · · × Hn,0 , where i : Fi → Hi,0 for each (Hi,1 , Hi,0 )-torsor Fi ,
i = 1, . . . , n.
Consider first the right action of elements of the form (1, . . . , hi , . . . , 1), where hi ∈ Hi,1 .
The action on P is given by sending a point (v1 , . . . , vn , u) to
(v1 , . . . , vi hi , . . . vn , u ×i si (y1 , . . . , hi , . . . , yn )),
For the left action, we have
(11.1) (v1 , . . . , hi vi , . . . vn , si (y1 , . . . , hi , . . . , yn ) ×i u)

= (v1 , . . . , vi hyi i , . . . vn , u ×i si (y1 , . . . , hyi i , . . . , yn )),
which follows from Definition 7.1, adapted to the multi-variable case. These are actions
thanks to the multiplicativity properties of sj established in Lemma 7.3. These actions are
compatible for i 6= j. Acting with hi and then with hj , assuming for example that i < j,
we have

u ×i si (y1 , . . . , hi , . . . , yn ) ×j sj (y1 , . . . , yi ∂hi , . . . , hj , . . . , yn ) =

u×i si (y1 , . . . , hi , . . . , yn ) ×j sj (y1 , . . . , yi . . . , hj , . . . , yn )×i sj (y1 , . . . , ∂hi , . . . , hj , . . . , yn ) ,
where we used the multiplicativity property of sj . Now, using the interchange property for
×i and ×j , and the compatibility between the trivializations si and sj , the right hand side
becomes

u ×j sj (y1 , . . . , hj , . . . , yn ) ×i si (y1 , . . . , hi , . . . , yn ) ×j si (y1 , . . . , hi , . . . , ∂hj , . . . , yn ) ,
which coincides with the action of hj first, followed by that of hi . This ensures that
following formula for the right action of a generic point (h1 , . . . , hn ) is well defined:

(11.2) v1 h1 , . . . vn hn , · · · u ×1 s1 (h1 , y2 , . . . , yn ) ×2 s2 (y1 ∂h1 , h2 , . . . , yn ) · · ·

×n sn (y1 ∂h1 , . . . , yn−1 ∂hn−1 , hn ) .
The reader will have no difficulty in writing the corresponding formulas for the left action.
Since each Fi is an (Hi,1 , Hi,0 )-torsor, the action is free.
Note also that the G1 -actions on P (both left and right) are compatible with the action
of H1,1 × · · · × Hn,1 thanks to the relations (7.4) in Lemma 7.3 (property 2). In particular,
the G1 -actions happen by way of those on the last element of the tuple; denoting the class
of a tuple by brackets we have:
[v1 , . . . , vn , u] g = [v1 , . . . , vn , u g],

g [v1 , . . . , vn , u] = [v1 , . . . , vn , g u] = [v1 , . . . , vn , u g u ].
Finally, the equivariant section from the quotient of P to G0 is defined to be
(v1 , . . . , vn , u) 7−→ (u).
Next, we define m1 + · · · + mn partial product structures ×i,j for i = 1, . . . , n and

j = 1, . . . , mi as follows. Using our established index convention, let us use the notation
(z1,1 , . . . , zi,j , . . . , zn,mn ) ∈ K1,1,0 × · · · × Ki,j,0 × · · · × Kn,mn ,0
for a point of the base of P/(H1,1 × · · · × Hn,1 ). To begin with, consider the special case
of two points (v1 , . . . , vj , . . . , vn , u) and (v1 , . . . , vj0 , . . . , vn , u0 ) of P such that
pi (vi ) = (zi,1 , . . . , zi,j , . . . , zi,mi ), pi (vi0 ) = (zi,1 , . . . , zi,j

0
, . . . , zi,mi ),
in Ki,j,0 , for fixed i and j ∈ {1, . . . , mi }, everything else being equal. Define:
(11.3) (v1 , . . . , vj , . . . , vn , u)×i,j (v1 , . . . , vj0 , . . . , vn , u) := (v1 , . . . , vj ×i,j vj0 , . . . , vn , u×i u0 ).
The symbol ×i,j on the right hand side of (11.3) denotes the j th product structure (within
j = 1, . . . , mi ) of Fi , and the resulting point of P projects onto the point
0
(z1,1 , . . . , zi,j zi,j , . . . , zn,mn ).
In general, let us consider points e = [v1 , . . . , vn , u] and e0 = [v10 , . . . , vn0 , u0 ] of P/(H1,1 ×
0
· · · × Hn,1 ) above (z1,1 , . . . , zi,j , . . . , zn,mn ) and (z1,1 , . . . , zi,j , . . . , zn,mn ), respectively. (As
0
before, only the zi,j and zi,j coordinates are different.) For j = 1, . . . , n, j 6= i, there exist
unique hj ∈ Hj,1 such that vj0 = vj hj . As a result,
[v10 , . . . , vn0 , u0 ] = [v1 , . . . , vi0 , . . . , vn , u00 ],
where u00 is related to u0 by an application of (11.2). Then we define e ×i,j e0 as the class
[v1 , . . . , vj ×i,j vj0 , . . . , vn , u ×i u00 ],
which is computed using (11.3) above. We must show that this is independent of the
various choices involved through the use of (11.2). The computation is quite elaborate,
but otherwise not illuminating nor eventful, therefore we omit it. We also omit the easy
verification that each of these partial multiplication laws is associative.
To claim that we have constructed a genuine multi-extension, we must prove the partial
multiplication laws just defined obey pairwise compatibility (interchange) laws. This we
prove explicitly. Like in the proof of Proposition 10.2, there are two distinct cases to
address, depending on whether the two partial product laws have the same first index.
The easiest is when they do not, so we treat it first. Thus, let i 6= k ∈ {1, . . . n}. For
brevity let Q = P/(H1,1 × · · · × Hn,1 ) and consider:
ei,j = [v1 , . . . , vi , . . . , vk , . . . , vn , ui ] ∈ Qz1,1 ,...,zi,j ,...,zk,l ,...,zn,mn

e0i,j = [v1 , . . . , vi0 , . . . , vk , . . . , vn , u0i ] ∈ Qz1,1 ,...,zi,j
0 ,...,z
k,l ,...,zn,mn
ek,l = [v1 , . . . , vi , . . . , vk0 , . . . , vn , uk ] ∈ Qz1,1 ,...,zi,j ,...,zk,l

0 ,...,z
n,mn
e0k,l = [v1 , . . . , vi0 , . . . , vk0 , . . . , vn , u0k ] ∈ Qz1,1 ,...,zi,j

0 ,...,z 0 ,...,z
k,l n,mn .
We first compute the products with ×i,j :
ei,j ×i,j e0i,j = [v1 , . . . , vi ×i,j vi0 , . . . , vk , . . . , vn , ui ×i u0i ],

ek,l ×i,j e0k,l = [v1 , . . . , vi ×i,j vi0 , . . . , vk0 , . . . , vn , uk ×i u0k ].
Then
(ei,j ×i,j e0i,j ) ×k,l (ek,l ×i,j e0k,l )

= [v1 , . . . , vi ×i,j vi0 , . . . , vk ×k,l vk0 , . . . , vn , (ui ×i u0i ) ×k (uk ×i u0k )]
= [v1 , . . . , vi ×i,j vi0 , . . . , vk ×k,l vk0 , . . . , vn , (ui ×k uk ) ×i (u0i ×k u0k )
0
{(u0i ), (uk )}−(uk ) ]
0
= (ei,j ×k,l ek,l ) ×i,j (e0i,j ×k,l e0k,l ) {(u0i ), (uk )}−(uk ) ,
where in the third line we have used the interchange law for the ith and k th product laws
in E, Proposition 4.2.
The other case is the one with the same first index, say ×i,j and ×i,l , for j 6= l. In this
situation we consider instead:
ei,j = [v1 , . . . , vi , . . . , vn , ui ] ∈ Qz1,1 ,...,zi,j ,...,zi,l ,...,zn,mn

e0i,j = [v1 , . . . , vi0 , . . . , vn , u0i ] ∈ Qz1,1 ,...,zi,j
0 ,...,z ,...,z
i,l n,mn
ei,l = [v1 , . . . , wi , . . . , vn , ti ] ∈ Qz1,1 ,...,zi,j ,...,zi,l

0 ,...,z
n,mn
e0i,l = [v1 , . . . , wi0 , . . . , vn , t0i ] ∈ Qz1,1 ,...,zi,j

0 ,...,z 0 ,...,z
i,l n,mn .
Proceeding in an analogous way we first get:
ei,j ×i,j e0i,j = [v1 , . . . , vi ×i,j vi0 , . . . , vk , . . . , vn , ui ×i u0i ],

ek,l ×i,j e0k,l = [v1 , . . . , wi ×i,j wi0 , . . . , vn , tk ×i t0k ].
And then:
(ei,j ×i,j e0i,j ) ×i,l (ei,l ×i,j e0i,l )

= [v1 , . . . , (vi ×i,j vi0 ) ×i,l (wi ×i,j wi0 ), . . . , vn , (ui ×i u0i ) ×i (ti ×i t0i )]
− (wi0 )
= [v1 , . . . , (vi ×i,j vi0 ) ×i,l (wi ×i,j wi0 ) {i (wi ), i (vi0 )}Hii , . . . , vn ,
0
0 −i (wi )
(ui ×i u0i ) ×i (ti ×i t0i )×i si (y1 , . . . , {i (wi ), i (vi )}Hi , . . . , yn )]
0 0
= [v1 , . . . , (vi ×i,l wi ) ×i,j (vi ×i,l wi ), . . . , vn ,
−(t0 )
(ui ×i ti ) ×i (u0i ×i t0i ) {(u0i ), (ti )}G i ],
where in the intermediate equality we used the action by an element of Hi,1 , whereas in
the last one the interchange law (within the multi-extension Fi ) and (9.2) were used. But
the last line equals
−(t0i )
(ei,j ×i,l ei,l ) ×i,j (e0i,j ×i,l e0i,l ) {(u0i ), (ti )}G ,
as wanted.
The final step is to show that Q, in addition to being a multi-extension of K1,1,0 ×
· · · × Kn,mn ,0 by G• , is in fact a butterfly, that is, it carries a trivialization for each of its
pullbacks to K1,1,0 × · · · × Ki,j,1 × · · · × Kn,mn ,0 . Such trivializations can be defined as
follows.
For each i = 1, . . . , n and for each j = 1, . . . , mi , let si,j be the trivialization si,j : Ki,1,0 ×
· · · × Ki,j,1 × · · · × Ki,mi ,0 → Fi . Observe that the fiber of P above
(v1 , . . . , si,j (zi,1 , . . . , ki,j , . . . , zi,mi ), . . . , vn ) ∈ F1 × · · · × Fi × · · · × Fn
(where vk is a point of Fk , for k 6= i) can be identified with G1 . This is because that point
maps via (1 , . . . , n ) to (y1 , . . . , 1, . . . , yn ), and the fiber of E above it is identified with G1 .
(In turn, this follows from the fact that biextensions in particular, and multi-extensions in
general, are canonically trivialized over identity sections, see sect. 4.) It follows that the
fiber of Q over (z1,1 , . . . , ∂ki,j , . . . , zn,mn ) is similarly identified with G1 . Thus, we obtain
a trivializing section ŝi,j of the pullback of Q to K1,1,0 × · · · × Ki,j,1 × · · · × Kn,mn ,0 by
sending the point (z1,1 , . . . , ki,j , . . . , zn,mn ) to [v1 , . . . , si,j (zi,1 , . . . , ki,j , . . . , zi,mi ), . . . , vn , ei ],
where ei is the unit (central) section of the restriction of E to H1,0 × . . . {1} × · · · × Hn,0 .
Observe this does not depend on the choice of vj ∈ Fj , for j 6= i, since Fj /Hj is just
(zj,1 , . . . , zj,mj ). Note that  : Q → G0 composed with any of ŝi,j is identically equal to 1,
since so is the result of applying E : E → G0 to ei . The conditions (Compatibility and
Restriction) of Definition 7.1 are easy to verify and so this task is left to the reader.
This ends the proof.
11.3. Proposition. The composition in Definition 11.2 gives rise to a well-defined
composition functor
MExt(K1,1,• , . . . ; H1,• ) × · · · × MExt(. . . , Kn,mn ,• ; Hn,• ) × MExt(H1,• , . . . , Hn,• ; G• )

−→ MExt(K1,1,• , . . . , Kn,mn ,• ; G• ),
where we are using the same convention for the indices as above and in sect. A. Same
with MExt in place of MExt.
Proof. We need only check that if f : E → E 0 and gi : Fi → Fi0 are butterfly isomorphisms,
then we get a well defined isomorphism of k = m1 + · · · + mn -butterflies
f (g1 , . . . , gn ) : E(F1 , . . . , Fn ) −→ E 0 (F10 , . . . , Fn0 ).
This is easy to check from the construction of the juxtaposition product provided above.
Associativity only holds up to isomorphism. This is due to the same phenomenon,
ultimately a manifestation of the same lack of strict associativity for fiber products,
observed in the unary case (see [AN09, Noo05]). We omit the proof.
11.4. Proposition. The composition in Definition 11.2 is associative up to coherent
isomorphism.
12. The bi-multicategories of group-like stacks and crossed modules

By using Proposition 10.2 we can rely on the framework recalled in the technical section A
to promote the 2-category of braided group-like stacks to a 2-multicategory. In particular,
using Definition A.1, we define the categories {H , G }, where to each n we assign
Hom(H n ; G ) = Hom(H , . . . , H ; G ).
| {z }
n−times
We ambiguously use the same symbol to denote the stack obtained using Hom in place of
Hom, as which version will be in use will be clear from the context. From Definition A.3
we immediately obtain:
12.1. Proposition. For group-like stacks K , H , and G , we have the assembly map
(12.1) ◦ : {H , G } o {K , H } −→ {K , G }
which to the object (F ; G1 , . . . , Gn ) assigns the composition F (G1 , . . . , Gn ).
In particular, M G := {G , G } is a monoid for this composition, whenever G is braided,
with identity object given by (id; ) (the empty string in the second slot).
12.2. Remark. As alluded in sect. A, the formalism includes the symmetry structure,
which makes it slightly more general than needed in the sequel. All objects of interest shall
have augmentations factoring through the discrete subcategory N of S, which amounts to
ignoring the permutation structure and hence symmetry conditions.
From Propositions 11.3 and 11.4 and the formalism of sect. A the bicategory of braided
crossed modules of T equipped with butterflies as morphisms (see [AN09, Theorem 5.3.6])
is promoted to a genuine bi-multicategory. In particular, we obtain objects {H• , G• } and
an analog of Proposition 12.1 with assembly map
(12.2) ◦ : {H• , G• } o {K• , H• } −→ {K• , G• }
for braided crossed modules K• , H• , and G• .
The equivalence between the 2-category BGrSt of (braided) group-like stacks and the
bicategory BXMod of (braided) crossed modules (cf. [AN09, Theorem 5.3.6]) lifts to
the present case. Let MBGrSt (resp. MBXMod) the bi-multicategory with morphisms
Hom(H1 , . . . , Hn ; G ) (resp. MExt(K1,1,• , . . . , Kn,mn ,• ; G• )).
12.3. Theorem. There is an equivalence of bi-multicategories
∼
Ma : MBXMod −→ MBGrSt
induced by the associated stack functor.
Proof. Recall that a functor F : M C → MD between bi-multicategories is an equivalence
if for each tuple (x; y1 , . . . , yn ) of objects of MC the functor
F(x;y1 ,...,yn ) : HomC (y1 , . . . , yn ; x) → HomD (F (y1 ), . . . , F (yn ); F (x))
is an equivalence of categories, and the underlying ordinary homomorphism [F ]1 : C → D
is essentially surjective.9
The associate stack functor at the level of the underlying bicategories a : BXMod →
BGrSt is an equivalence, hence, in particular, essentially surjective. The equivalence is
proved in [AN09, Theorem 5.3.6] without the braiding hypothesis, however Theorem 4.3.1
in loc. cit. applies to the braided case as well by §7, ibid., and presentations by braided
crossed modules are easily obtained by using Lemma 2.4. Thus essential surjectivity
follows.
The equivalence at the level of the multi-arrow categories directly follows from Theo-
rem 10.1. Finally we check that the functor u of Theorem 10.1 preserves the (horizontal)
compositions up to coherent isomorphism. Like the single variable case, this follows from
the fiber product construction of the quasi-inverse to u.
9
[F ]1 is the homomorphism we obtain by restricting F to arrows of arity equal to one.
12.4. Remark. The situation becomes decidedly more pleasant if we confine ourselves
to symmetric objects. We can consider a symmetric variant of the above based on the
bicategories SXMod and SGrSt. The equivalence in Theorem 12.3 restricts to between
these new entities:
∼
Ma : MSXMod −→ MSGrSt,
which, by Theorem 10.1, both ought to be regarded as multicategories enriched over
symmetric group-like groupoids. Furthermore, using the version of {−, −} based on Hom,
SGrSt becomes closed. Indeed one has the analog of [Kel72, Theorem 2], in that
HomC (K o H , G ) ' HomC (K , {H , G }),
where C = SGrSt (or even C = SGrSt/S using extraordinary structure briefly discussed
in sect. A—see [Kel72] for all the details). A proof can be obtained along the same lines,
proceeding from Theorem 10.1 and sect. A. We will not pursue this further.
Part III
13. Bimonoidal structures and weakly categorical rings
For a braided stack R, a biadditive bifunctor m : R × R → R gives a binary law which
is automatically distributive with respect to the monoidal structure of R by virtue of
(3.1) in Definition 3.1. If m is itself part of a monoidal structure, that is, there exists
µ : m(m, id) ⇒ m(id, m) satisfying the standard pentagon identity, then R is said to
be bimonoidal. In fact it satisfies the axioms of a categorical ring in the usual sense
[JP07], except possibly the requirement for the underlying group-like structure of R to
be symmetric. (One axiom was apparently missing in [JP07], and the gap was filled in
[Qua13], thus completing the classification.)
From sections 10 and A, the binary operation m and id are objects of MR = {R, R},
and µ : m(m, id) ⇒ m(id, m) is a morphism of MR, with the pentagon identity expressing
the equality between two morphisms thereof. Rather than using the classical, “biased,”
definition of the monoidal structure, it is more convenient to exploit the multi-categorical
structure comprising all the multilinear functors. Now, biased and unbiased definitions yield
equivalent structures [Lei04], hence we can simply define it in terms of the multicategory
MBGrSt.
13.1. Definition. [See Definition B.1] Let R be a braided stack. A (weak) categorical
ring structure on R is a lax monoidal functor (t, θ) : N → MR. A (weak) ring-like
stack is a braided stack R equipped with such a structure, namely a pseudo-monoid in the
2-multicategory MBGrSt.10
10
In the sequel we shall use “weak”, when referring to bimonoidal structures, in this sense; it will not
refer to the laxness of the monoidal structures.
In plainer language, N represents all the n-ary operations in a “disembodied” form.

Thus, an object n of N corresponds to an operation with n abstract variables. The functor
t assigns to n an actual t(n) : R × · · · × R → R in MExt(R n ; R). Given n operations
of various arities, they can be substituted into one of arity n, as hinted above. This set
of operations is represented by the object (n; m1 , . . . , mn ) of N o N (see Definitions A.3
and B.1) and the substitution, resulting in an operation of arity m1 + · · · + mn , gives a
morphism N o N → N which has the properties of a monoidal structure. This is mapped
by t into the composition of actual multi-additive multi-functors of R, satisfying the same
combinatorial structure. In summary, the functor (t, θ) gives a collection of multi-additive
t(n) subject to isomorphisms of functors of the form
θn;m1 ,...,mn : t(m1 + · · · + mn ) → t(n)(t(m1 ), . . . , t(mn )),
controlling the compositions.

13.2. Remark. The 2-multicategory MBGrSt has of course more structure than the
abstract situation discussed in sections A and B. In particular, the objects of BGrSt,
being themselves categories, have an internal structure, which we can use to define the
wreath product N o G for any G following sect. A, or, literally, the arguments of [Kel72,
§2.1]. Briefly, an object of N o G consists of a string (n; x1 , . . . , xn ) (where the label n is in
effect redundant) with xi an object of G , i = 1, . . . , n. There are no morphisms between
objects with different integer labels, and a morphism (n; x1 , . . . , xn ) → (n; x01 , . . . , x0n ) is
simply an n-tuple of morphisms in G .
It follows that by a special case of the adjunction discussed in [Kel72], which in the
present case can be verified by hand, the lax monoidal functor providing the exterior
monoidal structure on R corresponds to
(t̃, θ̃) : N o R −→ R,
which is required to satisfy the diagrams displayed in Remark B.3, which define the
structure of a pseudo algebra over N.
We drop the “weak” attribute is the underlying braiding is symmetric. In such a case,
we will see the fibers of R are categorical rings in the usual sense, so that R becomes the
stack analog of the categorical rings described in e.g. [JP07].
13.3. Remark. For a symmetric R, we can define the ring-like structure in the same way,
but working in SGrSt. In such case the pseudo-algebra structure of R can be verified
directly from the adjunction mentioned in Remark 12.4 above.
A pseudo monoid R in MBGrSt ought to correspond to one in MBXMod. More
precisely, any pseudo monoid R ought to be equivalent to one whose underlying stack is
the one associated to a presentation by a braided crossed module, i.e. an object of BXMod.
This will be made precise below. First, we can write the analog of Definition 13.1, namely:
13.4. Definition. A (weakly) ring-like crossed module is a pseudo monoid in the bi-
∂
multicategory MBXMod, that is, a braided crossed module R• : R1 → R0 equipped with a
pseudo monoidal functor
(t, θ) : N −→ MR• .
We drop the “weak” attribute if the underlying braiding is symmetric.
An operation of arity n is realized in this case by an n-butterfly En = t(n), an object
of Hom(R•n ; R• ), composed according to (B.2), realized by the juxtaposition product
(13.1) θn;m1 ,...,mn : Em1 +···+mk −→ En (Em1 , . . . , Emn ),
plus the coherence condition expressed by the full diagram (B.4), which reads
(13.2)
Ek (El1 , . . . , Elk ) o Eh / En (Ej , . . . , Ej )
1 n

En (Em1 , . . . , Emn )(El1 , . . . , Elk ) / En (Em1 (El1 , . . . ), . . . , Emn (. . . , Elk ))
The bottom arrow is the association isomorphism in the butterfly juxtaposition. For n ≤ 4,
with all the sequences emanating from n = 4 as in sect. B, we obtain the “generalized
pentagon” diagram (C.2).
Theorem 12.3 and the fact that pseudo monoids are preserved by pseudo functors
(Definition B.5) guarantee that a weakly ring-like R• gives rise, through the associated
stack construction, to a weakly ring-like stack. More interesting is the converse direction,
namely:
13.5. Proposition. If (R, t, θ) is a weakly categorical ring and R1 → R0 → R a
presentation, then R• : R1 → R0 is a pseudo monoid in MBXMod.
∼
Proof. From Theorem 12.3 we get an equivalence MaR : {R• , R• } → {R, R}.
Here is an explicit construction of a quasi-inverse to MaR . By Theorem 10.1, or
rather the generalization of the proof of Theorem 8.1, the functor u has a quasi-inverse v
computed by the fiber product construction which yields an n-butterfly En = v(t(n)); for
a composition t(n)(t(m1 ), . . . , t(mn )) we obtain a morphism
∼
v(t(n)(t(m1 ), . . . , t(mn ))) −→ En (Em1 , . . . , Emn ).
Thus, for any
θn;m1 ,...,mn : t(m1 + · · · + mn ) → t(n)(t(m1 ), . . . , t(mn ))
in MR, we obtain
∼ ∼
Em1 +···+mn −→ v(t(n)(t(m1 ), . . . , t(mn ))) −→ En (Em1 , . . . , Emn ),
eventually arriving at (13.2).
The notion of morphism for both weakly categorical rings and crossed modules is
straightforward. In each instance, the notion is a special case of that of morphism of
pseudo-monoid examined in sect. B.
13.6. Notation. From now on it will be convenient to refer to the monoidal structure
of R as a group-like object as the “intrinsic” or “internal” one, and denote it by a plus.
Correspondingly, the relative unit object will be denoted by 0R , or simply 0, if no confusion
is bound to arise. (This retroactively justifies the choice of the attribute “biadditive” for
the functor m.) The second monoidal structure (m, µ) will be referred to as the “extrinsic”
or “external” one. The result of the application of m will be denoted by a juxtaposition.
The corresponding unit object, if it exists, will be denoted by IR or simply I.
14. The presentation of a categorical ring

In this section we show that the presentation of a (weak) categorical ring or ring-like
stack is a crossed module with certain additional properties. In particular, its zeroth and
first homotopy sheaves are, respectively, a ring and a bimodule over that ring, as in the
standard case.
Let R be as above. Let A = π0 (R) and M = π1 (R) be the sheaves of connected
components and of automorphisms of the (additive) identity object. Since the definition
of πi applies to the underlying (group-like) stack, by the usual arguments A and M are
also zeroth and first homotopy sheaves of any crossed module R• used to present R. In
fact the homotopy kernel of the projection $ : R → A is identified with the Picard stack
M = Tors(M ) of M -torsors. One has the familiar diagram
M
ı / R1 / 0
∂
q
0 / R0 / A
π

M /R $ / A
with homotopy-exact columns and where the top two levels of each column (the part
within the box above) provide a presentation of the item immediately below. The colored
sequence in it is the standard extension of A by M by way of the crossed module R• . Of
course, the top two rows, considered as a sequence of complexes of length one, is exact
only in the sense that the group-like categories or stacks they determine form a short exact
sequence.
For objects X1 , X2 , . . . , Xn of R define the result of applying the n-ary operation
t(n) simply by X1 X2 · · · Xn . Then, by defining [X1 ][X2 ] · · · [Xn ] := [X1 X2 · · · Xn ], we
immediately get that $ : R → A is a morphism of pseudo monoids, so that A is a
ring—unital if R possesses a unit object IR for the external monoidal structure. (It follows
that M behaves like a bilateral ideal in R, namely the monoidal structure on R, and
more precisely its binary operation, restricts to a pair of actions:
R × M −→ M and M × R −→ M .)
While there is no direct map R0 × R0 → R0 , the exterior monoidal structure m2 =
t(2) : R × R → R is “covered” by the diagram
p
E 
(14.1)
R0 × R0 R0
which is part of the structure of the biextension. Given a pair (x, y) ∈ R0 × R0 , and
the choice of a point in the fiber e ∈ Ex,y , we get a “value” (e) ∈ R0 . The latter is of
course only defined up to shifting e by the action of R1 , namely (e r) = (e) + ∂r, where
r ∈ R1 . The connected component q((e))) ∈ A is well-defined and independent of all
choices. Moreover, by Theorem 8.1, or rather its proof, we have Ex,y = m2 (π(x), π(y)),
and therefore $(Ex,y ) = $(π(x))$(π(y)) = q(x)q(y). We conclude that, with e ∈ Ex,y as
above, q((e)) = q(x)q(y). Put differently, the diagram displayed above, if interpreted as
a “virtual ring structure” for R0 , covers the multiplication map of A. This consideration
extends to any number of variables.
Hence the following standard fact holds:
14.1. Proposition. M is an A-bimodule.
We recall the main idea of the proof, as it will be needed to translate the above fact in
terms of the presentation of R and the multi-extensions associated to its second monoidal
structure.
Proof Proof of 14.1 (Sketch). Recall that the standard way to relate the automor-
phisms of 0R to those of any other object X of R is to use the diagram
idX +α
X + 0R / + 0R
X

X /X
α̃
Left and right multiplications by an object Y (i.e. the additive functors m2 (Y, −) and
m2 (−, Y )) translate the above diagram into the corresponding ones for automorphisms
Y α and αY . If β : Y → Z, so that [Y ] = [Z] ∈ A, then we arrive at the square
XY + 0R
id +αY / XY + 0R
idX β+id0 idX β+id0

XZ + 0R / XZ + 0R
id +αZ
from which we conclude that αY = αZ. The situations for Y α, Zα is analogous.

Combining this with Proposition 13.5 we have:

14.2. Corollary. Every categorical ring R has a presentation
0 /M / R1 ∂ / R0 /A / 0,
where R• : R1 → R0 is a braided crossed module, A = π0 (R) a ring and M = π1 (R) an

A-bimodule.
In fact, something more can be said about the presentation, despite R0 having the
weaker structure given by (14.1). In addition, we will need to describe the A-bimodule
structure of M in terms of (14.1) and the biextension E.
Consider the morphism examined in Remark 8.5: for an arrow x → x0 = x+∂r of [R1 →
R0 ]∼ , with x ∈ R0 and r ∈ R1 , we have the morphism of (R1 , R0 )-torsors ρy : Ex,y → Ex0 ,y ,
e 7→ e ×1 s1 (r, y). If ∂r = 0, so r ∈ π1 (R) = M , then ρy is an automorphism of Ex,y , hence
ρ ∈ M as well (depending on x, y), since in general AutTors(R1 ,R0 ) (E) ' MR0 ×R0 . Similarly,
we construct an automorphism λx : Ex,y → Ex,y corresponding to the automorphism of
y ∈ R0 given by r ∈ R1 .
14.3. Lemma-Definition. The automorphism ρy (resp. λx ) is identified with a section
−ry (resp. −xr) of M such that if y, y 0 ∈ R0 (resp. x, x0 ∈ R0 ) and q(y) = q(y 0 ) ∈ A (resp.
q(x) = q(x0 ) ∈ A), then ry = ry 0 (resp. xr = x0 r).
With the above notations, define ar := xr and rb := ry, where a = q(x), b = q(y) ∈ A.
14.4. Remark. The apparently bizarre choice of inserting an inverse (denoted additively)
is due to the point raised in Remark 2.2.
Proof Proof of the Lemma. The identification between the automorphisms of E and
M is standard: for any section e ∈ Ex,y we have ρy (e) = e rρy (e), with rρy (e) ∈ M . On the
other hand, this must be equal to e ×1 s1 (r, y), the latter coming from the composition
Ex,y −→ Ex,y ∧R1 E0,y −→ Ex,y .
Recall from sect. 4 that, analogously to the standard case of abelian biextensions [Gro72],
for x, y ∈ R0 , the fibers Ex,0 and E0,y are canonically identified with the unit (R1 , R0 )-torsor:
the identification
∼
R1 −→ E0,y , 0R1 7−→ ey ,
sends the unit section of R1 (which we write as 0 according to the current additive
convention) to the central section ey ∈ E0,y . Note that as a result (ey ) = 0R0 . By the
above, we have
(14.2) s1 (r, y) = ey (−ry),
where ry := −rρy (ey ) ∈ R1 , and the juxtaposition stands for the right R1 -action. In fact,
we have ry ∈ M , since  ◦ s1 is trivial (cf. sect. 7).
Now, assume we have y → y 0 = y + ∂u, u ∈ R1 . We have
7 E0,y
ey
R1 u e 7−→ e ×2 s2 (0, u)
ey0 '
E0,y0
Using the identities (7.3) and the restriction condition 1 in Definition 7.1 we calculate:
s1 (r, y) ×2 s2 (0, u) = s1 (r, y) ×2 s2 (∂r, u)

= s1 (r, y) ×2 s1 (r, ∂u)
= s1 (r, y + ∂u)
= s2 (r, y 0 ).
Using (14.2) the above calculation gives
ey (ry) ×2 s2 (0, u) = ey ×2 (ry) s2 (0, u)

= ey ×2 s2 (0, u) (ry)
= ey0 (ry 0 ) .
The fact that the identification is canonical gives ey ×2 s2 (0, u) = ey0 and hence ry = ry 0 .
The situation with ex ∈ Ex,0 , xr, and x0 r is entirely analogous.
14.5. Remark. In the extension of A by M above, R1 and R0 are only groups, in
general, even though the crossed module R• they form comes equipped with a braiding.
In particular, unless certain strong triviality conditions on the biextension E2 = t(2) ∈
Biext(R• , R• ; R• ) hold, R0 is not a ring, in general. This remark applies to stock categorical
rings as well, which therefore admit a presentation of the above type with R• a braided
symmetric crossed module. This is in sharp contrast with the so-called Ann-Categories,
see [Qua03, QHT08, Qua13], whose underlying categories are Picard groupoids. In a
companion paper [Ald15] we prove that Picard stacks with a ring-like structure of this
type admit presentations given by crossed bimodules, namely crossed modules such as in
the presentation above where R0 is a ring and R1 is an R0 -bimodule, with ∂ a bimodule
morphism satisfying an appropriate version of the Pfeiffer identity. The general comparison
between these structures is intricate and it will be addressed elsewhere.
15. Decomposition of categorical rings and the cohomology of rings

Let R, A, and M be as above. In general, a decomposition [Bre94a] of R consists in
the choice of local data (objects, morphisms) subordinate to a hypercovering of A leading
to the calculation of cohomological invariants of R. We use the pseudo-monoidal structure
carried by a presentation of R for this purpose.
We let L•i (A), i = 2, 3 be the sharp truncations of the iterated bar complexes B(A, i),
i = 2, 3 of A as an abelian group. By [ML58, §11] it carries a product structure, defined
explicitly up to cells in B(A, 2). For i = 2, 3 we let Hi3 (A, M ) denote the (hyper)cohomology
groups computed using the multiplicative bar construction over Li• (A). From the previous
reference we have the H33 (A, M ) ' HML3 (A, M ), the third Mac Lane cohomology of A
with values in M . Furthermore, we let H e 23 (A, M ) denote the group of twisted classes
satisfying (15.2) below.
15.1. Theorem. There is a bijective correspondence between equivalence classes of weak
ring-like stacks R, with A = π0 (R) and M = π1 (R), and twisted classes in H e 3 (A, M )
2
defined below (cf. Definition 15.9). In the unital case, i.e. when the external monoidal
structure of R has a unit element, A is likewise unital and the underlying braiding of R
is necessarily symmetric. Hence the weak ring-like structure of R is fully ring-like, and
[R] ∈ H e 23 (A, M ) ' HML3 (A, M ).
The statement follows from Proposition 15.15, Corollary 15.16, and Proposition 15.17
below. The rest of this section is devoted to the details of the proof of the theorem.
15.2. The bar complex. We establish the notation for the iterated bar complex of the
abelian group A. We quote, with minor changes in the notation, from [Dug14, §7]. The
part of B ∞ (A) of interest is the following complex, written homologically:
∂1 ∂2 ∂3
L3• (A) : Z[A] o Z[A2 ] o Z[A3 ] ⊕ Z[A2 ] o Z[A4 ] ⊕ Z[A3 ] ⊕ Z[A3 ] ⊕ Z[A2 ]
We place the complex in homological degrees [0, 3], with generators:
Degree Generators
0 [a]
1 [a |1 b]
2 [a |1 b |1 c] [a |2 b]
3 [a |1 b |1 c |1 d] [a |1 b |2 c] [a |2 b |1 c] [a |3 b]
The differential is given by:
∂1 [a |1 b] = [b] − [a + b] + [a]
∂2 [a |1 b |1 c] = [b |1 c] − [a + b |1 c] + [a |1 b + c] − [a |1 b]
∂2 [a |2 b] = [a |1 b] − [b |1 a]
∂3 [a |1 b |1 c |1 d] = [b |1 c |1 d] − [a + b |1 c |1 d]
+ [a |1 b + c |1 d] − [a |1 b |1 c + d] + [a |1 b |1 c]
∂3 [a |1 b |2 c] = [a |1 b |1 c] − [a |1 c |1 b] + [c |1 a |1 b] − [b |2 c] + [a + b |2 c] − [a |2 c]
∂3 [a |2 b |1 c] = [a |1 b |1 c] − [b |1 a |1 c] + [b |1 c |1 a] + [a |2 c] − [a |2 b + c] + [a |2 b]
∂3 [a |3 b] = [a |2 b] + [b |2 a].
The subcomplex L2• (A) is the one obtained by dropping the component Z[A2 ] of L33 (A),
namely the one generated by symbols [a |3 b]. The degrees are shifted in a way compatible
with the reset needed to form the infinite bar construction, so in effect we have Li• (A) =
B(A, i)[−i], for i = 2, 3, and therefore
H2 (L2• (A)) ' H4 (K(A, 2)), H2 (L3• (A)) ' H5 (K(A, 3)).
In the stable situation, this shift is compatible with the degrees in the Q construction.
15.3. Product structure. Let now A be considered with its ring structure. We
describe the product structure on L2• (A) and L3• (A) (see [ML58, §11]). The non zero
products among the generators, up to products taking values in degrees ≤ 2,11 are the
following:
[a][b] = [ab]
[a][b |1 c] = [ab |1 ac] [a |1 b][c] = [ac |1 bc]
[a][b |1 c |1 d] = [ab |1 ac |1 ad] [a |1 b |1 c][d] = [ad |1 bd |1 cd]
[a][b |2 c] = [ab |2 ac] [a |2 b][c] = [ac |2 bc]
and the most interesting one12 :
[a |1 b][c |1 d] = [ac |1 bc |1 ad + bd] − [ac |1 ad |1 bc + bd] + [ad |1 bc |1 bd] − [bc |1 ad |1 bd] − [bc |2 ad].
With this product L2• (A) and L3• (A) become DGAs. Furthermore, both are equipped with
the augmentation η : Lk• (A) → A given by η([a]) = a, for a ∈ A, and zero in all other
degrees. Thus, they become augmented DGAs to which we can apply the (reduced) bar
construction B̄i,• (A) := B̄N (Li• (A), η) of loc. cit. (more details below). We are interested
in the resulting cohomology groups H 3 (Hom(B̄i,• (A), M )), where M is an A-bimodule as
above. As remarked, for i = 3 we have H 3 (Hom(B̄3,• (A), M )) ' HML3 (A, M ).
15.4. The multiplicative bar construction and third cohomology of rings.
To avoid typographical clutter, the generators of B̄i,• (A) are denoted Ju1 , . . . , un K, where
the uk s are homogeneous elements of Li• (A). The degree is
degJu1 , . . . , un K = n + |u1 | + · · · + |un |,

11
The remaining part of the product structure does not appear in [ML58] and it is unknown, to the
author, at least, whether it is explicitly available.
12
There appears to be a discrepancy in the second slot of the second term of the right hand side,
presumably a misprint in [ML58].
where we let |uk | = deg uk . We quote [ML58, p. 323] the expression for the differential in
the bar complex; it is ∂tot = ∂ 0 + ∂ 00 , where
n
X
0
∂ Ju1 , . . . , un K = − (−1)εi−1 Ju1 , . . . , ∂ui , . . . , un K
i=1
n−1
X
00
∂ Ju1 , . . . , un K = η(u1 )Ju2 , . . . , un K + (−1)εi Ju1 , . . . , ui ui+1 , . . . , un K
i=1
+ (−1)εn Ju1 , . . . , un−1 Kη(un ),
where we have set εi = degJu1 , . . . , ui K. On the right hand side of the expression for ∂ 0 the
inner ∂ui indicates the differential in the complex Li• (A).
The cells of total dimension (=degree) 3 are:
n Generators
3 J[a], [b], [c]K
2 J[a |1 b], [c]K J[a], [b |1 c]K
1 J[a |1 b |1 c]K J[a |2 b]K
Therefore a 3-dimensional cochain over B̄i,• (A) with values in M is a 5-tuple ξ =
(f, α1 , α2 , f+ , g+ ), where
f : A3 −→ M
α1 : A2 × A −→ M, α2 : A × A2 −→ M
f+ : A3 −→ M, g+ : A2 −→ M.
To write the cocycle condition δξ = ξ ◦ (∂ 0 + ∂ 00 ) = 0 in explicit form we need the cells of
dimension four:
n Generators
4 J[a], [b], [c], [d]K
3 J[a |1 b], [c], [d]K J[a], [b |1 c], [d]K J[a], [b], [c |1 d]K
2 J[a |1 b], [c |1 d]K J[a |1 b |1 c], [d]K J[a |2 b], [c]K J[a], [b |1 c |1 d]K J[a], [b |2 c]K
1 J[a |1 b |1 c |1 d]K J[a |1 b |2 c]K J[a |2 b |1 c]K J[a |3 b]K†
The element marked with a (†) would only figure in the bar complex B̄3,• (A). In explicit
form, the cocycle condition on the five components of ξ is quite complex, consisting of
several equations. For ease of reading we arrange them in different groups we have (see
below for specific comments):
(15.1a) a f (b, c, d) − f (ab, c, d) + f (a, bc, d) − f (a, b, cd) + f (a, b, c) d = 0
f (b, c, d) − f (a + b, c, d) + f (a, c, d) = α1 (ac, bc; d) − α1 (a, b; cd) + α1 (a, b; c) d

−f (a, c, d) + f (a, b + c, d) − f (a, b, d) = a α1 (b, c; d) − α1 (ab, ac; d)
(15.1b)
− α2 (a; bd, cd) + α2 (a; b, c) d
f (a, b, d) − f (a, b, c + d) + f (a, b, c) = a α2 (b; c, d) − α2 (ab; c, d) + α2 (a; bc, bd)
(15.1c) f+ (ac, bc, ad + bd) − f+ (ac, ad, bc + bd)

+ f+ (ad, bc, bd) − f+ (bc, ad, bd) − g+ (bc, ad)
= α1 (a, b; d) − α1 (a, b; c + d)
+ α1 (a, b; c) + α2 (b; c, d) − α2 (a + b; c, d) + α2 (a; c, d)
f+ (ad, bd, cd) − f+ (a, b, c) d = − α1 (b, c; d) + α1 (a + b, c; d)

− α1 (a, b + c; d) + α1 (a, b; d)
f+ (ab, ac, ad) − a f+ (b, c, d) = α2 (a; c, d) − α2 (a; b + c, d)
(15.1d) + α2 (a; b, c + d) − α1 (a; b, c)
g+ (ac, bc) − g+ (a, b) c = −α1 (a, b; c) + α1 (b, a; c)
g+ (ab, ac) − a g+ (b, c) = α2 (a; b, c) − α2 (a; c, b)
f+ (b, c, d) − f+ (a + b, c, d) + f+ (a, b + c, d) − f+ (a, b, c + d) + f+ (a, b, c) = 0

f+ (a, b, c) − f+ (a, c, b) + f+ (c, a, b) = g+ (b, c) − g+ (a + b, c) + g+ (a, c)
(15.1e)
f+ (a, b, c) − f+ (b, a, c) + f+ (b, c, a) = −g+ (a, c) + g+ (a, b + c) − g+ (a, b)
g+ (a, b) + g+ (b, a) = 0 .
15.5. Remarks.
1. The last group of equations (15.1e) is closed and it is the condition for the pair
(f+ , g+ ) to be an Eilenberg Mac Lane cocycle. Thus (f+ , g+ ) represents a class in
H 4 (K(A, 2), M ) or H 5 (K(A, 3), M ), depending on whether the last term is included.
2. The block (15.1d) can be re-written as follows. Let λa and ρa denote the left and
right multiplications by a ∈ A. Also, write a f+ and f+ a for the action of a ∈ A on
the values of f+ . Same for g+ . Then we can rewrite the block as
ρ∗d f+ − f+ d = −α1 (−, −; d) ◦ ∂3

λ∗a f+ − a f+ = α2 (a; −, −) ◦ ∂3
ρ∗c g+ − g+ c = −α1 (−, −; c) ◦ ∂2
λ∗a g+ − a g+ = α2 (a; −, −) ◦ ∂2
The block (15.1d) expresses the invariance of the class of (f+ , g+ ) under left and
right multiplication in A.
3. Equation (15.1a) has the familiar form of a Hochschild cocycle. (As it might be
expected, it arises from the associativity constraint on the exterior monoidal structure
of R, as it will be shown below.) The behavior of f with respect to the additivity,
i.e. its failure to be multilinear, is expressed by the equation block (15.1b).
4. As it will be explained below, the meaning of the somewhat more intricate rela-
tion (15.1c), is that it expresses the compatibility (interchange law) between the
partial composition laws of the exterior monoidal structure of R. It relates to the
last of the relations in the Mac Lane’s product structure of sect. 15.3.
It is well known that H5 (K(A, 3)) ' A/2A and that H 5 (K(A, 3), M ) ' Hom(A/2A, M ) '
Hom(A, 2 M ) [EML54, §23] (see also [Dug14, §7.3]). The former isomorphism is given by
a 7→ [a |2 a], the latter sends the class of [(f+ , g+ )] to the (linear) map a 7→ g+ (a, a). In
fact, dropping the last of equations (15.1e), the same assignment gives a quadratic map
from A to M . (Recall that q : A → M is quadratic if q(na) = n2 q(a) for all n ∈ Z, and if
the associated symmetric function ∆q(a, b) = q(a + b) − q(a) − q(b), a, b ∈ A, is bilinear.)
Thus H 5 (K(A, 3), M ) ' Hom(Γ2 (A), M ), where Γ2 (A) is the Whitehead functor, i.e. the
degree four component of the divided power algebra Γ• (A) over A.
Recall that m ∈ M is central if am = ma for all a ∈ A. Let M A be the A-module of
central elements of M . Assume that ξ represents a class in H 3 (Hom(B̄2,• (A), M )), that is,
the pair (f+ , g+ ) satisfies, as part of the full set ξ = (f, α1 , α2 , f+ , g+ ), equations (15.1d)
and (15.1e), except the last one. We have the following observation.
15.6. Proposition. Let A be unital. We have an isomorphism
H 3 (Hom(B̄2,• (A), M )) ' H 3 (Hom(B̄3,• (A), M )) ' HML3 (A, M ).
Moreover, the map HML3 (A, M ) → Hom(A, 2 M ) [ML58, §11] factors through 2 M A '
HomA (A, 2 M A ).
Proof. Let q be the map a 7→ g+ (a, a). The last two equations of (15.1d) imply that q
is an A-bimodule homomorphism. As such, it is determined by a central element in M .
This proves the first isomorphism. The second follows from [ML58].
15.7. Remark. Because the complexes B̄2,• (A) and B̄3,• (A) are equal below degree 3,
their cohomologies coincide for n ≤ 2. Combined with [Bre99, §6], we have that the
statement of Proposition 15.6 holds in general when H • is interpreted as hypercohomology.
The following variant of the previous constructions will be useful below. Given a
5-tuple ξ as above, consider the map βξ : B2,4 (A)2 → M given by (the subscript 2 refers to
the elements in B2,4 (A) with n = 2):

g+ (ab, ac) + g+ (ac, ab) Ju1 , u2 K = J[a], [b |2 c]K,

βξ (Ju1 , u2 K) = g+ (ac, bc) + g+ (bc, ac) Ju1 , u2 K = J[a |2 b], [c]K,

0 all other cases.

Further, we define βξ to be zero on the other components of B2,4 (A). Consider the twisted
cocycle equation condition:
(15.2) Dξ := δξ + βξ = 0.
This replaces the last two equations in the block (15.1d) with
−g+ (bc, ac) − g+ (a, b) c = −α1 (a, b; c) + α1 (b, a; c)

(15.3)
−g+ (ac, ab) − a g+ (b, c) = α2 (a; b, c) − α2 (a; c, b)
and it still drops the last one from (15.1e). Observe that for i = 3 equation (15.2) is
vacuously equal to the set (15.1).
15.8. Remark. Note that replacing ξ with ξ 0 = ξ + δν has the effect of adding to g+ the
alternation of a map h+ : A × A → M , so that βξ does not depend on the particular choice
of g+ within its class modulo coboundaries. Hence we get a well defined class of solutions
of equation (15.2) modulo adding coboundaries.
15.9. Definition. Let us denote by H e 3 (A, M ) the hypercohomology group of classes
2
satisfying (15.2) in degree three, relative to A (cf. Remark 15.7).
There is an evident map HML3 (A, M ) → H e 3 (A, M ), which, at least in the cases we
2
like to consider, is an isomorphism. Indeed we have
15.10. Lemma. Let A be unital. Then (15.2) and (15.1) are equivalent. Therefore
e 3 (A, M ) ' HML3 (A, M ).
H2
Proof. Choosing a or c = 1 in (15.3) shows that the last equation (15.1e) holds. (Again,
use Remark 15.7 for the general situation.)
15.11. Decomposition of R. For convenience of notation, let us write [∂ : R1 → R0 ] ≡
[∂ : R → Λ]. Recall that since we denote the interior monoidal structure of R additively,
we do the same for R and Λ in the presentation.
15.12. Remark. [Warning] In the following we use a set theoretic-type notation. How-
ever, in this sheaf theoretic context, the notation a ∈ A, or a1 , . . . , an ∈ A means these
are generalized points of A of the form, say, a : U → A, etc. where U is an object of the
topos. In fact, in order to properly carry out the hypercohomology calculations we will
have to choose hypercovers of A and of various simplicial objects associated to it, such as
the various Eilenberg-Mac Lane objects K(A, 2) and K(A, 3). We will proceed formally as
in the pointwise case, and systematically appeal to the hypercohomology spectral sequence
as in Remark 15.7. The actual hypercohomology arguments based are made precise in
Appendix E.
For a point a ∈ A, we let Λa = a∗ Λ be the fiber. For any multi-extension En in the
weak ring-like structure of R → Λ denote by Ea1 ,...,an the pullback of E to Λa1 × · · · × Λan ,
where a1 , . . . , an ∈ A.
There is an obvious morphism Λa1 × · · · × Λan → Λa1 +···+an covering the n-fold iteration
of + : A × A → A. If we assume a choice for a point xa ∈ Λa has been made for all points
a ∈ A, i.e. we have a section x of q : Λ → A, then xa + xb (the image of (xa , xb ) under
Λa × Λb → Λa+b ) and xa+b will in general be different. This gives rise to σ : A × A → R by

way of
xa+b = xa + xb + ∂σa,b .
In fact the objects Xa := π(xa ) of R provide a decomposition of the sequence M → R → A,
where R is considered as a gerbe over A [Bre94a, §7]. Note that the above corresponds to
the morphism
σa,b : Xa+b −→ Xa + Xb
of (R, Λ)-torsors. Therefore we have the classical fact that the difference between
the two possible comparisons between Xa + Xb + Xc and Xa+b+c , and the application
of equations (2.2) to xa , xb , and xc determines a pair (f+ , g+ ) satisfying the cocycle
equations (15.1e) above, hence a class in H 4 (K(A, 2), M ) or H 5 (K(A, 3), M ) (see, e.g.
[JS93, Bre99]). For future reference, the relevant relations are (cf. [Bre94a]):
σb,c + σa,b+c − f+ (a, b, c) = σa,b xc + σa+b,c ,

(15.4)
−g+ (a, b) + σa,b = {xa , xb } + σb,a .
(There is an obvious generalization to n variables, but the classical situation is sufficient

to describe the decomposition with respect to the “+” operation of R.)
For the multiplication, i.e. the exterior monoidal structure, the reasoning at the end of
sect. 13 implies we have instead
Ea1 ,...,an
p 
(15.5)
Λa1 × · · · × Λan Λa1 ...an
covering the multiplication in A. Using the choice of a section of the fibers Λa as above,
we have the isomorphism of (R, Λ)-torsors
∼
(15.6) ea1 ,...,an : Ea1 ,...,an −→ Xa1 ···an = (R, xa1 ···an )
which follows from (15.5) and again the end of sect. 13. We can assume this isomorphism
is realized by the choice of a point ea1 ,...,an ∈ Ea1 ,...,an such that (ea1 ,...,an ) = xa1 ···an .
We can write the morphism (13.1) with respect to this choice of local data. According
to the proof of Theorem 11.1, let a1,1 , . . . , a1,m1 , . . . , an,1 , . . . , an,mn ∈ A. For i = 1, . . . , n
define bi = ai,1 · · · ai,mi . Consider the points ei = eai,1 ,...,ai,mi ∈ Eai,1 ,...,ai,mi and eb1 ,...,bn ∈
Eb1 ,...,bn . Then [e1 , . . . , en , eb1 ,...,bn ] is a point of the composition En (Em1 , . . . , Emn ) over
Λa1,1 × · · · × Λan,an , and ([e1 , . . . , en , eb1 ...,bn ]) = (eb1 ,...,bn ) = b1 · · · bn . In other words, we
have an isomorphism of (R, Λ)-torsors
∼
En (Em1 , . . . , Emn )a1,1 ,...,an,mn −→ (R, xb1 ···bn ).
On the other hand, since Em1 +···+mn covers the multiplication
(a1,1 , . . . , an,mn ) → a1,1 · · · an,mn = b1 · · · bn ,

we have another isomorphism
∼
Em1 +···+mn a1,1 ,...,an,mn
−→ (R, xb1 ···bn )
determined by a chosen section of Em1 +···+mn . Thus, the morphism (13.1) amounts to an
automorphism
∼
(15.7) fn;m1 ,...,mn (a1,1 , . . . , an,mn ) : (R, xb1 ···bn ) −→ (R, xb1 ···bn ),
which, using standard facts about (R, Λ)-torsors, we identify with an element of M .
15.13. Cocycle computations. We compute the full class determined by R from the
multi-extension structure of the presentation R → Λ → R and prove that it satisfies the
full set of equations (15.1).
15.13.1. Setup. First, consider the following (not necessarily commutative) diagrams of
(R, Λ)-torsor morphisms
Ea,c ∧ Eb,c / Exa +xb ,xc / Ea+b,c

(15.8a)
−1
σac,bc
(R, xac ) ∧ (R, xbc ) (R, xac + xbc ) / (R, x(a+b)c )
and
Ea,b ∧ Ea,c / Exa ,xb +xc / Ea,b+c

(15.8b)
−1
σab,ac
(R, xab ) ∧ (R, xac ) (R, xab + xac ) / (R, xa(b+c) )
arising from the two partial product laws of E = E2 . The vertical arrows are the
identification (15.6). Both diagrams in (15.8) can be construed as defining an automorphism
of their respective lower right corners, which can be identified with an element of M : let
them be −α1 (a, b; c) and α2 (a; b, c) respectively.
On the one hand the top rows can be written as
(15.9) ea,c ∧ eb,c 7→ ea+b,c g1 (a, b; c), ea,b ∧ ea,c 7→ ea,b+c g2 (a; b; c),
by invoking (4.5). On the other hand, following the bottom part, we have
−1 −1
ea,c ∧ eb,c 7→ σac,bc , ea,b ∧ ea,c 7→ σab,ac .
Comparing the two we get the relations
(15.10) −α1 (a, b; c) = g1 (a, b; c) + σac,bc , α2 (a; b, c) = g2 (a; b, c) + σab,ac .

15.13.2. The relations (15.1d). Consider the diagrams:

σa+b,c σb+c,d
Exa+b +xc ,xd / Ea+b+c,d Exa ,xb+c +xd / Ea,b+c+d
O O
xc x d
σa,b σb,c
(15.11a) Exa +xb +xc ,xd Exa ,xb +xc +xd

σb,c σc,d

Exa +xb+c ,xd / Ea+b+c,d Exa ,xb +xc+d / Ea,b+c+d
σa,b+c σb,c+d
and
σa,b σb,c
Exa +xb ,xc / Ea+b,c Exa ,xb +xc / Ea,b+c
O O O O
(15.11b) η1 η2
Exb +xa ,xc / Ea+b,c Exa ,xc +xb / Ea,b+c

σb,a σc,b
Both diagrams (15.11) are parts of more extended ones, giving rise to relations linking f+
and g+ to the other quantities comprising a 5-tuple satisfying relations (15.1) as follows.
First observe that by applying Lemma 14.3, the right vertical give the automorphisms
corresponding to f+ (a, b, c) d (resp. a f+ (b, c, d)) for (15.11a), and g+ (a, b) c (resp. a g+ (b, c))
for (15.11b).
Then from (15.8) and (15.11a) form the obvious associativity diagrams for the mor-
phisms Ea,d ∧ Eb,d ∧ Ec,d → Ea+b+c,d and Ea,b ∧ Ea,c ∧ Ea,d → Ea,b+c+d . Using the cocycle
decomposition (4.6) (and Lemma 14.3 for the right vertical arrows of (15.11)) we arrive
at:
g1 (a + b, c; d) + g1 (a, b; d)xcd = g1 (a, b + c; d) + g1 (b, c; d) + f+ (a, b, c) d,
(15.12)
g2 (a; b + c, d) + g2 (a; b, c)xad = g2 (a; b, c + d) + g2 (a; c, d) + a f+ (b, c, d),
(The difference with equations (4.6a) and (4.6c) arises because the top rows of (15.8),
contrary to the actual partial multiplication morphisms, lack associativity.) Using (15.10),
(15.12), and the first of (15.4), we obtain the first two of the cocycle relations (15.1d).
The commutativity diagrams obtained from (15.11b) and (15.8) can be analyzed
in an analogous manner, utilizing the second of (15.4). However, we do not directly
obtain the other two equations in the block (15.1d); instead, we arrive at their “flipped”
counterpart (15.3).
15.13.3. The interchange relation (15.1c). We check the compatibility law (2.5)
after pulling back to Λa × Λb × Λc × Λd . Use the diagram
(15.13)
(Ea,c ∧ Eb,c ) ∧ (Ea,d ∧ Eb,d ) / (Ea,c ∧ Ea,d ) ∧ (Eb,c ∧ Eb,d )
+ s
Exa +xb ,xc ∧ Exa +xb ,xd I Exa ,xc +xd ∧ Exb ,xc +xd
s ( v +
Ea+b,c ∧ Ea+b,d II Exa +xb ,xc +xd II Ea,c+d ∧ Eb,c+d
+ v ( s
Exa+b ,xc +xd III Exa +xb ,xc+d
$ z
*
Ea+b,c+d t
where the triangles commute by definition of the morphisms determined by the top rows
of (15.8); the pentagon (marked I) commutes by the compatibility law; the squares II and
the square III are obviously commutative by functoriality (an explicit calculation uses
Proposition 4.2 and the equations (7.3)). Thus the diagram commutes, and we can use
the cocycle equation (4.7) directly, written additively, which gives:
g2 (a + b; c, d) + g1 (a, b; c)x(a+b)d + g1 (a, b; d)

= g1 (a, b; c + d) + g2 (a; c, d)xb(c+d) + g2 (b; c, d) − {xbc , xad }xbd .
Inserting equations (15.10) and using (15.4) finally gives the third block (15.1c) of the
cocycle relation (15.1).
15.13.4. The relations (15.1b) and (15.1a). We specialize the expression (15.7) for
the morphism (13.1) to the tuples (2; 2, 1) and (2; 1, 2). The morphism µ given by (C.1)
corresponds to the element
f (a, b, c) = −f(2;1,2) (a, b, c) + f(2;2,1) (a, b, c) ∈ M.
More precisely, using the isomorphism with (R, xabc ) as a reference trivialization, we can
identify f (a, b, c) with an automorphism of E3 pulled back to Λa × Λb × Λc , hence with
a section of M over it. Explicitly, locally on Λa × Λb × Λc , we have the morphism of
(R, Λ)-torsors
(15.14) µa,b,c : E2 (E2 , I)a,b,c −→ E2 (I, E2 )a,b,c ,
which, using the composition of multi-extensions given in section 11, we can write as
(15.15) µ([ea,b , 0c , eab,c ]) = [0a , eb,c , ea,bc ] − f (a, b, c),
where [ea,b , eab,c ] ∈ E2 (E2 , I)a,b,c and [eb,c , ea,bc ] ∈ E2 (I, E2 )a,b,c . (Recall the additivity in
the notation; we extend it to the action of R on torsors. The brackets denote the class
under the action of R. The identity map is represented by the trivial butterfly, and here
0a represents the unit section of the underlying trivial torsor Λ × R at the point xa , say.)
We have ([ea,b , eab,c ]) = ([eb,c , ea,bc ]) = xabc , so we can see directly that f (a, b, c) ∈ M .
As an isomorphism of tri-extensions, µ is a homomorphism for each of the three partial
laws. Writing these conditions for (15.14), we must compute the maps along the following
diagram
×1 σa,b
E2 (E2 , I)a,c,d ∧R E2 (E2 , I)b,c,d / E2 (E2 , I)x +x ,x ,x / E2 (E2 , I)a+b,c,d
a b c d
µ∧µ µ µ

E2 (I, E2 )a,c,d ∧R E2 (E2 , I)b,c,d / E2 (I, E2 )x +x ,x ,x / E2 (I, E2 )a+b,c,d
×1 a b c d σa,b
and the other two expressing the compatibility (or lack thereof) with the second and third
partial laws. Because there are some new elements compared to the calculations which
have appeared thus far, we sketch some of the details. In particular, to compute the two
horizontal maps in the second square above we need the explicit form of the trivializations
s1 for both tri-extensions. Similarly for the other two diagrams. According to the last part
of section 11, the form of s2,1
1 for E2 (E2 , I) is:
s2,1
1 (r, xc , xd ) = [s1 (r, xc ), 0xd , 1xd ], r = σa,b ,
where 0xd denotes the unit section of Λ × R computed at xd ∈ Λ, and 1xd denotes the unit
section of E2 |{0}×Λ ' Λ × R similarly computed at xd . Thus, the map denoted by σa,b in
the top row is given by
e 7−→ e ×1 s1 (σa,b , xc , xd ).
Similarly, for the analogous map in the bottom row we must use the expression
s11,2 (r, xc , xd ) = [0∂r , ec,d , s1 (r, xc )], r = σa,b ,
with a similar interpretation of the notation.

Thus, the upper path to the lower right corner gives

µ [ea,c , eac,d ] ×1 [eb,c , ebc,d ] ×1 [s1 (σa,b , xc ), 0d , 1xd ]
= µ[ea,c ×1 eb,c ×1 s1 (σa,b , xc ), 0d , eac,d ×1 ebc,d ]
= µ[ea+b,c + g1 (a, b; c), 0d , eac,d ×1 ebc,d ]
= µ[ea+b,c , 0d , eac,d ×1 ebc,d ×1 s1 (−g1 (a, b; c), xd )]
= µ[ea+b,c , 0d , eac,d ×1 ebc,d ×1 s1 (σac,bc + α1 (a, b; c), xd )]
= µ[ea+b,c , 0d , eac+bc,d ] + g1 (ac, bc, d) − α1 (a, b; c)d
= [0a+b , ec,d , ea+b,cd ] + g1 (ac, bc; d) − f (a + b, c, d) − α1 (a, b; c)d
where in the next to last we have used (7.3), the first of (15.9), and Lemma 14.3; to obtain
the last we have used (15.15).
On the other hand, the lower path to the lower right corner gives, with similar
calculations
µ[ea,c , eac,d ] ×1 µ[eb,c , ebc,d ] ×1 [0∂σa,b , ec,d , s1 (σa,b , xcd )]
= [0a+b , ec,d , ea+b,cd ] + g1 (a, b; cd) − f (a, c, d) − f (b, c, d).
Comparing the two expressions and using (15.10) yield the first equation of block (15.1b).
The others are obtained via identical means.
The last equation (15.1a) becomes now the easiest to obtain, as a condition satisfied by
the automorphism µ upon considering the five possible pullbacks to the quadri-extension
E4 , as per the pentagon diagram (C.2). We leave the details to the reader.
15.14. The class of a ring-like stack. Assembling the steps in sect. 15.13, we have
the following
15.15. Proposition. Let R be a (weakly) ring-like stack with π0 (R) = A and π1 (R) = M .
A decomposition of R determines a twisted cocycle ξ = ξR satisfying (15.2) with the same
A-bimodule structure. An equivalence R → R 0 gives rise to two twisted cocycles ξR and
ξR 0 differing by a coboundary, hence equivalence classes are in one-to-one correspondence
with elements of H e 3 (A, M ).
2
Proof. The first statement is a consequence of the preceding calculations. The statement
about the equivalence follows from the definition of morphism of pseudo-monoid and the
fact that the structure is preserved across pseudo-functors between multi-bicategories
(cf. B.4 and B.5), in particular between 13.5
Since the only difference between all these complexes occurs in degrees 3 and 4, the
statements carry over to the hypercohomology situation.
Let us use the notation [R] for the class determined by R. Thus of [R] = [ξR ]. As we
have seen, a consequence of Proposition 15.6 and Lemma 15.10 is that in the unital case
e 23 (A, M ) lift to HML3 (A, M ).
the classes in H
15.16. Corollary. Let R be as above, with in addition a unit object for the exterior
∼
monoidal structure. Then A is unital and there exists an equivalence R → R 0 where the
underlying categorical group structure of R 0 is braided symmetric. Hence [R] = [R 0 ] ∈
HML3 (A, M ).
We briefly address the question of recovering R (up to equivalence) from [R]. Consider
a class [ξ] ∈ He 3 (A, M ). A portion of ξ will represent a class in H 4 (K(A, 2), M ), possibly
2
lifting into the stable range. Let ξ+ denote this projection. Standard techniques [Bre94a,
§7.6–7] allow to reconstruct a braided (possibly symmetric) stack R = Rξ from ξ+ ,
equipped with $ : R → A fitting into the standard short exact sequence M → R → A,
with M = Tors(M ). Briefly, ξ+ determines a 2-gerbe over a simplicial model of K(A, 2)
or K(A, 3), suitably re-scaled so that the relevant class appears in degree three. R is
obtained by gluing trivial gerbes with band M → 0 over A along ξ+ . (We must supplement
the cocycles in loc. cit. with those parts pertaining to the braiding structure.) The class
of R is that of ξ+ , and therefore it is equipped with a decomposition (15.4).
15.17. Proposition. Let R, ξ, and ξ+ be as above. Let R → Λ → R be a presentation

by a braided crossed module. Then R → Λ carries a bi-extension whose class is ξ.
Proof Proof (Sketch). The main point is to reverse the computation of the cohomology
class, using the equations for the cocycle ξ (either (15.1) or their twisted form (15.2))
to obtain a well-defined biextension E → Λ × Λ providing (R, Λ) with a pseudo-monoid
structure.
Starting with the implementation of (15.5), for all a, b ∈ A define Ea,b → Λa × Λb as
Ea,b = Λa × Λb × R equipped with  : Ea,b → Λ given by 0R 7→ xab . From (15.10), given ξ
and the decomposition (15.4), we compute the nonabelian cocycles g1 and g2 we can use
to define partial laws
×1 : Ea,c ∧R Eb,c −→ Ea+b,c , ×2 : Ea,b ∧R Ea,c −→ Ea,b+c ,
which are well defined by equations (15.12) and (15.13).

Note that, unless a = 0, the pullback of ∂ : R → Λ along Λa → Λ is trivial, whereas it
is isomorphic to Im ∂ for a = 0. This ensures (again from equation (15.5) and choosing
a normalized section x) the triviality conditions required by Definition 7.1 to have a
full-fledged butterfly are satisfied.
Finally, we use (15.14) and (15.15) to define µ : E2 (E2 , I) → E2 (I, E2 ); the identity
g1 (ac, bc; d) − f (a + b, c, d) − α1 (a, b; c)d = g1 (a, b; cd) − f (a, c, d) − f (b, c, d),
and its companions found above, together with (15.1a) and (15.1b), ensure E satisfies the
required pentagonal structure.
Part IV
A. Multi-variable compositions
In this technical addendum we give a brief treatment of multivariable functor calculus
in a bicategory. Our approach is descriptive and explicit, and it is aimed at a definition
of pseudo-monoid suitable for the applications in the text to multi-additive functors and
multi-linear butterflies (cf. sects. 10 and following).
We resort to multi categorical-based ideas, in fact we borrow some of Kelly’s clubs
formalism (see [Kel72, Kel74]), which is convenient in the present context.13 We include
permutations, even though this is slightly more general than needed in the main part of the
text. Permutations can be included at nearly no additional cost, covering the symmetric
monoidal case, which is what the formalism was originally designed to do. As a result, the
formalism can still be used to symmetrize the (external) monoidal structures described
in the text. Unlike [Kel72], we need our objects to inhabit a bicategory, as opposed to a
13
See also [Tho95] for further applications to (symmetric) monoidal categories.
2-category, due to the fact that crossed modules equipped with butterflies as morphisms
form a genuine bicategory equivalent to the 2-category of group-like stacks.
We define a bi-multicategory C the structure defined by the following data:14
• A class of objects x, y, . . .
• For each tuple (y1 , . . . , yn ; x) of objects, a groupoid of arrows HomC (y1 , . . . , yn ; x).
The cells are denoted
f
'7
(y1 , . . . , yn ) α x.
g
• For each object x, a functor ıx : 1 → HomC (x; x), where 1 is the singleton category.
The resulting distinguished object is the identity arrow idx : (x) → x.
• Compositions functors
HomC (y1 , . . . , yn ; x) × HomC (z1,1 , . . . , z1,m1 ; y1 ) × · · · × HomC (zn,1 . . . , zn,mn ; yn )
−→ HomC (z1,1 , . . . , zn,mn ; x)
(f ; g1 , . . . , gn ) −→ f (g1 , . . . , gn ).
• Associativity data for the composition and the identities as in a bicategory.

• For each n and for each tuple (y1 , . . . , yn ; x), an action by Σn , that is, a functor
ξ ∗ : HomC (yξ(1) , . . . , yξ(n) ; x) −→ HomC (y1 , . . . , yn ; x)
such that (ξη)∗ ' η ∗ ξ ∗ and the composition functors are equivariant for this action.
Moreover, these isomorphisms are subject to appropriate coherence conditions.
The definition of pseudo- (or lax-)functor F : C → B between multi-bicategories is mutatis
mutandis the same as for bicategories. For the symmetric structure, we add the condition
that the functors
Fy1 ,...,yn ;x : HomC (y1 , . . . , yn ; x) → HomB (F (y1 ), . . . , F (yn ); F (x))
preserve the Σn -action, for all n.

The last item in the data list intuitively affords for cells of the form
(y1 , . . . , yn )
%
(A.1)
9x

(yξ(1) , . . . yξ(n) )
14
Without the extra structure given by permutations, the name “bi-multicategory” appears in [Pis14],
where it denotes the bicategory-analog of a multicategory: a Cat-enriched multicategory with weakly
associative composition. For enrichment over simplicial sets, see also [Rob11], which also incorporates
permutations.
Dropping all HomC -groupoids except those of the form HomC (y; x), i.e. those of valence
(or arity) one, we obtain an ordinary bicategory, the underlying bicategory of C. Also, a
2-multicategory is a bi-multicategory in which all the associativity and identity data are
strict. If all the groupoids HomC (−; −) are discrete we obtain an ordinary multicategory.
The following is the analog of the “generalized functor category” in Kelly [Kel72, §2].
A.1. Definition. Let x, y be two objects of C. Then {y, x} denotes the category whose
objects are pairs (n, f ), where n is a natural number and f is an object of HomC (y, . . . , y; x)
(y is repeated n times); morphisms (n, f ) → (m, g) only exist if n = m and consist of a
permutation ξ ∈ Σn and a morphism α : f → ξ ∗ (g). The composition of morphisms is
dictated by the composition in C. Explicitly, if (η, β) : (n, g) → (n, h), then (η, β) ◦ (ξ, α) =
(ηξ, ξ ∗ (β) ◦ α) modulo the (unnamed) coherence morphism ξ ∗ η ∗ ' (ηξ)∗ .
A.2. Remark. Let S be the skeletal category of finite sets with permutations as morphisms.
(S can be obtained as the core, i.e. the largest subgroupoid, of the Segal category Γ of
finite sets.) Thus an object of S can be identified with a natural number. For any two
objects x, y of C there results an obvious projection functor p : {y, x} → S. In fact, {y, x}
can be obtained as the Grothendieck construction applied to the functor F : S → Cat
defined by F (n) = HomC (y, . . . , y; x) and F (ξ) = ξ ∗ .
For the following definition is the analog of the operations defined in [Kel72, §2.1, §2.2].
A.3. Definition. For all objects x, y, z of C:
1. {y, x} o {z, y} is the category whose objects are lists (f ; g1 , . . . , gn ), where n = p(f );
morphisms (f ; g1 , . . . , gn ) → (f 0 ; g10 , . . . , gn0 ) (there are no morphisms if p(f ) 6= p(f 0 ))
consist of a morphism α : f → ξ ∗ (f 0 ) and for 1 ≤ i ≤ n, morphisms βi : gξ(i) →
∗
ηξ(i) (gi0 ), where ξ = p(f ) ∈ Σn , p(gi ) ηi = p(gi ) ∈ Σni . Such a morphism is denoted
(α; β1 , . . . , βn ).
2. There is functor, called the assembly map
◦ : {y, x} o {z, y} −→ {z, x},

sending the object (f ; g1 , . . . , gn ) to the composition f (g1 , . . . , gn ) and the morphism
(α; β1 , . . . , βn ) to the pasting of α and the n-tuple (ηξ(1) , . . . , ηξ(n) ), which is denoted
α(β1 , . . . , βn ) for simplicity, following Kelly, again.
Note that the details of the constructions stated in a direct fashion in Definitions A.1
and A.3 can be deduced from the combinatorics of the composition in the bi-multicategory
C. Similarly, the associativity for the composition given by the assembly map is also
inherited from the ambient bi-multicategory, hence, for objects x, y, z, w of C we have:
ass. / {y, x} o {z, y} o {w, z}

{y, x} o {z, y} o {w, z}
EM
(A.2) ◦oid id o ◦

{z, x} o {w, z} / {w, x} o {y, x} o {w, y}
A.4. Definition. [Kelly [Kel72, Kel74]] A Club is a bi-multicategory with one object.
Given a club with unique object ∗, set k = {∗, ∗}. By abuse of notation we refer to k
itself as a club. Then the assembly map and (A.2) give a monoidal structure ◦ : k o k → k.
Examples of clubs are:
• S itself, with projection id : S → S.
• The natural numbers identified with a discrete subcategory N of S, with projection

ι : N ,→ S.
In both cases k = S and k = N, the composition morphism k o k → k sends the object
(n; m1 , . . . , mn ) to m1 +· · ·+mn . In the case of N, however, we dispense with the symmetric
group action.
A.5. Remark. [Notation] Kelly [Kel72] uses the notations ◦ (resp. µ) where we use o
(resp. ◦). The symbol o usually denotes the so-called wreath product.
The wreath product F oD (or simply CoD) of two categories C and D, the first equipped
with a functor F : C → Γ,15 is defined in a way somewhat similar to Definition A.3 (see
e.g. [Ber07]). In particular, objects are tuples of the form (c; d1 , . . . , dn ), where F (c) has
cardinality n and d1 , . . . , dn are objects of D. Morphisms, however are slightly different.
B. (Unbiased) Monoids and monoidal structures

We keep the notations and environment of section A. C is a bicategory identified with the
underlying bicategory of a bi-multicategory MC. We add another piece of notation: for an
object x, let Mx = {x, x}. Having observed that this is a monoid for the composition, a
monoid object in C is defined in the expected way. Recall that N is a club, as discussed in
the previous section. Recall that a lax monoidal functor is one that respects the monoidal
structures up to coherent natural transformation.16
B.1. Definition. A (pseudo-)monoid in C is an object x of MC equipped with a lax
monoidal functor (t, θ) : N → Mx. A symmetric (pseudo-)monoid would is the same thing
with the club S, instead.
In the next diagrams we spell out the conditions for (t, θ) to be a lax monoidal functor.
Explicitly, the lax monoidal functor is given by a diagram
NoN ◦ / N
(B.1) tot θ t

Mx o Mx / Mx
◦
15
Γ is the Segal category of finite sets.
16
The distinction between lax and op-lax, i.e. the direction of the 2-arrows in the natural transformations
is immaterial as we work with bicategories whose 2-arrows are isomorphisms.
From it we have the coordinate expression for the natural transformation θ, which assigns
to the object (n; m1 , . . . , mn ) ∈ N o N the morphism
(B.2) θn;m1 ,...,mn : t(m1 + · · · + mk ) −→ t(n)(t(m1 ), . . . , t(mn ))
in Mx. These data are subject to be compatible with the associativity conditions for the
assembly maps of both N and Mx, namely they must satisfy the following commutative
diagram of natural transformations:
(B.3)
(Mx o Mx) o Mx ass / Mx o (Mx o Mx)
h 6
(tot)ot to(tot)
(N o N) o N ass / N o (N o N)
◦oId ◦oId Id o ◦ Id o ◦
X` >F
NoN / No NoN toθ
θot
◦ ◦

θ t θ
v tot tot (
Mx o Mx / Mx o Mx o Mx
◦ ◦
In (B.3) the top quadrangle and the small pentagon are strictly commutative. For the
quadrangle, it follows from the functoriality of the associator, whereas for the small
pentagon the associativity morphism reads
((n; m1 , . . . , mn ); l1 , . . . , lk ) 7−→ (n; (m1 ; l1 , . . . ), . . . , (mn ; . . . , lk )),
where k = m1 + · · · + mn ; both paths evaluate to h = l1 + · · · + lk = j1 + · · · + jn , where for

i = 1, . . . , n we have ji = lmi−1 +1 + · · · + lmi−1 +mi . Lastly, the back face of the diagram, that
is the large pentagon, is just (A.2) specialized to the same object. Thus, the commutativity
in diagram (B.3), when expressed in coordinates, reads as follows:
(B.4)
t(k)(t(l1 ), . . . , t(lk )) o t(h) / t(n)(t(j1 ), . . . , t(jn ))

t(n)(t(m1 ), . . . , t(mn ))(t(l1 ), . . . , t(lk )) / t(n)(t(m1 )(t(l1 ), . . . ), . . . , t(mn )(. . . , t(lk ))
B.2. Remark. By construction, the pseudo-monoid structure on x given by the club

morphism t : N → Mx possesses n-ary operations t(n) for arbitrary values of n. If x has an
internal structure (namely objects, as in the main text), then this implements a monoidal
structure in “unbiased” form (see e.g. [Lei04]), with the n-ary unprivileged operations
of all degrees. In this case diagram (B.3) expresses the coherence condition in unbiased
form, namely, that the two possible ways to remove parentheses from an expression must
coincide as it is apparent from the coordinate version (B.4).
B.3. Remark. Consider the singletons {x} as a copy of the singleton category 1. Any
category with products gives rise to a multicategory in a standard way [Lei04, Example
2.16]. Applied to {x}, there is singleton worth of morphisms from x to x of arbitrary
arity. The definition of the o-operation introduced in Appendix A can be copied in order
to define a category N o x (dropping the braces from the notation for convenience). One
could show (for example, by formally checking the adjunction properties of o and {−, −}
as in[Kel72]) that (t, θ) in Definition B.1 correspond to a functor
t̃ : N o x −→ x
and a natural transformation
(N o N) o x / N o (N o x)
◦ox θ̃ Not̃

Nox / xo Nox
t̃ t̃
By definition, (t̃, θ̃) equip x with a structure of N-(pseudo-)algebra. We will not work
with this variant.
One can consider morphisms of monoids as follows. Denote N o x (cf. Remark B.3) by
Tx. Let f : y → x be a (unary) arrow. Then f determines a functor Tf : Ty → Tx and
two functors
f∗ {y, y} −→ {y, x} Tf ∗ {x, x} −→ {y, x}
by post-composing with f or pre-composing with Tf , respectively.
B.4. Definition. Let x, y be (pseudo-)monoids in C. A morphism of monoids is a
pair (f : y → x, λ) fitting the diagram
ty
N / {y, y}
tx λ f∗

{x, x} / {y, x}
Tf ∗
and compatible with the pseudo-algebra condition (B.1).

The compatibility between the above diagram and (B.1) means that in the coordinates
of (B.2) we have
f θy 1
f ◦ ty (m1 + · · · + mk ) /f ◦ (ty (n)(ty (m1 ), . . . , ty (mn ))) / (f ◦ ty (n))(ty (m1 ), . . . , ty (mn ))
λm1 +···+mn λn ◦Id

tx (m1 + · · · + mk ) ◦ Tf (tx (n) ◦ Tf)(ty (m1 ), . . . , ty (mn ))
θx Tf 2

(tx (n)(tx (m1 ), . . . , tx (mn ))) ◦ Tf / tx (n)((tx (m1 ), . . . , tx (mn )) ◦ Tf) o tx (n)(f ◦ ty (m1 ), . . . , f ◦ ty (mn ))
3 Id ◦λm1 ,...,mn
where the numbered arrows result from associativity isomorphism for composition.
Pseudo-monoids behave in the expected manner with respect to pseudo-functors.
Specifically, let F : MC → MB be a pseudo-functor, which by virtue of our identification,
we think of as coming from an pseudo-functor F : C → B between bicategories. It is clear
that for any objects x, y of C, F induces a functor
F(y);x : {y, x} −→ {F (y), F (x)}.
B.5. Definition. Let x be a pseudo-monoid in C. Then F (x) acquires the structure of a

pseudo-monoid in B by virtue of the composition:
t / Mx F• /
N MF (x),
with the full pseudo-monoid structure for MF (x) results from the composition
tot F• oF•
NoN / Mx o Mx / MF (x) o MF (x)
AI DL
◦ θ ◦ ε
◦
F•

N t / Mx / MF (x)
C. Pentagons
Pentagon diagrams express the coherence condition in a monoidal category. This condition
is replaced by a diagram of the form (B.3) or (B.4) if the monoidal structure is given
in unbiased form [Lei04]. The actual pentagon can be recovered if these diagrams are
specialized to arities equal to 4 and the monoidal category comes from a monoid inhabiting
a 2-category. This shows the equivalence of the biased and unbiased definitions. A slight
generalization of the pentagon occurs if the monoid inhabits a bicategory, and both these
situations arise in the main text. In addition, conditions arising from pentagons are cocycle
conditions in appropriate cohomology theories. It is useful to compute them once and for
all in the general setting of a monoid object in a bicategory.
To begin with, observe that by applying (B.1) and (B.2) to the objects (2; 2, 1)
and (2; 1, 2) ∈ N o N we obtain morphisms θ2;2,1 : t(3) → t(2)(t(2), t(1)), θ2;1,2 : t(3) →
t(2)(t(1), t(2)), and combining these two we obtain
(C.1) µ : t(2)(t(2), t(1)) −→ t(2)(t(1), t(2)).
The pentagons (or the diagrams related to them) arise from the decomposition of the
operation t(4), by way of (B.4), down to terms only involving the binary and unary
operations t(2) and t(1), respectively. We can assume that the unary operation t(1) : x → x
coincides with the identity idx . These decompositions can be encoded by working our way
along the small pentagon in the diagram (B.3), so for instance one of them corresponds to
the sequences
((2; 2, 1); 2, 1, 1) / (2; (2; 2, 1), 1)

(3; 2, 1, 1) / 4o (2; 3, 1)
(We have simply written 1 in place of the more accurate but cumbersome expression (1; 1).)
There are six distinct sequences including the one above. Their starting points, counted
from the upper left corner of the small pentagon in the diagram (B.3), are:
1. ((2; 2, 1); 2, 1, 1),
2. ((2; 2, 1); 1, 2, 1),
3. ((2; 1, 2); 1, 2, 1),
4. ((2; 1, 2); 1, 1, 2),
5. ((2; 1, 2); 2, 1, 1),
6. ((2; 2, 1); 1, 1, 2).
Let us also use the notations mi = t(i), i = 1, . . . , 4, with the special provision m1 = idx =
m2 (id(m2 ), m2 (id, id)) / m2 (id, m2 )(m2 , id, id) 958
O < O g
(C.2)
µ
m2 (m2 (id, id), id(m2))

h
5 m2 (m2 , id)(m2 , id, id)
7 _
m2 (m2 , m2 ) o m3 (m2 , id, id)

> Y E
~ s
m2 (m2 , id)(id, hid, m2 ) 6 1 m2 (m7 2 (m2 , id), id)
µ m3 (id, id, m2 ) o m4 / m2 (m3 , id) µ
v '
ETTORE ALDROVANDI
m2 (id, m2 )(id, id, m2 )

`
4 2 m2 (m2 (id, m2 ), id)

m2 (id, m3 ) m3 (id, m2 , id)
v '
3 m2 (m2 , id)(id, m2 , id)
id. With these, applying (B.4) and (B.3), we obtain the diagram on page 958:
m2 (id, m2 (id,h m2 ))
µ µ
w
m2 (id, m2 (m2 , id)) / m2 (id, m2 )(id, m2 , id)
The six petals correspond to the indicated sequences and are numbered accordingly. We
have marked the (magenta) arrows resulting from the morphism (C.1). Note that there
are five of them. In particular:
• If MC is unital, the petals 5 and 6 collapse to the red arrows.
• If furthermore MC is in fact a 2-multicategory, i.e. the 2-categorical analog of a

multicategory, so that C itself is a genuine 2-category, the associativity morphisms
at the tops of all petals reduce to identities. As a result, the outer perimeter reduces
to a standard pentagon
(C.3)
m2 (m2 ,X m2 ) l
µ
m2 (m2 (m2 , id), id)

:
m4 / m2 (m3 , id) µ
$
m2 (m
5 2
(id, m2 ), id)

m2 (id, m3 ) m3 (id, m2 , id)
µ
x
m2 (id, m2 (id, m2 ))
f
µ
y
m2 (id, m2 (m2 , id))
D. Some lemmas on symmetric braidings

D.1. Lemma. Let (C, ⊗, c) be a strictly associative braided monoidal category. For every
x, y, z, w, the diagram
1⊗cy,z ⊗1
x⊗y⊗z⊗w / x⊗z⊗y⊗w
cx,y ⊗cz,w cx⊗z,y⊗w

y⊗x⊗w⊗z / y⊗w⊗x⊗z
1⊗cx,w ⊗1
commutes if and only if the braiding c is symmetric.

Proof. The naturally occurring diagram relating cx⊗z,y⊗w to cx,y ⊗ cz,w is
1⊗cz,y ⊗1
x⊗y⊗z⊗wo x⊗z⊗y⊗w
cx,y ⊗cz,w cx⊗z,y⊗w

y⊗x⊗w⊗z / y⊗w⊗x⊗z
1⊗cx,w ⊗1
which is valid with respect to any braiding. This is equal to the diagram in the statement
if and only if cy,z is the inverse of cz,y .
D.2. Lemma. Let (C, ⊗, c) be a strictly associative braided monoidal category. If the
braiding c is symmetric, the following diagram
1⊗cyz,uv ⊗1
x⊗y⊗z⊗u⊗v⊗w / x⊗u⊗v⊗y⊗z⊗w
cx,y ⊗1⊗1⊗cv,w cx.u ⊗1⊗1⊗cz,w

y⊗x⊗z⊗u⊗w⊗v u⊗x⊗v⊗y⊗w⊗z
1⊗cxz,uw ⊗1 1⊗cxv,yw ⊗1

y⊗u⊗w⊗x⊗z⊗v / u⊗y⊗w⊗x⊗v⊗z
cy.u ⊗1⊗1⊗cz,v
commutes.
Proof. Use Lemma D.1 to replace cyz,uv with cy,u ⊗ cz,v and similarly for cxz,uw and cxv,yw .
This converts the diagram in the statement into the following one
1⊗cy,u ⊗cz,v ⊗1
x⊗y⊗u⊗z⊗v⊗w / x⊗u⊗y⊗v⊗z⊗w
cx,y ⊗1⊗1⊗cv,w cx.u ⊗1⊗1⊗cz,w

y⊗x⊗u⊗z⊗w⊗v u⊗x⊗y⊗v⊗w⊗z
1⊗cx,u ⊗cz,w ⊗1 1⊗cx,y ⊗cv,w ⊗1

y⊗u⊗x⊗w⊗z⊗v / u⊗y⊗x⊗w⊗v⊗z
cy.u ⊗1⊗1⊗cz,v
which consists of two juxtaposed copies of the same kind of hexagon:
1⊗cy,u
x⊗y⊗u / x⊗u⊗y
cx,y ⊗1 cx.u ⊗1

y⊗x⊗u u⊗x⊗y
1⊗cx,u 1⊗cx,y

y⊗u⊗x / u⊗y⊗x
cy.u ⊗1
The latter commutes in any braided monoidal (strictly associative) category.

A permutative category is a symmetric monoidal category (C, ⊗, c) in which associa-
tivity and unitality hold strictly. Thus the diagrams in Lemmas D.1 and D.2 commute
if and only if the category (C, ⊗, c) is permutative. On the other hand, the statements
of Lemma D.1 and Lemma D.2 hold in general. Requiring that C be strictly associative
merely simplifies the diagrams and overall legibility. For example, in the general case the
diagram in the statement of Lemma D.1 would need replacing with:
cx⊗z,y⊗w
x ⊗ (z ⊗ (y ⊗ w)) o (x ⊗ z) ⊗ (y ⊗ w) / (y ⊗ w) ⊗ (x ⊗ z) / y ⊗ (w ⊗ (x ⊗ z))
O O
x ⊗ ((z ⊗O y) ⊗ w) y ⊗ ((w ⊗O x) ⊗ z)
1⊗cy,z ⊗1 1⊗cx,w ⊗1
x ⊗ ((y ⊗ z) ⊗ w) y ⊗ ((x ⊗ w) ⊗ z)

x ⊗ (y ⊗ (z ⊗ w)) o (x ⊗ y) ⊗ (z ⊗ w) cx,y ⊗cz,w
/ (y ⊗ x) ⊗ (w ⊗ z) / y ⊗ (x ⊗ (w ⊗ z))
with a similarly modified proof. A similar, but more complicated modifications apply to
the diagram in the proof of Lemma D.2.
E. Hypercohomology computations
In order to carry out a full, explicit cohomology computation for the class of a ring-like
stack R, it is necessary to resolve various simplicial objects related to the ring A, notably
the Eilenberg-Mac Lane’s K(A, n) for n = 2, 3, by way of certain bisimplicial objects.
In the main text, and specifically in sect. 15, we simply referred to the spectral sequence
for the hypercohomology of the simplicial object K(A, 1)17 and focus on its E 0,3 -term,
showing that it is isomorphic to HML3 (A, M ) proper. While this is appropriate to show
the agreement with Mac Lane cohomology, it is important to provide the main outline of
the full calculation.
17
As done in ref. [Bre99, §6], discussing the passage from a monoidal category to a monoidal stack
It is easier to illustrate the full computation in the (known) cases of H 3 (K(A, 1), M ),
or H 4 (K(A, 2), M ) first, that is, by just considering the underlying structure of braided
(symmetric) stack of R. The full case just calls for the addition of more data following
the same pattern.
As it is well known, R is only a gerbe over A (see e.g. [LMB00]), and the decomposition
of $ : R → A requires choosing local objects. Thus we need to choose covers ξn : Un,0 → An ,
which we complete to an hypercover of K(A, 1). Following ref. [Fri82], this means an
augmented bisimplicial object
U•,• −→ K(A, 1)• ,
such that each Un,• → An is a hypercover. (In fact, and it is convenient to do so, U can
be arranged so that its horizontal zero level U•,0 → K(A, 1) is an acyclic fibration: this
ensures it becomes finer in the horizontal direction so as to guarantee the existence of
local choices.) We denote by di,h (resp. di,v ) the face maps of U in the horizontal (resp.
vertical) direction, where the “horizontal direction” is the direction of K(A, 1).
Let us use the conventions:
1. a = ξ1 : U1,0 → A,
2. (a, b) = ξ2 : U2,0 → A × A,
3. (a, b, c) = ξ3 : U3,0 → A × A × A,
etcetera, where of course the meaning of the letters a, b, c, . . . depends on the level n and
it is determined, in effect, by the simplicial face maps. Thus, at level n = 2 we have
a = d2 ξ2 = ξ1 dh,2
b = d0 ξ2 = ξ1 dh,0
a + b = d1 ξ2 = ξ1 dh,1 ,
whereas at level n = 3 we have, with the same conventions:
a = d2 d3 ξ3 = ξ1 dh,2 dh,3
b = d0 d3 ξ3 = ξ1 dh,0 dh,3
c = d0 d1 ξ3 = ξ1 dh,0 dh,1
a + b = d1 d3 ξ3 = ξ1 dh,1 dh,3
b + c = d0 d2 ξ3 = ξ1 dh,0 dh,2
a + b + c = d1 d2 ξ3 = ξ1 dh,1 dh,2 .
The choice of an object X of the pullback Ra = ξ1∗ R determines the class of R,

as a gerbe over A with band M , in H 2 (A, M ). This class is obtained by computing
it in the standard way along the hypercover U1,• → A. As it is well known, the class
itself is represented by an object of α ∈ M (U1,2 ), which is simply the defect in the
commutativity of the triangle formed by the three possible pullbacks to U1,2 of the
∼
isomorphism ϕ : d∗1,v X → d∗0,v X defined over U1,1 (see [Bre94b, Bre94a]). As a result, α is
a Čech cocycle relative to the hypercover U1,• :
(E.1) d∗0,v α + d∗1,v α − d∗2,v α + d∗3,v α = 0 .
In the horizontal direction, by successively pulling back X along the horizontal face maps at
vertical level n = 0 we obtain objects Xa , Xa+b , . . . (recall the convention above) providing
decompositions for the various pullbacks d∗i,h . . . d∗j,h R over An .
At level n = 2, over U2,0 , we get an isomorphism
∼
σa,b : Xa + Xb −→ Xa+b ,
∼
which is only compatible with the various isomorphisms ϕa : d∗1,v Xa → d∗0,v Xa up to an
automorphism, say δ ∈ M (U2,1 ), of d∗0,v Xa+b . A calculation of the pullbacks to U2,2, yields
(E.2) d∗0,v δ − d∗1,v δ + d∗2,v δ = −d∗0,h α + d∗1,h α − d∗2,h α .
Observe that modulo α, the element δ gives the class in H 1 (A × A, M ) of the sheaf
Hom(Xa + Xb , Xa+b ).
Also, still over U2,0 , the braiding of R gives the diagram
Xa + Xb / Xb + Xa
σa,b σb,a

Xa+b / Xa+b
g+ (a,b)
defining g+ (a, b) ∈ M (U2,0 ). If the braiding is in addition symmetric, then the previous
diagram yields
g+ (a, b) + g+ (b, a) = 0,
namely g+ is antisymmetric under the map determined by the pullback of U2,• by the swap
map of A2 . If we denote this operation by τ , then the above relation would be written
g+ + τ ∗ g+ = 0
Similarly, an analysis of the associativity of the monoidal structure of R yields
f+ (a, b, c) ∈ M (U3,0 ), as an automorphism of the object Xa+b+c of d∗2,h d∗1,h R over A3 .
Under pullback of both f+ and g+ to U3,1 we obtain:
d∗0,v f+ − d∗1,v f+ = d∗0,h δ − d∗1,h δ + d∗2,h δ − d∗3,h δ

(E.3)
d∗0,v g+ − d∗1,v g+ = −δ + τ ∗ δ.
Finally, the familiar relations describing a (symmetric) braiding arise by analyzing the
behavior of f+ when pulled back to U4,0 and that of g+ when pulled back to U3,0 , yielding
the standard cocycle relation
d∗0,h f+ − d∗1,h f+ + d∗2,h f+ − d∗3,h f+ + d∗4,h f+ = 0,

plus the two equations arising from Mac Lane’s hexagonal diagrams (and the antisymmetry
condition above, if R is symmetric).
Again, observe that modulo δ, f+ and g+ define objects in H 0 (A4 , M ) and H 0 (A3 , M )
satisfying the standard cocycle relations for a class in H 4 (K(A, 2), M ) (or H 5 (K(A, 3), M )).
This is the relevant part in the discussion of the class arising from a braided (or symmetric)
monoidal stack as above.
In summary, the quadruplet (f+ , g+ , δ, α) defines a cocycle of degree 3 with respect to
the total complex defined as follows. Let B(A, k) be the iterated bar construction on A,
where k = 2, 3. If k = 1, then B(A, 1) is simply the complex Z[K(A, 1)]∼ .18 With a mild
abuse of language, for each (vertical) level n let B(U•,n , k) denote the complex obtained in
a way analogous to B(A, k) from the abelian sheaves Z[U•,n ]. Thus we get an augmented
simplicial object
/ / ξ
··· / B(U•,1 , k) / B(U•,0 , k) / B(A, k).
/
Then we form (neglecting B(A, k)) the total complex with respect to the vertical direction
to obtain the necessary complex.
A glance at the complex used in the main text, section 15, namely the bar complex
built on top of a DGA structure on the complexes L2 (A) (or L3 (A)), reveals that the same
construction with the hypercover
U•,• → K(A, 1)
would also work in this case, with the difference that in total degree 3 we have the
quintuplet used in the paper, namely (f, α1 , α2 , f+ , g+ ) plus the automorphisms α and δ
as above arising from the decomposition of R over A and A × A, supplemented by an
additional one, ε ∈ M (U2,1 ), arising from the descent condition on the pullback of the
various biextensions Ea,b and their higher analogs.
To accommodate for these changes, we must form simplicial objects corresponding to
the bar complexes B̄k,• (A), k = 2, 3, in section 15.3. That is, for each complex B(U•,n , k)
above we apply the (normalized) bar construction B̄k (U•,n ) := B̄(B(U•,n , k), η) to form
the augmented simplicial object
/ / ξ
(E.4) ··· / B̄k (U•,1 ) / B̄k (U•,0 ) / B̄k,• (A) ,
/
for k = 2, 3. In addition to our notational conventions, we must consider appropriate

ones for the multiplicative structures: we consider products ab = (d2 ξ2 )(d0 ξ2 ), abc =
(d2 d3 ξ3 )(d0 d3 ξ3 )(d0 d1 ξ3 ), etc. and use the same notation for the corresponding maps at
any level in the vertical direction.
To explain the required steps to obtain the full class in some more detail, consider the
biexact bifunctors
ma,b : Ra × Rb −→ Rab
18
A down-shifted version is the complex denoted L2 (A) or L3 (A) in the paper.
defined over U2,0 (again, recall the above convention about naming the components of
ξ1 , ξ2 , . . . ) corresponding to the biextension Ea,b introduced in sect. 15.11, and their higher
arity analogs. In terms of local objects, we have an isomorphism
∼
ma,b : Xa Xb −→ Xab ,
which we can safely indicate with the same name, between objects of Rab . Analogously to
the case of the sum operation, ma,b is only compatible with the (horizontal) pullbacks of
∼
ϕ : d∗1,v X → d∗0,v X up to an automorphism ε ∈ M (U2,1 ). For added clarity, let us specify
the labels as ε(a, b). Comparing the three possible pullbacks to U2,2 yields the relation
(E.5) d∗0,v ε(a, b) − d∗1,v ε(a, b) + d∗2,v ε(a, b) = −aα(b) + α(ab) − α(a)b ,
where a notation like α(ab) means the pullback of α ∈ M (U1,2 ) along the analog B̄k (U2,2 ) →
B̄k (U1,2 ) of the map ab, as mentioned above.
The remaining three relations are found by analyzing the behavior of the diagrams (15.14), (15.8)
which define the quantities f (a, b, c), α1 (a, b; c), and α2 (a; b, c), under pullback from U3,0
to U3,1 . They can be readily computed by interpreting f (a, b, c), α1 (a, b; c), and α2 (a; b, c)
as 2-arrows in the diagrams
Ra × Rb × Rc
ma,b ×1 1×mb,c
x f (a,b,c)
&
Rab × Rc +3 Ra × Rbc
mab,c ma,bc
& x
Rabc
and
Ra × Rb × Rc Ra × Rb × Rc
ma,c ×mb,c σa,b ×1 ma,b ×ma,c 1×σb,c
x −α1 (a,b;c)
& x α2 (a;b,c)
&
Rac × Rbc +3 Ra+b × Rc , Rab × Rac +3 Ra × Rb+c ,
mac,bc ma+b,c mab,ac ma,b+c

& x & x
R(a+b)c Ra(b+c)
and then comparing, for each diagram, the two possible pullbacks to U3,1 . We obtain the
following relations:
(E.6)
d∗0,v f (a, b, c) − d∗1,v f (a, b, c) = −aε(b, c) + ε(ab, c) − ε(a, bc) + ε(a, b)c
d∗0,v α1 (a, b; c) − d∗1,v α1 (a, b; c) = −δ(ac, bc) + δ(a, b)c − ε(a, c) + ε(a + b, c) − ε(b, c)
d∗0,v α2 (a, b; c) − d∗1,v α2 (a, b; c) = δ(ab, ac) − aδ(b, c) + ε(a, b) − ε(a, b + c) + ε(a, c) .
In summary, the relations (E.1), (E.2) and (E.5), and (E.3) and (E.6), together with the
ones found in the main text, sect. 15.13, over U4,0 , show the entire collection
((f, α1 , α2 , f+ , g+ ); (δ, −ε), α)
forms a cocycle of total degree 3 in the double complex obtained from (E.4) for k = 3
(minus the term in degree −1, of course). In particular, the relations (E.3) and (E.6)
feature the horizontal differential of (δ, −ε), which has horizontal degree 2, according to
sect. 15.4. Similarly, the relations (E.2), (E.5) feature the horizontal differential of α, with
horizontal degree 1.
References
[Ald15] Ettore Aldrovandi. Stacks of ann-categories and their morphisms. Theory and
Applications of Categories, 30(39):1256–1286, 2015-09-21. URL: http://www.
tac.mta.ca/tac/volumes/30/39/30-39abs.html, arXiv:1501.07592.
[AN09] Ettore Aldrovandi and Behrang Noohi. Butterflies i: Morphisms of 2-group
stacks. Advances in Mathematics, 221:687–773, 2009. arXiv:0808.3627,
doi:doi:10.1016/j.aim.2008.12.014.
[AN10] Ettore Aldrovandi and Behrang Noohi. Butterflies II: torsors for 2-group
stacks. Advances in Mathematics, 225:922–976, 2010. 10.1016/j.aim.2010.03.011,
arXiv:0909.3350 [math.AT].
[BDRR13] Nils A. Baas, Bjørn Ian Dundas, Birgit Richter, and John Rognes. Ring
completion of rig categories. J. Reine Angew. Math., 674:43–80, 2013.
[Ber07] Clemens Berger. Iterated wreath product of the simplex category and iterated
loop spaces. Advances in Mathematics, 213(1):230–270, 2007. URL: http://
www.sciencedirect.com/science/article/pii/S0001870806003938, doi:
http://dx.doi.org/10.1016/j.aim.2006.12.006.
[Boy10] Mitya Boyarchenko. Characters of unipotent groups over finite fields. Se-
lecta Math. (N.S.), 16(4):857–933, 2010. URL: http://dx.doi.org/10.1007/
s00029-010-0036-9, doi:10.1007/s00029-010-0036-9.
[Bre69] Lawrence Breen. Extensions of abelian sheaves and Eilenberg-MacLane algebras.
Invent. Math., 9:15–44, 1969.
[Bre83] Lawrence Breen. Fonctions thêta et théorème du cube, volume 980 of Lecture
Notes in Mathematics. Springer-Verlag, 1983.
[Bre90] Lawrence Breen. Bitorseurs et cohomologie non abélienne. In The Grothendieck
Festschrift, Vol. I, volume 86 of Progr. Math., pages 401–476. Birkhäuser Boston,
1990.
[Bre94a] Lawrence Breen. On the classification of 2-gerbes and 2-stacks. Astérisque,
(225):160, 1994.
[Bre94b] Lawrence Breen. Tannakian categories. In Motives (Seattle, WA, 1991),

volume 55 of Proc. Sympos. Pure Math., pages 337–376. Amer. Math. Soc.,
1994.
[Bre99] Lawrence Breen. Monoidal categories and multiextensions. Compositio Math.,

117(3):295–335, 1999.
[CS10] G. S. H. Cruttwell and Michael A. Shulman. A unified framework for generalized

multicategories. Theory Appl. Categ., 24:No. 21,580–655, 2010.
[Dat10] Swarnendu Datta. Metric groups attached to skew-symmetric biextensions.

Transform. Groups, 15(1):72–91, 2010. URL: http://dx.doi.org/10.1007/
s00031-010-9078-5, doi:10.1007/s00031-010-9078-5.
[Del73] Pierre Deligne. La formule de dualité globale, volume 3, pages vi+640. Springer-
Verlag, 1973.
[Del79] Pierre Deligne. Variétés de shimura: interprétation modulaire, et techniques

de construction de modèles canoniques. In Automorphic forms, representations
and L-functions (Proc. Sympos. Pure Math., Oregon State Univ., Corvallis,
Ore., 1977), Part 2, Proc. Sympos. Pure Math., XXXIII, pages 247–289. Amer.
Math. Soc., 1979.
[Dug14] Daniel Dugger. Coherence for invertible objects and multigraded homotopy
rings. Algebr. Geom. Topol., 14(2):1055–1106, 2014. URL: http://dx.doi.
org/10.2140/agt.2014.14.1055, doi:10.2140/agt.2014.14.1055.
[EML53] Samuel Eilenberg and Saunders Mac Lane. On the groups of H(Π, n). I. Ann.
of Math. (2), 58:55–106, 1953.
[EML54] Samuel Eilenberg and Saunders Mac Lane. On the groups H(Π, n). II. Methods
of computation. Ann. of Math. (2), 60:49–139, 1954.
[Fri82] Eric M. Friedlander. Étale homotopy of simplicial schemes, volume 104 of

Annals of Mathematics Studies. Princeton University Press, Princeton, N.J.;
University of Tokyo Press, Tokyo, 1982.
[Gir71] Jean Giraud. Cohomologie non abélienne. Springer-Verlag, 1971. Die

Grundlehren der mathematischen Wissenschaften, Band 179.
[GJ99] Paul G. Goerss and John F. Jardine. Simplicial homotopy theory, volume 174
of Progress in Mathematics. Birkhäuser Verlag, 1999.
[Gro72] A. Grothendieck. Biextensions de faisceaux de groupes. In Groupes de mon-

odromie en géométrie algébrique. I, volume 288 of Lecture Notes in Mathematics,
exposé VII, pages 133–217. Springer-Verlag, 1972. Séminaire de Géométrie
Algébrique du Bois-Marie 1967–1969 (SGA 7 I), Dirigé par A. Grothendieck.
Avec la collaboration de M. Raynaud et D. S. Rim.
[JP07] Mamuka Jibladze and Teimuraz Pirashvili. Third mac lane cohomology via
categorical rings. J. Homotopy Relat. Struct., 2(2):187–216, 2007. arXiv:
math/0608519.
[JS93] André Joyal and Ross Street. Braided tensor categories. Adv. Math., 102(1):20–
78, 1993.
[Kel72] G. M. Kelly. Many-variable functorial calculus. I. In Coherence in categories,

pages 66–105. Lecture Notes in Math.,Vol. 281. Springer, Berlin, 1972.
[Kel74] G. M. Kelly. On clubs and doctrines. In Category Seminar (Proc. Sem., Sydney,
1972/1973), pages 181–256. Lecture Notes in Math.,Vol. 420. Springer, Berlin,
1974.
[Lap72a] Miguel L. Laplaza. Coherence for distributivity. In Coherence in categories,

pages 29–65. Lecture Notes in Math.,Vol. 281. Springer, Berlin, 1972.
[Lap72b] Miguel L. Laplaza. A new result of coherence for distributivity. In Coherence

in categories, pages 214–235. Lecture Notes in Math.,Vol. 281. Springer, Berlin,
1972.
[Lei04] Tom Leinster. Higher operads, higher categories, volume 298 of London
Mathematical Society Lecture Note Series. Cambridge University Press,
Cambridge, 2004. URL: http://dx.doi.org/10.1017/CBO9780511525896,
doi:10.1017/CBO9780511525896.
[LMB00] Gérard Laumon and Laurent Moret-Bailly. Champs algébriques, volume 39

of Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of
Modern Surveys in Mathematics [Results in Mathematics and Related Areas.
3rd Series. A Series of Modern Surveys in Mathematics]. Springer-Verlag,
2000.
[Lod98] Jean-Louis Loday. Cyclic homology, volume 301 of Grundlehren der Mathe-
matischen Wissenschaften [Fundamental Principles of Mathematical Sciences].
Springer-Verlag, second edition, 1998. Appendix E by Marı́a O. Ronco, Chapter
13 by the author in collaboration with Teimuraz Pirashvili.
[ML58] Saunders Mac Lane. Extensions and obstructions for rings. Illinois J. Math.,
2:316–345, 1958.
[ML95] Saunders Mac Lane. Homology. Springer-Verlag, 1995. Reprint of the 1975
edition.
[Mum69] David Mumford. Bi-extensions of formal groups. In Algebraic Geometry

(Internat. Colloq., Tata Inst. Fund. Res., Bombay, 1968), pages 307–322.
Oxford Univ. Press, 1969.
[Noo05] Behrang Noohi. On weak maps between 2-groups, 2005. arXiv:arXiv:math/

0506313[math.CT].
[Pis14] Claudio Pisani. Sequential multicategories, 2014. arXiv:1402.0253.
[QHT08] Nguyen Tien Quang, D. D. Hanh, and N. T. Thuy. On the axiomatics of

ann-categories. JP J. Algebra Number Theory Appl., 11(1):59–72, 2008.
[Qua03] Nguyen Tien Quang. Structure of ann-categories and mac lane-shukla coho-
mology. East-West J. Math., 5(1):51–66, 2003.
[Qua13] Nguyen Tien Quang. Cohomological classification of Ann-categories. Math.

Commun., 18(1):151–169, 2013.
[Rob11] Marcy D. Robertson. The Homotopy Theory of Simplicially Enriched Multi-

categories. ArXiv e-prints, 2011-11. arXiv:1111.4146.
[Shu10] Michael Shulman. Extraordinary multicategories, 2010. 2010 Canadian

Mathematical Society Summer Meeting. URL: http://home.sandiego.edu/
~shulman/papers/cms2010talk.pdf.
[Tho95] R. W. Thomason. Symmetric monoidal categories model all connective spectra.
Theory Appl. Categ., 1:No. 5,78–118 (electronic), 1995.
Department of Mathematics, Florida State University

1017 Academic Way, Tallahassee, FL 32312-4510, USA
Email: aldrovandi@math.fsu.edu
This article may be accessed at http://www.tac.mta.ca/tac/
THEORY AND APPLICATIONS OF CATEGORIES (ISSN 1201-561X) will disseminate articles that
significantly advance the study of categorical algebra or methods, or that make significant new contributions
to mathematical science using categorical methods. The scope of the journal includes: all areas of pure
category theory, including higher dimensional categories; applications of category theory to algebra,
geometry and topology and other areas of mathematics; applications of category theory to computer
science, physics and other mathematical sciences; contributions to scientific knowledge that make use of
categorical methods.
Articles appearing in the journal have been carefully and critically refereed under the responsibility of
members of the Editorial Board. Only papers judged to be both significant and excellent are accepted for
publication.
Full text of the journal is freely available from the journal’s server at http://www.tac.mta.ca/tac/. It
is archived electronically and in printed paper format.
Subscription information Individual subscribers receive abstracts of articles by e-mail as they
are published. To subscribe, send e-mail to tac@mta.ca including a full name and postal address. For
institutional subscription, send enquiries to the Managing Editor, Robert Rosebrugh, rrosebrugh@mta.ca.
Information for authors The typesetting language of the journal is TEX, and LATEX2e is
required. Articles in PDF format may be submitted by e-mail directly to a Transmitting Editor. Please
obtain detailed information on submission format and style files at http://www.tac.mta.ca/tac/.
Managing editor. Robert Rosebrugh, Mount Allison University: rrosebrugh@mta.ca
TEXnical editor. Michael Barr, McGill University: barr@math.mcgill.ca
Assistant TEX editor. Gavin Seal, Ecole Polytechnique Fédérale de Lausanne:
gavin seal@fastmail.fm
Transmitting editors.
Clemens Berger, Université de Nice-Sophia Antipolis: cberger@math.unice.fr
Richard Blute, Université d’ Ottawa: rblute@uottawa.ca
Lawrence Breen, Université de Paris 13: breen@math.univ-paris13.fr
Ronald Brown, University of North Wales: ronnie.profbrown(at)btinternet.com
Valeria de Paiva: Nuance Communications Inc: valeria.depaiva@gmail.com
Ezra Getzler, Northwestern University: getzler(at)northwestern(dot)edu
Kathryn Hess, Ecole Polytechnique Fédérale de Lausanne: kathryn.hess@epfl.ch
Martin Hyland, University of Cambridge: M.Hyland@dpmms.cam.ac.uk
Anders Kock, University of Aarhus: kock@imf.au.dk
Stephen Lack, Macquarie University: steve.lack@mq.edu.au
F. William Lawvere, State University of New York at Buffalo: wlawvere@buffalo.edu
Tom Leinster, University of Edinburgh: Tom.Leinster@ed.ac.uk
Ieke Moerdijk, Utrecht University: i.moerdijk@uu.nl
Susan Niefield, Union College: niefiels@union.edu
Robert Paré, Dalhousie University: pare@mathstat.dal.ca
Jiri Rosicky, Masaryk University: rosicky@math.muni.cz
Giuseppe Rosolini, Università di Genova: rosolini@disi.unige.it
Alex Simpson, University of Ljubljana: Alex.Simpson@fmf.uni-lj.si
James Stasheff, University of North Carolina: jds@math.upenn.edu
Ross Street, Macquarie University: ross.street@mq.edu.au
Walter Tholen, York University: tholen@mathstat.yorku.ca
Myles Tierney, Université du Québec à Montréal : tierney.myles4@gmail.com
R. J. Wood, Dalhousie University: rjwood@mathstat.dal.ca

32 27 PDF

Uploaded by

Copyright:

Available Formats

32 27 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

32 27 PDF

Uploaded by

Copyright:

Available Formats

Theory and Applications of Categories, Vol. 32, No. 32, 2017, pp. 889–969.

BIEXTENSIONS, BIMONOIDAL FUNCTORS, MULTILINEAR

Abstract. We associate to a bimonoidal functor, i.e. a bifunctor which is monoidal in

Received by the editors 2015-08-05 and, in final form, 2017-07-28.

11 Compositions of n-butterflies 924

of pointed groupoids, where each Hi,• is a presentation of the corresponding Hi , and G•

We apply the previous observation to C = BGrSt or C = BXMod. We say that

where the exponent notation denotes the action.

2.4. Lemma. The coordinate χ : P ∧G Q → G has the expression:

(2.3) χu,v = {s(u), t(v)}−1 .

(2.4b) γ 2 : dv2 ∗ E ∧G dv0 ∗ E −→ dv1 ∗ E.

compatibility condition reads

λ1h,h0 ;k λ1h,h0 ;k0 λ2h;k,k0 λ2h0 ;k,k0

where the upper horizontal arrow is the “commuto-associativity” morphism

A natural transformation ϕ : F ⇒ F 0 between two biadditive morphisms is a natural

4. Biextensions by braided crossed modules

where f is the equivariant section into the trivial Π-torsor.

Dropping the commutativity of the partial multiplication laws, E becomes a biextension

4.5. Definition. A morphism of biextensions of H, K by (G, Π) is a morphism of the

The trivial biextension is I = (GH×K , eΠ ). More generally, a biextension will be

(4.3) ı(g)e = eı(g (e) ).

and the diagram

For a biextension which is globally of the form (GH×K , x), where x : H × K → Π

These cocycles are a coboundary if there exists u : H × K → G such that:

(4.8a) u(hh0 , k) g1 (h, h0 ; k) = u(h, k)u(h0 , k)

5. Biadditive morphisms and biextensions

This establishes a functor

and in fact one between stacks

(F ◦ dh2 ) (F ◦ dh0 ) =⇒ (F ◦ dh1 ) : H × H × K −→ G ,

The situation for the left action is completely analogous, with x0 = ∂g x.

6. Symmetry properties of biextensions

and similarly for the second partial multiplication.

of biextensions, which is evidently symmetric. We summarize the previous discussion as

1. (Restriction) s1 and s2 agree on H1 × K1 : (id, ∂)∗ s1 = (∂, id)∗ s2 .

2. (Compatibility) For all (h, z) ∈ H1 × K0 and (y, k) ∈ H0 × K1 , and e ∈ Ey,z , s1 and

A morphism ϕ : (E, s1 , s2 ) → (E 0 , s01 , s02 ) is a morphism of the underlying biextensions pre-

s1 (h, ∂k) = s2 (∂h, k).

Let s : H1 × K1 → E be the resulting morphism. The next lemmas characterize these

7.3. Lemma. The trivializations s1 and s2 , as maps s1 : H1 × K0 → E and s2 : H0 × K1 →

s(h, k) ×1 s(h0 , k) = s(hh0 , k)

pullbacks to H1 × K0 and H0 × K1 . (The required triviality conditions are the same as

(7.7a) u1 (hh0 , z) g1 (∂h, ∂h0 ; z) = u1 (h, z) u1 (h0 , z)

(7.7d) u2 (yy 0 , k) g1 (y, y 0 ; ∂k) = u2 (y, k) u2 (y 0 , k)

(7.8a) u(hh0 , k) g1 (∂h, ∂h0 ; ∂k) = u(h, k)u(h0 , k)

for all h, h0 ∈ H1 and k, k 0 ∈ K1 .

8. Biadditive morphisms and butterflies

In essence, the theorem states that any biadditive functor F : H × K → G can be

To get a biadditive functor from a butterfly in functorial way, we proceed as in ref.

[Gir71, Proposition III 1.2.7]. This extends to (G1 , G0 )-torsors by assigning to m : G1 → P

e 7−→ e ×1 s1 ({y 0 , y}H , z) e0 7−→ e0 ×2 s2 (y, {z 0 , z}K ).

Thus, for y, y 0 ∈ H0 and z, z 0 ∈ K0 the following diagrams must commute:

(H0 , yy 0 ) −→ (H0 , y 0 y) and (K0 , zz 0 ) −→ (K0 , z 0 z)

in Tors(H1 , H0 ) and Tors(K1 , K0 ), respectively. This requires writing η 1 and η 2 (going

e0 ×1 e = e ×1 e0 ×1 s1 ({y 0 , y}−1 {(e), (e0 )}G ;

and for e ∈ Ey,z , e0 ∈ Ey,z0 :

e0 ×2 e = e ×2 e0 ×2 s2 (y, {z 0 , z}−1 {(e), (e0 )}G .

of groupoids. Furthermore, if G is braided symmetric, u (from Theorem 8.6) becomes

Homb (H1 × H2 , K ; G ) ' Homb (H1 , K ; G ) × Homb (H2 , K ; G ) ,

γ 1 : d∗2 E ∧G1 d∗0 E −→ d∗1 E