Smash Products for Non-cartesian Internal Prestacks
Liang Ze Wong

TL;DR
This paper generalizes the smash product construction to prestacks internal to non-cartesian monoidal categories, unifying the Grothendieck construction and smash products for B-module algebras.
Contribution
It introduces a new smash product analogue for internal prestacks in non-cartesian monoidal categories, extending existing constructions.
Findings
Provides a unified framework for internal prestacks and smash products.
Recovers original prestacks via fibers or coinvariants.
Generalizes classical constructions to a broader categorical context.
Abstract
The smash product construction (or the Grothendieck construction) takes a functor (or prestack) and returns a fibration . In this paper, we develop an analogue of the smash product for prestacks internal to a non-cartesian monoidal category. Our construction simultaneously generalizes the Grothendieck construction for prestacks and smash products for -module algebras over a bialgebra . Further, taking fibers or coinvariants allows one to recover the original prestack.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAlgebraic structures and combinatorial models · Advanced Operator Algebra Research · Advanced Topics in Algebra
Smash Products for Non-cartesian Internal Prestacks
Liang Ze Wong
Abstract
The smash product construction (or the Grothendieck construction) takes a functor (or prestack) and returns a fibration . In this paper, we develop an analogue of the smash product for prestacks internal to a non-cartesian monoidal category. Our construction simultaneously generalizes the Grothendieck construction for prestacks and smash products for -module algebras over a bialgebra . Further, taking fibers or coinvariants allows one to recover the original prestack.
Contents
- 1 Introduction
- 2 Comonoids and comodules
- 3 Comonoidal internal categories
- 4 Internal Prestacks
- 5 Smash products
- 6 Coinvariants of comodule categories
- 7 Further work
1 Introduction
Given a group acting on another group via a homomorphism , we may form the semi-direct product , or simply . There is also a projection , and taking the kernel of allows us to recover . This paper synthesizes two classical generalizations of the semi-direct product.
The first is the Grothendieck construction [grothendieck1961categories]. Instead of a group acting on another group , we now have a category acting on a family of other categories via a functor sending to . Such functors are also known as (split) prestacks. The Grothendieck construction then takes a split prestack and returns a fibration whose fibers allow us to recover the categories that we started with.
The second generalization is the smash product construction [cohen1984group]. This time, instead of a group acting on another group, we start with a group acting on a -algebra . We may then form the smash product (or ), which is another -algebra. Instead of an algebra homomorphism , we have a -grading on whose identity component is the original algebra ; equivalently, we have a -comodule algebra whose coinvariant subalgebra is . More generally, given a Hopf algebra acting on another algebra (i.e. a -module algebra), we may form the smash product which is a -comodule algebra, and taking the coinvariant subalgebra of allows us to recover [blattner1985duality, van1984duality]. Although the antipode of the Hopf algebra is used in the definition of the smash product, it is not actually required: we may in fact form the smash product for a bialgebra acting on , which coincides with the usual smash product if is a Hopf algebra.
The starting point of this paper is the observation that categories and bialgebras are both examples of internal categories [aguiar1997internal]. In fact, they are comonoidal internal categories (which we define in Section 3), and we may thus define comodule categories and prestacks over them (Section 4). In Section 5, we define smash products of prestacks, and in Section 6 we show that taking coinvariants allows us to recover the original prestack. Some necessary lemmas regarding comonoids and comodules will be provided in Section 2.
The reader might find many of the statements and proofs in this paper rather technical and unmotivated. This is because they were developed in the following manner:
Identify a notion for ordinary categories (i.e. categories internal to ); 2. 2.
Define this notion for categories internal to an arbitrary monoidal category , in the language of comonoids and comodules; 3. 3.
Prove the necessary statements using string diagrams; 4. 4.
Transfer this proof into commutative diagrams.
Consequently, the results and proofs that end up in this paper are already one step removed from the original method of proof (string diagrams), and three steps removed from the original motivation (ordinary category theory)! Future versions of this paper might attempt to better motivate the results, and present them using string diagrams. For now, we encourage the reader to keep the original categorical constructions in mind and work out the statements and proofs for themselves in string diagrams.
2 Comonoids and comodules
In this section, we give a quick overview of comonoids and comodules. Throughout, we assume that is a symmetric monoidal category, where denotes the symmetry. We will further assume that is regular in the following sense:
Definition 2.1** ([aguiar1997internal]*Definition 2.1.1).**
A monoidal category is regular if it has all equalizers, and preserves them (in both variables). In other words, if {E}$${X}$$\scriptstyle{\mathsf{eq}} is the equalizer of
{X}$${Y,}$$\scriptstyle{f}$$\scriptstyle{g}
then is the equalizer of
[TABLE]
For comonoids in , let denote the category of left -, right -bicomodules, or -comodules. When either or is the monoidal unit , we write and . The maps in are comodule maps respecting both the and coactions. More generally, we have:
Definition 2.2**.**
Let be a comonoid map, and .
A (comodule) map over is a map such that the diagram on the left commutes, where denotes the respective right coactions.
[TABLE]
We use the diagram on the right as an abbreviation of the diagram on the left. In particular, the dotted arrows indicate that has a -coaction and has a -coaction. In the special case where and , we say that is a map over . We may similarly define maps over for left comodules. For bicomodules, we may define maps over , or simply maps over if . Thus, maps in , and are maps over .
Lemma 2.3**.**
Let be a comonoid map, and .
A map over is equivalently a -comodule map , where is the corestriction along .
Definition 2.4**.**
Let be comonoids, and let and . The cotensor over of and is the equalizer:
[TABLE]
Proposition 2.5** ([aguiar1997internal]*Proposition 2.2.1).**
When is a regular, has a right -coaction induced by the coaction on :
[TABLE]
Similarly, has a left -coaction making an object of .
Lemma 2.6** ([aguiar1997internal]*Lemma 7.1.1).**
Let and .
[TABLE]
Then there is a canonical isomorphism in
[TABLE]
natural in and .
It is further shown in [aguiar1997internal]*§2.2 that cotensoring extends to a functor
[TABLE]
In particular, if and are maps in and , there is a -map . More generally, we have:
Proposition 2.7**.**
Let and be comonoid maps, and let and , and suppose we have over and over .
[TABLE]
Then there is a map over .
Proof.
The map is induced by:
[TABLE]
This is a comodule map over if the left-most face of the following diagram commutes,
[TABLE]
where we have omitted and for brevity. But both composites that make up the left-most face are maps uniquely induced by the diagonal map , hence are equal. Similarly, is a comodule map over . ∎
Theorem 2.8** ([aguiar1997internal]*Theorem 2.2.1).**
There is a bicategory whose objects are comonoids in , and whose category of arrows from to is .
Corollary 2.9**.**
For a comonoid in , is a monoidal category.
We conclude this section with some useful lemmas.
Lemma 2.10**.**
Let and be comonoids in . If each is in , and and are maps over and , then is also a comonoid.
Proof.
By Proposition 2.7, since each is a map over , we have a map , which we may compose with the isomorphism from Lemma 2.6 to obtain a comultiplication:
[TABLE]
Similarly, since each is a comodule map over , we have a counit
[TABLE]
The reader may verify that these maps make a comonoid. ∎
Lemma 2.11**.**
Let and be comonoids, and suppose that is a -comodule with coaction . Then is a comonoid map if and only if is a map over :
[TABLE]
Proof.
Note that always preserves counits, so is a comonoid map if and only if it also preserves comultiplication.
The diagram in the lemma commutes precisely when the left pentagon in the following diagram commutes:
[TABLE]
The outer square then says that is a comonoid map111Note that we need to be symmetric, not just braided, for the bottom-right corner to commute!.
Conversely, if is a comonoid map, the left pentagon in the following diagram commutes:
[TABLE]
The outer square then says that is a map over . ∎
Lemma 2.12**.**
Let be comonoids, and be a -coaction that is also a comonoid map. Then is induced by a comonoid map .
Proof.
The counit is a comonoid map, so the composite
[TABLE]
is a comonoid map. The left square of the following diagram commutes because is a comonoid map; the upper-right square commutes because is a coaction.
[TABLE]
The outer diagram then says that induces . ∎
Remark 2.13**.**
The converse of Lemma 2.12 does not hold: given an arbitrary comonoid map , the coaction
[TABLE]
need not be a comonoid map, because is not a comonoid map (unless is cocommutative). Thus the two equivalent conditions in Lemma 2.11 are stronger than the condition in Lemma 2.12.
Remark 2.14**.**
Note that for any comonoid map inducing a coaction , the following diagram always commutes:
[TABLE]
Lemma 2.15**.**
Let be a cocommutative comonoid, and with coaction . Then is a map over :
[TABLE]
Proof.
We need the following diagram to commute:
[TABLE]
Since , this is equivalent to the following diagram commuting,
[TABLE]
whose bottom-right square commutes because is cocommutative. ∎
3 Comonoidal internal categories
We take our definition of a category internal to a regular monoidal category from [aguiar1997internal]. Importantly, is not required to be cartesian i.e. the monoidal product is not necessarily the cartesian product .
Definition 3.1** ([aguiar1997internal]*Definition 2.3.1).**
A -internal category consists of a comonoid in and a monoid in .
In detail, an internal category is a tuple with
a comonoid of objects , with comultiplication and counit ; 2. 2.
a comodule of maps , with coactions222 for ‘source’ and for ‘target’. and ; 3. 3.
and identity and composition comodule maps
[TABLE]
satisfying associativity and unitality.
For brevity, we will sometimes refer to an internal category using subtuples such as .
Remark 3.2**.**
The definition of an internal category does not require the comonoid of objects to be cocommutative. However, it does not seem possible to define internal prestacks or the internal Grothendieck construction without cocommutativity of objects. The internal categories that we subsequently consider will all have cocommutative comonoids of objects. In such a situation, the left coaction induces a right coaction . Similarly, the right coaction induces a left coaction .
Example 3.3**.**
Any monoid in gives rise to the ‘one-object’ internal category . Any comonoid in gives rise to the ‘discrete’ internal category . (These are denoted and \underaccent{\check}{C} in [aguiar1997internal]*Example 2.4.1.)
Definition 3.4** ([aguiar1997internal]*Definition 4.1.1).**
Let and be internal categories in . An internal functor from to is a tuple where is a comonoid map and is a map such that the following diagrams commute:
[TABLE]
Definition 3.5**.**
Let denote the category of internal categories and functors.
Remark 3.6**.**
It is also possible to define internal transformations between internal functors, making a -category, but we will not need the -category structure in this paper.
Recall that if is a symmetric monoidal category, its category of comonoids is also symmetric monoidal, with the same braiding and monoidal product. A similar result holds for internal categories.
Proposition 3.7** ([aguiar1997internal]*§7.1).**
* is a monoidal category, with product*
[TABLE]
and unit .
Definition 3.8**.**
A comonoidal internal category is a comonoid in .
Proposition 3.9**.**
Let be a comonoidal internal category. Then:
* is cocommutative (i.e. and are comonoid maps);* 2. 2.
* is a comonoid, and and are comonoid maps (hence are induced by comonoid maps and );* 3. 3.
* and (the comultiplication of ) are maps over :*
[TABLE]
[TABLE] 4. 4.
* is a comonoid, and and are comonoid maps.*
Proof.
Let and be internal functors making comonoidal. Then and are both counital comonoidal structures on such that is a comonoid map with respect to . By the Eckmann-Hilton argument, we have and is cocommutative.
The maps make a comonoid, and is a map over by definition of an internal functor. By Lemma 2.11, both and are comonoid maps, and by Lemma 2.15, they are comodule maps over .
Since and are comodule maps over and , Lemma 2.10 shows that is a comonoid. Finally, since and are internal functors, and are required to make the following diagrams commute:
[TABLE]
[TABLE]
But these are precisely the diagrams that make and comonoid maps. ∎
Remark 3.10**.**
The previous proposition effectively says that a comonoidal internal category is a ‘category internal to ’. We write the latter statement in quotes because our definition of internal category requires the ambient monoidal category to be regular, which need not be.
Corollary 3.11**.**
The following diagram commutes:
[TABLE]
Proof.
Follows from being a map over . ∎
Thus, although the comultiplicands of need not be the same (i.e. is not cocommutative), their sources are. The analogous statement for targets also holds.
Example 3.12**.**
If is a bimonoid, its one-object category is comonoidal.
If is a cocommutative comonoid, its discrete category is comonoidal.
4 Internal Prestacks
Definition 4.1**.**
Let be a comonoidal internal category. A right -comodule category is a right -comodule in .
In detail, this is the data of an internal category along with:
a -coaction that is also a comonoid map (hence is induced by a comonoid map ); 2. 2.
a -coaction that is also a map over ; 3. 3.
such that is an internal functor.
We henceforth refer to these as simply -comodule categories or -comodules.
Recall that if is comonoidal, then so is the discrete category .
Lemma 4.2**.**
Let be a comonoidal internal category and its subcategory of objects. Let be a -comodule category with coaction
[TABLE]
and let be the comonoid map that induces . Then:
* is a comonoid, with comultiplication ;* 2. 2.
The -coactions and on coincide with ; 3. 3.
* and are maps over :*
[TABLE] 4. 4.
The coactions and induce -coactions on ;
Proof.
By Lemma 2.11, since is a comonoid map, the comultiplication is a comodule map over , and the counit is a comodule map over . By Lemma 2.10, is a comonoid.
By Lemma 2.12, induces a comonoid map . Corestricting along makes a -bicomodule. Since is a map over , the left square in the following diagram commutes:
[TABLE]
The outer diagram then says that and coincide. A similar diagram (with an identity instead of ) shows that and coincide.
Again, the following diagram commutes, so is a map over :
[TABLE]
Similarly, is a map over . We may thus form the composites,
[TABLE]
which are seen to be coactions. ∎
Definition 4.3**.**
Let be a comonoidal internal category and its subcategory of objects. A prestack over (or a -module category) consists of:
An internal category with cocommutative; 2. 1.
A coaction ; 3. 2.
A comonoid map satisfying:
[TABLE] 4. 3.
A map satisfying:
[TABLE]
[TABLE] 5. 4.
and further satisfy:
[TABLE]
The map is given by the following lemma:
Lemma 4.4**.**
There is an action .
Proof.
We first observe that we have a map induced by:
[TABLE]
Next, since the left and right -coactions on coincide, the following diagram commutes,
[TABLE]
so the map {\iota\colon A\underset{C}{\diamond}A}$${A\underset{D}{\diamond}A}$${A\otimes A} is a comodule map over . The comultiplication is also a comodule map over , which we may combine with the above map to obtain a map :
[TABLE]
Finally, a routine diagram chase, repeatedly invoking the naturality of , allows us to verify that this map does indeed factor through , giving the desired map. ∎
Remark 4.5**.**
The prestacks we have defined should techincally be called split prestacks. However, as these are the only prestacks we consider in this paper, we omit the word ‘split’.
5 Smash products
Let be a prestack over , with actions and as above.
We make an object of , with left coaction induced by the comonoid map , and right coaction induced by the comonoid map ,
[TABLE]
where is the comultiplication of . We also have a right -coaction induced by the comonoid map :
[TABLE]
Lemma 5.1**.**
Let be an internal prestack over . Then:
The coaction is a bicomodule map over :
[TABLE] 2. 2.
The coaction is a bicomodule map over :
[TABLE] 3. 3.
The coaction is a comodule map over :
[TABLE]
Proof.
By Lemma 2.15, the top squares of the following diagrams commute:
[TABLE]
The remaining squares obviously commute. For the left square, since is cocommutative, the left vertical composite is . The right vertical composite is because by Lemma 4.2. This proves the first item.
For the second item, the left square commutes because
[TABLE]
The right square of the second item and the square in the third item commute by similar arguments. ∎
We are now in a position to define smash products of internal prestack.
Theorem 5.2**.**
Let be an internal prestack. There is an internal category
[TABLE]
which we call the smash product of with . Further, has the structure of a -comodule category.
Proof.
By Lemma 5.1, and are all maps over , allowing us to define the composite in Figure 1.
In fact, this composite factors through , giving the multiplication on .
The unit of is given by the composite:
[TABLE]
These maps are unital and associative (because of Items 3 and 4 in Definition 4.3, and the fact that and are internal categories), so is an internal category.
To see that has the structure of a -comodule category, note that already has a -coaction . We then take the -coaction on to be the composite in Figure 2.
∎
6 Coinvariants of comodule categories
Although we have not defined what a ‘cartesian fibered right -comodule category’ should be, we can still verify that the fibers of allow us to recover our original prestack . We first begin by defining the fibers of any right -comodule category.
Definition 6.1**.**
Let be a right -comodule category, with coaction functor . The coinvariant category is the coinduction along :
[TABLE]
Equivalently, is given by the equalizer:
[TABLE]
The following lemmas follow almost by definition:
Lemma 6.2**.**
The coinvariant category is a -comodule category.
Lemma 6.3**.**
The coinvariant category is given by .
Given an arbitrary -comodule category, it is unlikely that its coinvariant category has the structure of a prestack over . However, when the -comodule category is of the form for a prestack , we have:
Theorem 6.4**.**
Let be a prestack over and let be the corresponding right -comodule category. Then the coinvariant category is a prestack over , which is moreover isomorphic to .
Proof.
First observe that the comonoid of objects for and are all . On morphisms, recall that the -coaction on is given by the copy of sitting inside . Thus
[TABLE]
So and are isomorphic categories. We may then transfer the prestack structure of over to . ∎
7 Further work
In this paper, we have seen that the smash product (a.k.a. the Grothendieck construction) for split prestacks generalizes well to the non-cartesian internal setting, as long as one is willing to relax the definition of what it means to be a prestack.
Several assumptions were made that reduce the scope of the results in this paper. Firstly, although is not cartesian, it is assumed to be symmetric monoidal rather than merely braided monoidal. The symmetry assumption yielded certain convenient but not crucial lemmas (at least in the author’s opinion). It is thus believable that smash products as defined in this paper should still exist in the braided monoidal setting.
We have also assumed that the comonoids of objects of both the base and the prestack are cocommutative. This assumption seems to be more crucial, and not something that can be easily done away with. Indeed, Lemma 5.1 – the main technical result – holds only because we assumed that and were cocommutative.
It remains to be seen if a similar construction can be carried out for internal categories with non-cocommutative comonoids of objects, e.g. the quantum categories of [chikhladze2011category]. We note that smash products have been defined for weak bialgebras [nikshych2000duality], and that these are bimonoids in an appropriate duoidal category, so a possible next step would be to define smash products for prestacks internal to a duoidal category.
References
