This paper explores modules over categorified fiber bundles using entwining structures involving linear categories and coalgebras, establishing Frobenius and separability conditions, and introducing Galois extensions of categories.
Contribution
It introduces the concept of C-Galois extensions of categories and characterizes entwined modules over these extensions as modules over subcategories of coinvariants.
Findings
01
Frobenius and separability conditions for functors on entwined modules
02
Definition of C-Galois extensions of categories
03
Equivalence of entwined modules and modules over coinvariant subcategories
Abstract
In this paper, we study modules over quotient spaces of certain categorified fiber bundles. These are understood as modules over entwining structures involving a small K-linear category D and a K-coalgebra C. We obtain Frobenius and separability conditions for functors on entwined modules. We also introduce the notion of a C-Galois extension E⊆D of categories. Under suitable conditions, we show that entwined modules over a C-Galois extension may be described as modules over the subcategory E of C-coinvariants of D.
\begin{array}[]{ccc}\mbox{$R\longrightarrow S$ is split extension}&\qquad\Leftrightarrow&\mbox{Left adjoint $F:Mod\text{-}R\longrightarrow Mod\text{-}S$ is separable}\\
&&\\
\mbox{$R\longrightarrow S$ is separable extension}&\qquad\Leftrightarrow&\mbox{Right adjoint $G:Mod\text{-}S\longrightarrow Mod\text{-}R$ is separable}\\
&&\\
\mbox{$R\longrightarrow S$ is Frobenius extension}&\qquad\Leftrightarrow&\mbox{$(F,G)$ is Frobenius pair of functors}\\
\end{array}
\begin{array}[]{ccc}\mbox{$R\longrightarrow S$ is split extension}&\qquad\Leftrightarrow&\mbox{Left adjoint $F:Mod\text{-}R\longrightarrow Mod\text{-}S$ is separable}\\
&&\\
\mbox{$R\longrightarrow S$ is separable extension}&\qquad\Leftrightarrow&\mbox{Right adjoint $G:Mod\text{-}S\longrightarrow Mod\text{-}R$ is separable}\\
&&\\
\mbox{$R\longrightarrow S$ is Frobenius extension}&\qquad\Leftrightarrow&\mbox{$(F,G)$ is Frobenius pair of functors}\\
\end{array}
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.
Full text
Entwined modules over linear categories and Galois extensions
Department of Mathematics, Indian Institute of Science, Bangalore - 560012, India.
Abstract
In this paper, we study modules over quotient spaces of certain categorified fiber bundles. These are understood as modules over entwining structures involving a small K-linear category
D and a K-coalgebra C.
We obtain Frobenius and separability conditions for functors on entwined modules. We also introduce the notion of a C-Galois extension E⊆D of categories. Under suitable conditions, we show that entwined modules over a C-Galois extension may be described as modules over the subcategory E of
C-coinvariants of D.
MSC(2010) Subject Classification: 16W30, 18E05
Keywords: Entwining structures, entwined modules, rings with several objects, Frobenius conditions, separability conditions, coalgebra-Galois extensions
1 Introduction
The purpose of this paper is to study a theory of modules over quotient spaces of certain categorified fiber bundles. Suppose that X is an affine scheme over a field K and let G be an affine algebraic group scheme with a free action σ:X×G⟶X on X. Let Y be the quotient given by the coequalizer
[TABLE]
If X⟶Y is faithfully flat and the canonical map can:X×G⟶X×YX is an isomorphism, then X is said to be (see, for instance, [Mum], [HJS])
a principal fiber bundle over Y with group G.
The algebraic counterpart of (1.1) consists of an algebra A, a Hopf algebra H and a coaction ρ:A⟶A⊗H that makes A into a right H-comodule algebra. Let B:=A^{coH}=\{\mbox{a\in A|\rho(a)=a\otimes 1_{H}}\} be the algebra of coinvariants of A, i.e., B is given by the equalizer
[TABLE]
In this case, there is a canonical map can:A⊗BA⟶A⊗H determined by setting can(x⊗y)=x⋅ρ(y). If the Hopf algebra H
has bijective antipode, B⟶A is a faithfully flat extension and can:A⊗BA⟶A⊗H is an isomorphism, it was shown by Schneider [HJS] that
modules over B may be recovered as the category of “(A,H)-Hopf modules.”
We work with a small K-linear category D, a K-coalgebra C and an “entwining structure” ψ consisting of a collection of morphisms
[TABLE]
satisfying conditions that we lay out in Section 2. We consider the category M(ψ)DC of modules over the entwining structure
(D,C,ψ) (see Definition 2.2). These may be seen as modules over a “categorical quotient space” of D with respect to the coalgebra C and the entwining
ψ.
The notion of a C-Galois extension E⊆D of categories is introduced in Section 4. Additionally, a C-Galois
extension gives rise to a canonical entwining structure on D. Under certain conditions, we show that modules over the category E of C-coinvariants of D may be described as modules
over the canonical entwining structure.
Entwining structures for algebras were introduced by Brzeziński and Majid in [BrMj] and it was realized in Brzeziński [Brz00] that
entwined modules provide a unifying formalism for studying diverse concepts such as relative Hopf modules, Doi-Hopf and Yetter-Drinfeld modules as well as coalgebra Galois extensions. In fact, the study of entwining structures for algebras and entwined modules over them is well developed in the literature and we refer the reader, for instance, to [Abu], [Brz00] [Br02], [BCT1], [BCT], [CG], [Jia], [Sch] for more on this subject.
Our notion of modules over an entwining structure (D,C,ψ) builds on the analogy of Mitchell [Mit1] which says that a small K-linear category
should be seen as a “K-algebra with several objects.” In particular, the category M(ψ)DC also generalizes the “relative (D,H)-Hopf modules” studied
in our previous work in [BBR], where H is a Hopf algebra and D is an H-comodule category in the sense of Cibils and Solotar [CiSo]. In other words,
D is a small K-linear category whose morphism spaces are equipped with a coaction of H that is compatible with composition. When D has a single object, it reduces
to an ordinary H-comodule algebra and the relative (D,H)-Hopf modules
reduce to the usual notion of relative Hopf modules (see Takeuchi [Take]).
For Doi-Hopf modules, Frobenius and separability conditions were studied extensively in a series of papers [CMIZ], [CMZ1], [CMZ]. Later,
Brzeziński studied Frobenius and Maschke type theorems for entwined modules in [Brz]. In this paper, we proceed in a manner analogous
to the unified approach of Brzeziński, Caenepeel, Militaru and Zhu [uni] for studying Frobenius and separability conditions for entwined modules over (D,C,ψ).
The idea is as follows: the “categorical quotient space” of D with respect to C and ψ may be thought of as a subcategory of D and M(ψ)DC plays the role of modules over this subcategory. Although this “subcategory” of D need not exist in an explicit sense, we would like to study the properties of this extension
of categories. In particular, we would like to know if it behaves like a separable, split or Frobenius extension of small K-linear categories. For this, we turn to a pair of functors
[TABLE]
Here F is the left adjoint and behaves like an “extension of scalars” whereas its right adjoint G behaves like a “restriction of scalars.” We recall here (see [uni, Theorem 1.2]) that in the classical case of an extension R⟶S of rings inducing the pair of adjoint functors Mod\text{-}R\ext@arrow 3399\arrowfill@\mathrel{\raise 3.22916pt\hbox{\oalign{\scriptstyle\leftarrow\cr\vrulewidth=0.0pt,height=2.15277pt\hfil\scriptstyle\relbar\cr}}}\mathrel{\raise 3.22916pt\hbox{\oalign{\scriptstyle\relbar\cr\vrulewidth=0.0pt,height=2.15277pt\scriptstyle\relbar}}}\mathrel{\raise 3.22916pt\hbox{\oalign{\scriptstyle\relbar\hfil\cr\scriptstyle\vrule width=0.0pt,height=1.50694pt\smash{\rightarrow}\cr}}}{\text{\qquad F\qquad}}{\text{\qquad G\qquad}}Mod\text{-}S given
by extension and restriction of scalars, we have:
[TABLE]
It is therefore natural to study criteria for the separability of the functors F and G as well as conditions for (F,G)
to be a Frobenius pair of functors.
In this paper, we will always use the following convention: for f∈HomD(Y,X) and c∈C, we write ψYX(c⊗f)=fψ⊗cψ∈HomD(Y,X)⊗C with the summation omitted. We write h:Dop⊗D⟶VectK for the canonical D-D-bimodule h(Y,X)=HomD(Y,X). The entwining structure makes h⊗C into a D-D-bimodule by setting
[TABLE]
for any (Y,X)∈Ob(Dop⊗D), \phi:=(\phi^{\prime},\phi^{\prime\prime})\in Hom_{\mathcal{D}^{op}\otimes\mathcal{D}}\big{(}(Y,X),(Y^{\prime},X^{\prime})\big{)}, f∈HomD(Y,X) and c∈C. We consider a collection θ:={θX:C⊗C⟶EndD(X)}X∈Ob(D) of K-linear maps satisfying the following conditions:
[TABLE]
for any f∈HomD(Y,X). Let V1 be the K-space consisting of all such θ.
Our first result gives conditions for the functors F and G to be separable.
Theorem A**.**
(see 3.7, 3.8, 3.10 and 3.11) Let D be a small K-linear category, (C,ΔC,εC) be a K-coalgebra and let (D,C,ψ) be a right-right entwining
structure.
(a)
Let V=Nat(GF,1M(ψ)DC) be the space of natural transformations from GF to 1M(ψ)DC. Then:
(1)
There is an isomorphism V≅V1 of K-vector spaces.
(2)
The functor F is separable if and only if there exists θ∈V1 such that
[TABLE]
(b)
Let W=Nat(1Mod-D,FG) be the space of natural transformations from 1Mod-D to FG. Then:
(1)
There is an isomorphism of K-vector spaces from W to W1=Nat(h,h⊗C).
(2)
The functor G is separable if and only if there exists η∈W1=Nat(h,h⊗C) such that
(idh⊗εC)η=idh.
The next result gives conditions for (F,G) to be a Frobenius pair.
Theorem B**.**
(see 3.14) Let D be a small K-linear category, (C,ΔC,εC) be a K-coalgebra and let (D,C,ψ) be a right-right entwining
structure.
Then, (F,G) is a Frobenius pair if and only if there exist θ∈V1 and η∈W1 such that the following conditions hold:
[TABLE]
for any f∈HomD(X,Y), d∈C and η(X,Y)(f)=∑f^⊗cf.
More generally, the D-D-bimodule h⊗C may be treated as a functor h⊗C:D⟶M(ψ)DC by setting (see Lemma 2.4)
[TABLE]
for f∈HomD(Y,X) and g⊗c∈HomD(Z,Y)⊗C. Additionally, let C be a finite dimensional coalgebra and let C∗=Hom(C,K) be the linear dual of C. Then, we show that there is a functor
C∗⊗h:D⟶M(ψ)DC.
Theorem C**.**
(see 3.19)
Let (D,C,ψ) be an entwining structure and let C be a finite dimensional coalgebra. Then, the following statements are equivalent:
(i) (F,G) is a Frobenius pair.
(ii) C∗⊗h and h⊗C are isomorphic as functors from D to M(ψ)DC.
In the final part of this paper, we study coalgebra Galois extensions of categories in a manner analogous to Brzeziński [Brz00], Brzeziński and Hajac [BH] and Caenepeel [Caeni]. For this, we suppose that
every morphism space HomD(X,Y) carries the structure of a C-comodule ρXY:HomD(X,Y)⟶HomD(X,Y)⊗C, f↦∑f0⊗f1. This allows us to define a category E of C-coinvariants of D (see
Definition 4.5). Further, we say that D is a C-Galois extension of E if the canonical map
[TABLE]
is an isomorphism for each X∈Ob(D) (see Definition 4.7). We show that a C-Galois extension leads to a canonical entwining structure.
Theorem D**.**
(see 4.9)
Let D be a C-Galois extension of E.
Then,
there exists a unique right-right entwining structure (D,C,ψ) which makes HomD(−,Y) an object in M(ψ)DC for every Y∈Ob(D) with its canonical D-module structure and right C-coactions {ρXY}X∈Ob(D).
Conversely, under suitable conditions, an entwining structure (D,C,ψ) may be used to express D as a C-Galois extension. In that case, the category
M(ψ)DC reduces to the category of modules over the C-coinvariants of D.
Theorem E**.**
(see 4.12 and 4.21)
Let C be a K-coalgebra and D be a small K-linear category such that HomD(X,Y) has a right C-comodule structure ρXY for every X,Y∈Ob(D). Let E be the subcategory of C-coinvariants of D. If there exists a convolution invertible collection Φ={ΦXY:C⟶HomD(X,Y)}X,Y∈Ob(D) of right
C-comodule maps, then the following are equivalent:
(i) D is a C-Galois extension of E.
(ii) There exists a right-right entwining structure (D,C,ψ) such that HomD(−,Y) is an object in M(ψ)DC for every Y∈Ob(D) with its canonical D-module structure and right C-coactions {ρXY}X∈Ob(D).
(iii) For any f∈HomD(X,Y), the morphism ∑f0∘ΦZX′(f1)∈HomE(Z,Y) for every Z∈Ob(D), where Φ′ is the convolution inverse of Φ.
In this case, the categories M(ψ)DC and Mod-E are equivalent.
Notations: Throughout the paper, K is a field, C is a K-coalgebra with comultiplication ΔC and counit εC. We shall use Sweedler’s notation for the coproduct ΔC(c)=c1⊗c2, and for a coaction ρM:M⟶M⊗C, ρM(m)=m0⊗m1 with the summation omitted. We denote by C∗ the linear dual of C. Sometimes when the coaction is clear from context, we will omit the subscript.
2 Entwining structures
In this section, we introduce a categorical generalization of entwining structures and entwined modules. We prove that the category of entwined modules is a Grothendieck category.
We begin by recalling the definition of modules over a category (see, for instance, [Sten, Mit2]).
Definition 2.1**.**
A right module over a small K-linear category D is a K-linear functor Dop⟶VectK, where VectK denotes the category of K-vector spaces. Similarly, a left module over D is a K-linear functor D⟶VectK. The category of all right (resp. left) modules over D will be denoted by Mod-D (resp. D-Mod).
For each X∈Ob(D), the representable functors hX:=HomD(−,X) and Xh:=HomD(X,−) are examples of right and left modules over D respectively. Unless otherwise mentioned,
by a D-module we will always mean a right D-module.
Let C be a K-coalgebra and let D be a small K-linear category. Suppose that we have a collection of K-linear maps
[TABLE]
We use the notation ψXY(c⊗f)=fψ⊗cψ for c∈C and f∈HomD(X,Y). We will say that the tuple (D,C,ψ) is a (right-right) entwining structure if the following conditions hold:
[TABLE]
for each f∈HomD(X,Y), g∈HomD(Y,Z) and c∈C.
Throughout this paper, (D,C,ψ) will always be an entwining structure. A morphism between entwining structures (D′,C′,ψ′) and (D,C,ψ) is a pair (F,σ) where F:D′⟶D is a functor and σ:C′⟶C is a counital coalgebra map such that F(f′ψ′)⊗σ(c′ψ′)=F(f′)ψ⊗σ(c′)ψ for any c′⊗f′∈C′⊗HomD′(X′,Y′) where X′,Y′∈Ob(D′).
Definition 2.2**.**
Let M be a right D-module with a given right C-comodule structure ρM(Y):M(Y)⟶M(Y)⊗C on M(Y) for each Y∈Ob(D). Then, M is said to be an entwined module over (D,C,ψ) if the following compatibility condition holds:
[TABLE]
for every f∈HomD(Y,X) and m∈M(X). We denote by M(ψ)DC the category whose objects are entwined modules over (D,C,ψ) and whose morphisms are given by
[TABLE]
We now give an important example of entwining structures.
Example 2.3**.**
Let D be a right co-H-category (see [CiSo] or the description in [BBR, Definition 2.4] ) and C be a right H-module coalgebra. Then, the triple (D,C,ψ) is an entwining structure, where ψ is given by:
[TABLE]
Explicitly, we have ψXY(c⊗f):=f0⊗cf1 for any f∈HomD(X,Y) and c∈C. In this case, an entwined module is precisely a right D-module with a given right C-comodule structure on M(X) for each X∈Ob(D) and satisfying the following compatibility condition
[TABLE]
We will refer to these modules as (right-right) Doi-Hopf modules and their category will be denoted by MDC. If D is a right co-H-category with a single object, i.e., an H-comodule algebra, then MDC recovers the classical notion of Doi-Hopf modules (see [Doi]). In the particular case where C=H, the right-right
Doi-Hopf modules have been referred to as relative Hopf modules in [BBR, § 5].
Lemma 2.4**.**
Let (D,C,ψ) be an entwining structure and let M be a right D-module. Then, we may obtain an object M⊗C∈M(ψ)DC by setting
[TABLE]
for X∈Ob(D),f∈HomD(Y,X) and m⊗c∈M(X)⊗C. In fact, this determines a functor
from Mod-D to M(ψ)DC .
Proof.
The fact that M⊗C is a right D-module follows from (2.1). For each X∈Ob(D), it may be verified that M(X)⊗C has a right C-comodule structure given by
[TABLE]
It remains to check the compatibility condition in (2.5). By definition, we have
[TABLE]
∎
Lemma 2.5**.**
Let (D,C,ψ) be an entwining structure and N be a right C-comodule. Then, for each X∈Ob(D) we may obtain an object N⊗hX∈M(ψ)DC by setting
[TABLE]
for Y∈Ob(D), f∈HomD(Z,Y), n⊗g∈N⊗hX(Y). In fact, this determines a functor
from Comod-C to M(ψ)DC .
Proof.
By definition, it follows that N⊗hX is a right D-module. Further, for each Y∈Ob(D), we define a K-linear map σN⊗hX(Y)r:N⊗hX(Y)⟶N⊗hX(Y)⊗C as follows
[TABLE]
We now verify that the map defined in (2.9) makes N⊗hX(Y) a right C-comodule. We have
It remains to verify the condition in (2.5). We have
[TABLE]
∎
It follows from Lemma 2.4 and Lemma 2.5 that both hY⊗C and C⊗hY are objects in M(ψ)DC for every Y∈Ob(D).
Lemma 2.6**.**
Let (D,C,ψ) be an entwining structure. Then, for each Y∈Ob(D), we get a morphism ΨY:C⊗hY⟶hY⊗C in M(ψ)DC given by ΨY(X):=ψXY.
Proof.
First we verify that ΨY is a morphism of right D-modules. For any f∈HomD(X′,X), g∈HomD(X,Y) and c∈C, we have
[TABLE]
Next, we will show that ΨY(X) is C-colinear for every X∈Ob(D). We have
[TABLE]
∎
We now recall from [Mit1, § 3] and [Mit2] the notion of a finitely generated module over a category. Given M∈Mod-D, we set el(M):=X∈Ob(D)∐M(X) to be the collection of all elements of M. Since D is small, we note that
el(M) is a set. If m∈el(M) is such that m∈M(X), we will write
∣m∣=X.
Definition 2.7**.**
Let D be a small preadditive category and let M be a right D-module. For each
m∈el(M), we consider the corresponding morphism ηm:h∣m∣⟶M. A family of elements {mi∈el(M)}i∈I is said to be a generating set for M if the induced morphism
[TABLE]
is an epimorphism in Mod-D. In other words, every element m∈el(M) may be expressed as a sum m=∑i∈IM(fi)(mi), where each fi∈HomD(∣m∣,∣mi∣) and
all but finitely many {fi}i∈I are zero.
Lemma 2.8**.**
Let (D,C,ψ) be an entwining structure and let M be an entwined module. We consider an element
m∈el(M). Then, there exists a finite dimensional C-subcomodule Vm of M(∣m∣) containing m and a morphism ηm:Vm⊗h∣m∣⟶M in M(ψ)DC such that ηm(∣m∣)(m⊗id∣m∣)=m.
Proof.
By [SCS, Theorem 2.1.7], we know that there exists a finite dimensional C-subcomodule Vm⊆M(∣m∣) such that m∈Vm.
Now, we consider the D-module morphism ηm:Vm⊗h∣m∣⟶M defined by setting ηm(Y)(v⊗f):=M(f)(v) for any Y∈Ob(D),f∈HomD(Y,∣m∣) and v∈Vm. We also have
[TABLE]
This shows that ηm(Y) is C-colinear for each Y∈Ob(D). Hence, ηm is a morphism in M(ψ)DC such that ηm(∣m∣)(m⊗id∣m∣)=m.
∎
Proposition 2.9**.**
Let (D,C,ψ) be an entwining structure. Then, the category M(ψ)DC of entwined modules is a Grothendieck category.
Proof.
Given a morphism η:M⟶N in M(ψ)DC, let Ker(η) and Coker(η) be respectively the kernel and cokernel in Mod-D. Since Comod-C
is an abelian category, we know that Ker(η)(X), Coker(η)(X)∈Comod-C for each X∈Ob(D). It is easily
seen that Ker(η) and Coker(η) satisfy the compatibility condition in (2.5), i.e., Ker(η), Coker(η)∈M(ψ)DC. Since limits and colimits in M(ψ)DC are obtained from
those in Mod-D and Comod-C, it is clear that M(ψ)DC is a cocomplete abelian category satisfying (AB5).
for any M∈M(ψ)DC. As such, the collection {V⊗hX}, where X ranges over all objects in D and V ranges over all (isomorphism classes of) finite dimensional C-comodules gives a set
of generators for M(ψ)DC in the sense of [Grothen, Proposition 1.9.1].
∎
Corollary 2.10**.**
The category MDC of Doi-Hopf modules is a Grothendieck category.
3 Separability and Frobenius conditions
Let F:M(ψ)DC⟶Mod-D be the forgetful functor. The next result shows that the functor F has a right adjoint.
Lemma 3.1**.**
The forgetful functor F:M(ψ)DC⟶Mod-D has a right adjoint G:Mod-D⟶M(ψ)DC given by G(N):=N⊗C for each N∈Mod-D.
Proof.
From Lemma 2.4, we know that G(N)=N⊗C∈M(ψ)DC
for each N∈Mod-D. We define
\alpha:Hom_{{\mathscr{M}(\psi)}_{\mathcal{D}}^{C}}\big{(}\mathcal{M},\mathscr{G}(\mathcal{N})\big{)}\longrightarrow Hom_{Mod\text{-}\mathcal{D}}(\mathscr{F}(\mathcal{M}),\mathcal{N}) by setting
[TABLE]
for each ξ:M⟶N⊗C in M(ψ)DC, X∈Ob(D) and m∈M(X).
We also define
\beta:Hom_{Mod\text{-}\mathcal{D}}(\mathscr{F}(\mathcal{M}),\mathcal{N})\longrightarrow Hom_{{\mathscr{M}(\psi)}_{\mathcal{D}}^{C}}\big{(}\mathcal{M},\mathscr{G}(\mathcal{N})\big{)}
by setting
[TABLE]
for each η:M⟶N in Mod-D,
X∈Ob(D) and m∈M(X). First we check that α(ξ) and β(η) are morphisms in Mod-D and M(ψ)DC respectively. Using the fact that idN⊗εC:N⊗C⟶N and ξ are right D-module morphisms, for any f∈HomD(Y,X), we have
[TABLE]
We also have
[TABLE]
Moreover, it is easy to see that β(η)(X) is C-colinear for each X∈Ob(D). We now verify that α and β are inverses to each other.
[TABLE]
Further, we have \alpha\big{(}\beta(\eta)\big{)}(X)(m)=\eta(X)(m_{0})\varepsilon_{C}(m_{1})=\eta(X)(m). This
proves the result.
∎
We now describe the unit μ:1M(ψ)DC⟶GF and the counit ν:FG⟶1Mod-D of the adjunction in Lemma 3.1:
[TABLE]
[TABLE]
for each M∈M(ψ)DC, N∈Mod-D, X∈Ob(D).
We recall that a functor F:A⟶B between arbitrary categories is said to be separable if the natural transformation
[TABLE]
induced by F is a split monomorphism (see [NVV], [Rf, § 1]). The following result provides a characterization of separable functors.
Theorem 3.2**.**
[Rf, Theorem 1.2]*
Let F:A⟶B be a functor which has a right adjoint G:B⟶A. Let μ and ν be the unit and counit of this adjunction respectively. Then,*
(i)
F* is separable if and only if there exists υ∈Nat(GF,1A) such that υ∘μ=1A, the identity natural transformation on A.*
(ii)
G* is separable if and only if there exists ζ∈Nat(1B,FG) such that ν∘ζ=1B, the identity natural transformation on B.*
3.1 Separability conditions
Let (D,C,ψ) be an entwining structure.
We now investigate the separability of the forgetful functor F:M(ψ)DC⟶Mod-D. Since F has a right adjoint G, it follows from Theorem 3.2 that the functor F is separable if and only if there exists a natural transformation υ:GF⟶1M(ψ)DC such that υ∘μ=1M(ψ)DC, where μ is the unit of the adjunction as explained in (3.1). Throughout Section 3, V:=Nat(GF,1M(ψ)DC) will denote the K-space of all natural transformations from GF to 1M(ψ)DC. We will shortly give another useful interpretation of V. We start by proving few preparatory results required for this.
We recall from Lemma 2.4 and Lemma 2.5 that both hY⊗C and C⊗hY are objects in M(ψ)DC for every Y∈Ob(D). We define a functor h⊗C:D⟶M(ψ)DC as
[TABLE]
for f∈HomD(Y,X),g∈hY(Z) and c∈C. Similarly, we may also obtain a functor h⊗C⊗C:D⟶M(ψ)DC.
Lemma 3.3**.**
Let f∈HomD(Y,X). For any υ∈V and c,d∈C, we have
[TABLE]
In particular, we have
[TABLE]
Proof.
A morphism f:Y⟶X in D induces morphisms hY⊗C⟶hX⊗C and hY⊗C⊗C⟶hX⊗C⊗C in M(ψ)DC as explained in (3.4). Since υ:GF⟶1M(ψ)DC is a natural transformation, it follows that the following diagram commutes:
[TABLE]
Thus, we have
[TABLE]
We now consider the morphism ΨY:C⊗hY⟶hY⊗C in M(ψ)DC given by ΨY(X):=ψXY as in Lemma 2.6. Then, using the naturality of υ:GF⟶1M(ψ)DC and (2.2) we have the following commutative diagram
[TABLE]
Using the fact that ψYY(c⊗idY)=idY⊗c, we now have
By putting Y=X and taking f=idX, the result of (3.6) is clear from (3.9).
∎
Lemma 3.4**.**
For any υ∈V and Y∈Ob(D), we have υ(C⊗C⊗hY)=idC⊗υ(C⊗hY) as a morphism
of D-modules.
Proof.
For each d∈C, we define ηd:C⊗hY⟶C⊗C⊗hY by
[TABLE]
for each X∈Ob(D), g∈hY(X) and c∈C. It may be easily verified that ηd is a morphism of right D-modules. We now verify that ηd(X):C⊗hY(X)⟶C⊗C⊗hY(X) is right C-colinear. We have
[TABLE]
Thus, ηd:C⊗hY⟶C⊗C⊗hY is a morphism in M(ψ)DC. Therefore, using the naturality of υ, we have the following commutative diagram:
[TABLE]
Thus, for any g∈HomD(X,Y) and c,c′∈C, we get
[TABLE]
The result follows.
∎
We now proceed to give another interpretation of V=Nat(GF,1M(ψ)DC). We consider a collection θ:={θX:C⊗C⟶EndD(X)}X∈Ob(D) of K-linear maps satisfying the following conditions:
[TABLE]
for any f∈HomD(Y,X). Let V1 be the K-space consisting of all such θ.
Proposition 3.5**.**
Let υ∈V=Nat(GF,1M(ψ)DC). For each X∈Ob(D), we define a K-linear map
[TABLE]
*Then, θ:={θX}X∈Ob(D) is an element in V1.
*
Proof.
Since idhX⊗εC:hX⊗C⟶hX is a morphism of right D-modules, we have
[TABLE]
for f∈HomD(Y,X) and c,d∈C. Since υ(hX⊗C):hX⊗C⊗C⟶hX⊗C is a morphism of right D-modules, we also have
[TABLE]
The morphism fψψ:Y⟶X in D induces morphisms hY⊗C⟶hX⊗C and hY⊗C⊗C⟶hX⊗C⊗C in M(ψ)DC. Therefore, we have
[TABLE]
This proves (3.11). We now verify that θ satisfies (3.12). Using Lemma 2.5, we know that C⊗hY and C⊗C⊗hY belong to M(ψ)DC for each Y∈Ob(D). For each X∈Ob(D), it may be easily seen that C⊗hY(X) is also a left C-comodule with coaction given by ρYl(X):=ΔC⊗idhY(X). Moreover, it may be easily verified that the following diagram commutes:
[TABLE]
This shows that ρYl(X) is a morphism of right C-comodules. Further, for any g∈HomD(X,X′), we have the following commutative diagram:
[TABLE]
Thus, ρYl:C⊗hY⟶C⊗C⊗hY is a morphism of right D-modules. This shows that ρYl is a morphism in the category M(ψ)DC. Therefore, using the naturality of υ and Lemma 3.4, we have the following commutative diagram:
[TABLE]
For any c⊗idX⊗d∈C⊗hX(X)⊗C, we set a_{i}\otimes f_{i}:=\upsilon\big{(}C\otimes{\bf h}_{X}\big{)}(X)(c\otimes id_{X}\otimes d). Then, we have
[TABLE]
Now applying the map idC⊗εC⊗idhX to both sides, we get
[TABLE]
Therefore, we have
[TABLE]
Since υ(C⊗hY)(X) is a morphism of right C-comodules, we also have the following commutative diagram:
[TABLE]
Thus, we have
[TABLE]
Now, applying the map εC⊗idhX⊗idC to both sides, we get
[TABLE]
Therefore,
[TABLE]
It now follows from (3.15) and (3.16) that θ satisfies (3.12).
∎
Proposition 3.6**.**
Let θ∈V1. Then, we have an element υ∈Nat(GF,1M(ψ)DC) defined by
[TABLE]
for \mathcal{M}\in Ob\big{(}{\mathscr{M}(\psi)}_{\mathcal{D}}^{C}\big{)},~{}X\in Ob(\mathcal{D}),~{}m\in\mathcal{M}(X) and c∈C.
Proof.
We need to verify that υ(M):M⊗C⟶M is a morphism in M(ψ)DC and that υ is indeed a natural transformation. We first verify that υ(M) is a morphism of right D-modules. Let f∈HomD(Y,X). Then, we have
[TABLE]
We now verify that υ(M)(X):M(X)⊗C⟶M(X) is a morphism of right C-comodules for every X∈Ob(D). For each m⊗c∈M(X)⊗C, we have
[TABLE]
It remains to show that υ:GF⟶1M(ψ)DC is a natural transformation. Let η:M⟶N be a morphism in M(ψ)DC. Then, for every X∈Ob(D) and m⊗c∈M(X)⊗C, we have
[TABLE]
This proves the result.
∎
Proposition 3.7**.**
The K-spaces V=Nat(GF,1M(ψ)DC) and V1 are isomorphic.
Proof.
We define α:V⟶V1 by setting α(υ)=θ, where θ is the collection of K-linear maps {θX:C⊗C⟶EndD(X)}X∈Ob(D) defined by
[TABLE]
for c,d∈C. Then, α is a well-defined map by Proposition 3.5. We also define
β:V1⟶V by setting β(θ)=υ, where υ:GF⟶1M(ψ)DC is defined by
[TABLE]
for \mathcal{M}\in Ob\big{(}{\mathscr{M}(\psi)}_{\mathcal{D}}^{C}\big{)},~{}X\in Ob(\mathcal{D}),~{}m\otimes c\in\mathcal{M}(X)\otimes C. By Proposition 3.6, β is well-defined. We will now verify that α and β are inverses of each other. Let θ∈V1. Then, for any X,Y∈Ob(D), we have
[TABLE]
This proves that \big{(}\alpha\beta(\theta)\big{)}_{X}=\theta_{X} for all X∈Ob(D). Therefore,
(αβ)(θ)=θ. For any υ∈V, we now verify that (βα)(υ)=υ. We set θ=α(υ). Then, by definition we have
[TABLE]
For any m′∈M(X), it may be easily verified that ηm′:hX⟶M defined by ηm′(Y)(f):=M(f)(m′) for each f∈HomD(Y,X) is a morphism in Mod-D. By Lemma 2.4, this induces the morphism ηm′⊗idC:hX⊗C⟶M⊗C in M(ψ)DC defined by (ηm′⊗idC)(Y)(f⊗c):=M(f)(m′)⊗c for f∈HomD(Y,X) and c∈C. Since υ is a natural transformation, it follows easily that the following diagram commutes
[TABLE]
In particular, we have
[TABLE]
The comodule structure on entwined modules determines a morphism in M(ψ)DC as follows. We define ρ~:M⟶M⊗C given by
[TABLE]
for any M∈Ob(M(ψ)DC) and X∈Ob(D). We first verify that ρ~ is a morphism of right D-modules. For any f∈HomD(Y,X) and m∈M(X), we have
[TABLE]
It may be verified easily that ρ~(X):M(X)⟶M(X)⊗C is right C-colinear. Thus, ρ~:M⟶M⊗C is a morphism in M(ψ)DC. Therefore, we have the following commutative diagram
[TABLE]
Thus, we get
[TABLE]
Now applying idM(X)⊗εC on both sides, we obtain
[TABLE]
∎
Theorem 3.8**.**
Let F:M(ψ)DC⟶Mod-D be the forgetful functor and G:Mod-D⟶M(ψ)DC, N↦N⊗C be its right adjoint. Then, F is separable if and only if there exists θ∈V1 such that
[TABLE]
Proof.
We first recall from (3.1) that the unit of the adjunction is given by
[TABLE]
for M∈Ob(M(ψ)DC) and m∈M(X). Suppose that F is separable. Then, by Theorem 3.2, there exists υ∈V such that υ∘μ=1M(ψ)DC. Therefore, using Proposition 3.7, corresponding to υ∈V we can obtain an element θ∈V1 given by
\theta_{X}(c\otimes d)=\big{(}(id_{{\bf h}_{X}}\otimes\varepsilon_{C})\upsilon({\bf h}_{X}\otimes C)\big{)}(X)(id_{X}\otimes c\otimes d) for each c,d∈C.
Moreover, we have
[TABLE]
for any c∈C.
Conversely, suppose that θ∈V1 is such that θX∘ΔC=εC⋅idX for every
X∈Ob(D). Corresponding to θ∈V1 there exists υ∈V defined by
[TABLE]
for \mathcal{M}\in Ob\big{(}{\mathscr{M}(\psi)}_{\mathcal{D}}^{C}\big{)},~{}X\in Ob(\mathcal{D}),~{}m\in\mathcal{M}(X) and c∈C. Further, we have
[TABLE]
This shows that υ∘μ=1M(ψ)DC. Hence, F is separable by Theorem 3.2.
∎
Next we investigate the separability of the functor G:Mod-D⟶M(ψ)DC given by G(N)=N⊗C for any N∈Mod-D. Since G is a right adjoint of F, it follows from Theorem 3.2 that the functor G is separable if and only if there exists a natural transformation ω:1Mod-D⟶FG such that ν∘ω=1Mod-D, where ν is the counit of the adjunction as explained in (3.2).
We set W:=Nat(1Mod-D,FG) and proceed to give another interpretation of W.
We define h:Dop⊗D⟶VectK as
[TABLE]
for any (X,Y)∈Ob(Dop⊗D), \phi:=(\phi^{\prime},\phi^{\prime\prime})\in Hom_{\mathcal{D}^{op}\otimes\mathcal{D}}\big{(}(X,Y),(X^{\prime},Y^{\prime})\big{)} and f∈HomD(X,Y). Similarly, we define the functor h⊗C:Dop⊗D⟶VectK as
[TABLE]
for any (X,Y)∈Ob(Dop⊗D), \phi:=(\phi^{\prime},\phi^{\prime\prime})\in Hom_{\mathcal{D}^{op}\otimes\mathcal{D}}\big{(}(X,Y),(X^{\prime},Y^{\prime})\big{)}, f∈HomD(X,Y) and c∈C. By slight abuse of notation, we will make no distinction between functors Dop⊗D⟶VectK and functors D⟶Mod-D.
We observe that h⊗C:Dop⊗D⟶VectK corresponds to F∘(h⊗C) when viewed as a functor from D⟶Mod-D.
Given a natural transformation η:h⟶h⊗C, it is easy to see that
[TABLE]
is a morphism of right D-modules for each Y∈Ob(D). Similarly, for each X∈Ob(D),
[TABLE]
is a morphism of left D-modules.
Throughout the rest of this section, we set W1:=Nat(h,h⊗C), the K-space consisting of all natural transformations between the functors h and h⊗C.
Lemma 3.9**.**
Let η∈W1. We set η(X,X)(idX)=∑aX⊗cX for each X∈Ob(D) and η(Y,Z)(g):=∑g^⊗cg for any g∈HomD(Y,Z). Then,
[TABLE]
Proof.
Since η(−,Z):hZ⟶hZ⊗C is a morphism of right D-modules for each Z∈Ob(D), we have the following commutative diagram:
The K-spaces W=Nat(1Mod-D,FG) and W1=Nat(h,h⊗C) are isomorphic.
Proof.
We define a K-linear map γ:W⟶W1 by setting
[TABLE]
for any (X,Y)∈Ob(Dop⊗D). We now verify that the map is well-defined. Let \phi:=(\phi^{\prime},\phi^{\prime\prime})\in Hom_{\mathcal{D}^{op}\otimes\mathcal{D}}\big{(}(X,Y),(X^{\prime},Y^{\prime})\big{)}. Since ω(hY):hY⟶hY⊗C is a morphism of right D-modules, we have the following commutative diagram:
[TABLE]
The morphism ϕ′′:Y⟶Y′ in D induces a morphism ϕ′′:hY⟶hY′ of right D-modules. Therefore, using the naturality of ω, we get the following commutative diagram:
[TABLE]
We now observe that for f∈HomD(X,Y), we have
[TABLE]
Thus, by combining the diagrams (3.24) and (3.25), we obtain the following commutative diagram:
[TABLE]
This shows that η∈W1.
Conversely, let η∈W1=Nat(h,h⊗C). For any Y∈Ob(D),
[TABLE]
is a morphism of right D-modules. For any f∈HomD(X,Y), the naturality of η gives us the following commutative diagram:
[TABLE]
Now, for any M in Mod-D, we know that M=y∈el(M)colimh∣y∣. Similarly, M⊗C=y∈el(M)colim(h∣y∣⊗C) where the colimit is taken in Mod-D. Thus, the morphisms as in (3.26) induce a morphism ω(M):M⟶M⊗C of right D-modules. Moreover, for any morphism M⟶ζN in Mod-D, the commutative diagrams as in (3.27) induce the following equality:
[TABLE]
Therefore, for η∈W1 we have obtained a natural transformation ω:1Mod-D⟶FG in W. We will denote this K-linear map by δ:W1⟶W, i.e., δ(η)=ω determined by ω(hY):=η(−,Y) for each Y∈Ob(D). It may be easily verified that the morphisms
γ and δ are inverses of each other.
∎
Theorem 3.11**.**
Let F:M(ψ)DC⟶Mod-D be the forgetful functor and G:Mod-D⟶M(ψ)DC, N↦N⊗C be its right adjoint. Then G is separable if and only if there exists η∈W1=Nat(h,h⊗C) such that
[TABLE]
Proof.
Suppose that G is separable. Then, by Theorem 3.2, there exists ω∈W=Nat(1Mod-D,FG) such that ν∘ω=1Mod-D, where ν is the counit of the adjunction. Using Proposition 3.10, corresponding to ω∈W, there exists an element η∈W1 given by η(X,Y)=ω(hY)(X) for every (X,Y)∈Ob(Dop⊗D). The condition (3.28) now follows from the definition of the counit in (3.2).
Conversely, let η∈W1 be such that (idh⊗εC)η=idh. We consider ω:1Mod-D⟶FG given by ω(hY):=η(−,Y) for each Y∈Ob(D). Then, (idhY⊗εC)ω(hY)=(idhY⊗εC)η(−,Y)=idhY. Since F is a left adjoint and it is clear from the definition that G preserves colimits, we obtain that (idN⊗εC)ω(N)=idN for any N∈Mod-D, i.e, (id⊗εC)ω=1Mod-D. Therefore, G is separable by Theorem 3.2.
∎
3.2 Frobenius conditions
Let F:A⟶B be a functor which has a right adjoint G:B⟶A. Then, the pair (F,G) is called a Frobenius pair if G is both a right and a left adjoint of F. We recall the following characterization for Frobenius pairs (see [uni, § 1])
Theorem 3.12**.**
Let F:A⟶B be a functor which has a right adjoint G. Then, (F,G) is a Frobenius pair if and only if there exist υ∈Nat(GF,1A) and ω∈Nat(1B,FG) such that
[TABLE]
for all M∈A and N∈B.
Lemma 3.13**.**
For any ω∈W=Nat(1Mod-D,FG), N∈Comod-C and Y∈Ob(D), we have ω(N⊗hY)=idN⊗ω(hY).
Proof.
For each n∈N, we define ζn:hY⟶N⊗hY by
[TABLE]
for any X∈Ob(D) and f∈HomD(X,Y). It may be easily verified that ζn is a morphism of right D-modules. Therefore, using the naturality of ω, we have the following commutative diagram:
[TABLE]
Let f∈HomD(X,Y). We set ω(hY)(X)(f)=∑f^⊗cf. Then, we have
[TABLE]
The result follows.
∎
Theorem 3.14**.**
Let F:M(ψ)DC⟶Mod-D be the forgetful functor and G:Mod-D⟶M(ψ)DC, N↦N⊗C be its right adjoint. Then, (F,G) is a Frobenius pair if and only if there exist θ∈V1 and η∈W1 such that the following conditions hold:
[TABLE]
for any f∈HomD(X,Y), d∈C and η(X,Y)(f)=∑f^⊗cf.
Proof.
Suppose there exist θ∈V1 and η∈W1 such that (3.31) and (3.32) hold. Then, using the isomorphisms V≅V1 and W≅W1 as in Propositions 3.7 and 3.10, there exist υ∈V and ω∈W corresponding to θ∈V1 and η∈W1 respectively.
We also know by Proposition 2.9 that the collection {N⊗hY}, where N ranges over all (isomorphisms classes of) finite dimensional C-comodules and Y ranges over all objects in D, forms a generating set for M(ψ)DC. Therefore, we first verify the condition (3.29) for M=N⊗hY∈M(ψ)DC, where N∈Comod-C and Y∈Ob(D). For any n⊗f∈N⊗HomD(X,Y), we have
[TABLE]
This proves (3.29) for the generators of M(ψ)DC. As explained in the proof of Proposition 2.9, for any M in M(ψ)DC, there is an epimorphism
[TABLE]
in M(ψ)DC.
The morphism η:=m∈el(M)⨁ηm induces the following commutative diagram:
[TABLE]
From (3.33), it follows that F(υ(Vm⊗h∣m∣))ω(F(Vm⊗h∣m∣))=idF(Vm⊗h∣m∣) for each m∈el(M). Thus, by the commutative diagram (3.34), we have
[TABLE]
Since F is a left adjoint, it preserves epimorphisms. Since η is an epimorphism, so is F(η). Therefore, (3.35) implies that F(υ(M))∘ω(F(M))=idF(M). This proves (3.29) for any M∈Ob(M(ψ)DC).
Next, we verify the condition (3.30). From the definition, it is clear that G preserves colimits. Since any D-module may be expressed as the colimit of representable functors, it is enough to verify the condition (3.30) for representable functors. For any f⊗d∈hY(X)⊗C, we have
[TABLE]
This proves (3.30). Therefore, (F,G) is a Frobenius pair.
Conversely, suppose (F,G) is a Frobenius pair. Then, there exist υ∈V and ω∈W satisfying (3.29) and (3.30). Then, using the isomorphisms V≅V1 and W≅W1 as in Propositions 3.7 and 3.10, there exist θ∈V1 and η∈W1 corrresponding to υ∈V and ω∈W respectively. We will now verify the conditions (3.31) and (3.32). Taking M=C⊗hY in (3.29), for any d∈C and f∈HomD(X,Y) we have
[TABLE]
Applying εC⊗idhY(X) on both sides, we get
[TABLE]
This proves (3.32).
Now, taking N=hY in (3.30), we have
[TABLE]
Applying idhY(X)⊗εC on both sides, we get (3.31). This proves the result.
∎
3.3 Frobenius conditions in the case of a finite dimensional coalgebra
We continue with (D,C,ψ) being an entwining structure.
For each Y∈Ob(D), we obtain an object Hom(C,hY) in Mod-D by setting
[TABLE]
for any X∈Ob(D), g∈HomD(X′,X), \phi\in Hom_{K}\big{(}C,{\bf h}_{Y}(X)\big{)} and x∈C.
Using (3.36), we now define a functor Hom(C,h):D⟶Mod-D as follows:
[TABLE]
for any f∈HomD(Y,X), \phi\in Hom_{K}\big{(}C,{\bf h}_{Y}(Z)\big{)} and x∈C.
For the rest of this section, we assume that C is finite dimensional. Then, for each Z∈Ob(D), we have an isomorphism
[TABLE]
Let {di}1≤i≤k be a basis for C and
{di∗}1≤i≤k be its dual basis.
Lemma 3.15**.**
Let C be a finite dimensional coalgebra. Then, we have a functor
[TABLE]
Proof.
For each Y∈Ob(D), it is clear that C∗⊗hY∈Mod-D. We consider
f∈HomD(Y,X) and an element c∗⊗g∈C∗⊗hY(Z). By the isomorphism in (3.38),
c∗⊗g corresponds to the element \phi_{c^{*}\otimes g}\in Hom_{K}\big{(}C,{\bf h}_{Y}(Z)\big{)} given by ϕc∗⊗g(x)=c∗(x)g for each x∈C. From the action in (3.37), the element f\cdot\phi_{c^{*}\otimes g}\in\big{(}Hom(C,{\bf h}_{X})\big{)}(Z) is given by
[TABLE]
Again, using the isomorphism in (3.38), the element in C∗⊗hX(Z) corresponding
to f⋅ϕc∗⊗g is given by ∑i=1kc∗(diψ)di∗⊗fψg. It may be easily verified that (C∗⊗h)(f):C∗⊗hY⟶C∗⊗hX is a morphism of right D-modules. The result now follows.
∎
Since C is a coalgebra, its vector space dual C∗ is an algebra with the convolution product (c^{*}\mbox{\tiny\bullet }d^{*})(x):=\sum c^{*}(x_{1})d^{*}(x_{2}) for c∗,d∗∈C∗ and x∈C. Let N be any left C∗-module. Then, we have a K-linear map ρ:N⟶Hom(C∗,N) defined by ρ(n)(c∗):=c∗n for n∈N and c∗∈C∗.
In general, there is an embedding N⊗C↪Hom(C∗,N) given by (n⊗x)(c∗):=c∗(x)n for x∈C. Since C is finite dimensional, this embedding is also a surjection. This gives us a K-linear map ρ:N⟶N⊗C which makes N a right C-comodule (see, for instance, [SCS, § 2.2]). Then, ρ(n)=∑i=1kdi∗n⊗di. In particular, C∗ becomes a right C-comodule with
[TABLE]
Considering the element εC∈C∗, the coassociativity of the coaction
ρC∗ may be used to verify that
[TABLE]
Proposition 3.16**.**
Let C be a finite dimensional coalgebra. Then, we have a functor:
[TABLE]
Proof.
From (3.40), we know that C∗ is a right C-comodule.
Applying Lemma 2.5, it follows that each C∗⊗hY is an object in M(ψ)DC. Accordingly, the right C-comodule structure on C∗⊗hY(Z) for any Z∈Ob(D) is given by the following composition:
[TABLE]
Explicitly, we have
\sigma^{r}_{C^{*}\otimes{\bf h}_{Y}(Z)}(c^{*}\otimes g)=\sum\limits_{i=1}^{k}d_{i}^{*}\mbox{\tiny\bullet }c^{*}\otimes g_{\psi}\otimes d_{i}^{\psi}
for each c∗⊗g∈C∗⊗hY(Z). We consider f∈HomD(Y,X). By Lemma 3.15, this induces a morphism C∗⊗hY⟶C∗⊗hX in Mod-D. In order to show that
C∗⊗h:D⟶M(ψ)DC is a functor, it therefore suffices to show
that each morphism
[TABLE]
is right C-colinear. For any c∗⊗g∈C∗⊗hY(Z), we have
[TABLE]
∎
Since C is finite dimensional, the right C-comodule structure on C∗⊗hY(X) induces a right C-comodule structure on Hom(C,hY(X)) for each X,Y∈Ob(D) which we now explain. Let ϕ∈Hom(C,hY(X)). Then, ϕ corresponds to the element ∑1≤i≤kdi∗⊗ϕ(di)∈C∗⊗hY(X). We know by Proposition 3.16 that
[TABLE]
The element \sum_{i=1}^{k}d_{j}^{*}\mbox{\tiny\bullet }d_{i}^{*}\otimes(\phi(d_{i}))_{\psi}\otimes d_{j}^{\psi}\in C^{*}\otimes{\bf h}_{Y}(X)\otimes C
corresponds to the element ϕ0⊗ϕ1∈Hom(C,hY(X))⊗C given by
[TABLE]
for x∈C.
It now follows from (3.36), (3.37), (3.43) and Proposition 3.16 that we have a functor
[TABLE]
We also recall from (3.3) and (3.4), the functor h⊗C:D⟶M(ψ)DC, defined as follows:
[TABLE]
for f∈HomD(Y,X) and g⊗c∈hY(Z)⊗C. We now set V2:=Nat(h⊗C,C∗⊗h).
Proposition 3.17**.**
Let C be a finite dimensional coalgebra. Then,
[TABLE]
Proof.
Since C is finite dimensional, we know that C^{*}\otimes{\bf h}_{Y}(X)\cong Hom_{K}\big{(}C,{\bf h}_{Y}(X)\big{)} for each X,Y∈Ob(D). We first define a K-linear map ΥXY:hY(X)⊗C⟶C∗⊗hY(X) given by
[TABLE]
for any f∈HomD(X,Y) and c,d∈C. In other words, we have
[TABLE]
where {di}1≤i≤k is a basis for C and {di∗}1≤i≤k is its dual basis.
We now define α′:V1⟶V2 by setting α′(θ)=Υ with Υ:h⊗C⟶C∗⊗h defined as follows:
[TABLE]
for any X,Y∈Ob(D).
We now verify that α′ is a well-defined map. For this, we first check that ΥY:hY⊗C⟶C∗⊗hY is a morphism in M(ψ)DC for every Y∈Ob(D). For any g∈HomD(X′,X), we need to show that the following diagram commutes:
[TABLE]
For any f⊗c∈hY(X)⊗C, we have
[TABLE]
This shows that ΥY is a morphism of right D-modules for every Y∈Ob(D). Next we verify that ΥY(X):hY(X)⊗C⟶C∗⊗hY(X) is right C-colinear for every X,Y∈Ob(D). We have
[TABLE]
Finally, we verify that Υ is a natural transformation from h⊗C to C∗⊗h, i.e.,
the following diagram commutes for any g∈HomD(Y,Y′):
[TABLE]
For any f⊗c∈hY(X)⊗C, we have
[TABLE]
This proves that Υ∈V2.
For the converse, we first observe that the functors C∗⊗h and Hom(C,h) are isomorphic which follows from (3.38). We define β′:V2⟶V1 by setting β′(Υ)=θ with θX:C⊗C⟶EndD(X) defined as follows:
[TABLE]
for any X∈Ob(D) and c,d∈C. We will now verify that θ satisfies (3.11) and (3.12). For each X∈Ob(D), we know that ΥX:hX⊗C⟶Hom(C,hX) is a morphism of right D-modules. Therefore, for any f∈HomD(Y,X), we have the following commutative diagram:
[TABLE]
Since Υ:h⊗C⟶Hom(C,h) is a natural transformation, the following diagram also commutes for any f∈HomD(Y,X):
This proves (3.12).
It remains to show that α′ and β′ are inverses of each other. For every θ∈V1 and c,d∈C, it follows from
(3.45) that
[TABLE]
Finally, for any Υ∈V2, f∈HomD(X,Y) and c,d∈C, we have
[TABLE]
This proves the result.
∎
Proposition 3.18**.**
Let C be a finite dimensional coalgebra. Then, we have isomorphisms
[TABLE]
Proof.
Given an η:h⟶h⊗C, we want to define Φ:C∗⊗h⟶h⊗C. For each Y∈Ob(D), we first define a K-linear map ΦYY:C∗⊗hY(Y)⟶hY(Y)⊗C by the following composition:
[TABLE]
i.e., ΦYY(c∗⊗idY)=∑aY⊗c∗(cY2)cY1, where ∑aY⊗cY=η(Y,Y)(idY) as in the notation of Lemma 3.9.
We observe that an element c∗⊗f∈C∗⊗hY(X) may be written as c∗⊗f=(C∗⊗hY)(f)(c∗⊗idY). For each X∈Ob(D), we now define ΦXY:C∗⊗hY(X)⟶hY(X)⊗C as follows:
[TABLE]
for any c∗⊗f∈C∗⊗hY(X).
We define γ′:W1⟶W2 by setting γ′(η)=Φ with Φ:C∗⊗h⟶h⊗C given by
[TABLE]
for every X,Y∈Ob(D). We now verify that γ′ is a well-defined map. For this, we first check that ΦY:C∗⊗hY⟶hY⊗C is a morphism of right D-modules for every Y∈Ob(D), i.e., the following diagram commutes for any g∈HomD(X′,X):
[TABLE]
We have
[TABLE]
Next we verify that ΦY(X):C∗⊗hY(X)⟶hY(X)⊗C is right C-colinear for any X,Y∈Ob(D):
[TABLE]
It follows that ΦY:C∗⊗hY⟶hY⊗C is a morphism in M(ψ)DC.
To show that Φ∈Nat(C∗⊗h,h⊗C), it remains to verify that the following diagram commutes:
[TABLE]
for any g∈HomD(Y,Z). For any X∈Ob(D) and c∗⊗f∈C∗⊗hY(X), we have
[TABLE]
Conversely, we define δ′:W2⟶W1 by setting δ′(Φ)=η with η:h⟶h⊗C given by
[TABLE]
for any (X,Y)∈Ob(Dop⊗D) and f∈HomD(X,Y).
We now verify that η∈W1. Let ϕ:(X,Y)⟶(X′,Y′) be a morphism in Dop⊗D given by ϕ′:X′⟶X and ϕ′′:Y⟶Y′ in D. Then, using the fact that ΦY:C∗⊗hY⟶hY⊗C is a morphism of right D-modules, we have
[TABLE]
for any f∈HomD(X,Y). This shows that the following diagram commutes:
[TABLE]
Now using the naturality of Φ:C∗⊗h⟶h⊗C, we also have
[TABLE]
for any g∈hY(X′). Thus, we get the following commutative diagram:
[TABLE]
It now follows from (3.51) and (3.52) that the following diagram commutes:
[TABLE]
This shows that η∈W2.
It remains to check that γ′ and δ′ are inverses of each other.
First we verify that \big{(}(\delta^{\prime}\circ\gamma^{\prime})(\eta)\big{)}(X,Y)=\eta(X,Y) for all X,Y∈Ob(D). For this, we set Φ=γ′(η). Then, for any f∈HomD(X,Y), we have
[TABLE]
Next, we will show that \big{(}(\gamma^{\prime}\circ\delta^{\prime})(\Phi)\big{)}_{Y}(X)=\Phi_{Y}(X) for any X,Y∈Ob(D). Since C∗⊗hY(X) and hY(X)⊗C are right C-comodules for any X,Y∈Ob(D), they are also left C∗-modules. The left actions are respectively given by
[TABLE]
for any d∗,c∗∈C∗, f∈hY(X) and x∈C. Moreover, since ΦY(X):C∗⊗hY(X)⟶hY(X)⊗C is right C-colinear, it is also left C∗-linear.
We now set η=δ′(Φ). Then, for any c∗⊗f∈C∗⊗hY(X), we have
[TABLE]
This proves the result.
∎
Theorem 3.19**.**
Let (D,C,ψ) be an entwining structure and assume that C is a finite dimensional coalgebra. Let F:M(ψ)DC⟶Mod-D be the functor forgetting the C-coaction and G:Mod-D⟶M(ψ)DC given by N↦N⊗C be its right adjoint. Then, the following statements are equivalent:
(i) (F,G) is a Frobenius pair.
(ii) There exist η∈W1 and θ∈V1 such that the corresponding morphisms
γ′(η)=Φ:C∗⊗h⟶h⊗Candα′(θ)=Υ:h⊗C⟶C∗⊗hgiven by
[TABLE]
where f∈hY(X), c∗∈C∗ and d∈C, are inverses of each other.
(iii) C∗⊗h and h⊗C are isomorphic as objects of the category DM(ψ)DC of functors from D to M(ψ)DC.
Proof.
(i) ⇒ (ii) By assumption, there exist η∈W1 and θ∈V1 satisfying (3.31) and (3.32). Then, α′(θ)=Υ and γ′(η)=Φ are morphisms in DM(ψ)DC in the notation of Proposition 3.17 and Proposition 3.18. Since ΥXY:hY(X)⊗C⟶C∗⊗hY(X) and ΦXY:C∗⊗hY(X)⟶hY(X)⊗C are right C-colinear, they are also left C∗-linear. Using this fact and (3.53), we have
[TABLE]
for any c∗⊗f∈C∗⊗hY(X). Thus, Υ∘Φ=idC∗⊗h.
Using the naturality of Υ and Φ, we have
[TABLE]
for any f⊗c∈hY(X)⊗C. Thus, Φ∘Υ=idh⊗C. This proves (ii).
(ii) ⇒ (iii) is obvious since both Φ and Υ are morphisms in DM(ψ)DC.
(iii) ⇒ (i) Let Φ:C∗⊗h⟶h⊗C denote the isomorphism in DM(ψ)DC. We consider the following morphism of (Dop⊗D)-modules
[TABLE]
for any f∈HomD(X,Y). We now set η=Φ∘Λ∈W1 and θ=β′(Φ−1)∈V1 where β′ is as in Proposition 3.17. If η(X,Y)(f)=∑f^⊗cf, then
[TABLE]
Using the isomorphism as in (3.38) and evaluating the equality in (3.55) at d∈C, we get (3.32). We also have
for any f∈HomD(X,Y) and d∈C. Applying to both sides the composition HomD(X,Y)⊗HomD(X,X)⟶HomD(X,Y), we obtain ∑f^∘(θX(cf⊗d))=∑faX∘θX(cX⊗d)=εC(d)f. This proves (3.31). Therefore, (F,G) is a Frobenius pair by Theorem 3.14. This completes the proof.
∎
4 Categorical Galois extensions and entwining structures
Let D be a small K-linear category. Let (D,C,ψ) be a right-right entwining structure. We denote by DMD the category of D-D bimodules, i.e., the category whose objects are functors from Dop⊗D to VectK and whose morphisms are natural transformations between these functors. We recall the functors h and h⊗C in DMD from (3.20) and (3.21) respectively:
[TABLE]
for any (X,Y)∈Ob(Dop⊗D), \phi:=(\phi^{\prime},\phi^{\prime\prime})\in Hom_{\mathcal{D}^{op}\otimes\mathcal{D}}\big{(}(X,Y),(X^{\prime},Y^{\prime})\big{)} and f∈HomD(X,Y), c∈C. We refer, for instance, to [EV, § 2.2] for the tensor product which makes DMD a monoidal category with h∈DMD as the unit object.
Definition 4.1**.**
A D-coring C is a coalgebra object in the monoidal category DMD. Explicitly, a D-coring is a functor C:Dop⊗D⟶VectK with two morphisms
[TABLE]
satisfying the coassociativity and counit axioms in DMD. A right C-comodule consists of a right D-module M equipped with a morphism ρM:M⟶M⊗DC of right D-modules satisfying
[TABLE]
A morphism η:(M,ρM)⟶(N,ρN) of right C-comodules is a morphism η:M⟶N of right D-modules satisfying
[TABLE]
The category of right C-comodules will be denoted by Comod-C.
Lemma 4.2**.**
Let (D,C,ψ) be a right-right entwining structure. Then, the functor h⊗C is a D-coring.
Proof.
It may be verified that (h⊗C)⊗D(h⊗C)≅h⊗C⊗C. This gives us morphisms
[TABLE]
in DMD. Using the coassociativity and counitality of the K-coalgebra C, it may be verified that idh⊗ΔC and idh⊗εC satisfy the coassociativity and counit axioms in the category DMD. Thus, h⊗C is a coalgebra object in DMD.
∎
Proposition 4.3**.**
Let (D,C,ψ) be a right-right entwining structure. Then, the category M(ψ)DC of entwined modules is identical to the category Comod-(h⊗C).
Proof.
Let M∈M(ψ)DC. It may be verified that M⊗C≅M⊗D(h⊗C) as right D-modules. Then,
by Lemma 2.4, M⊗C∈M(ψ)DC and we have
[TABLE]
for any f∈HomD(X,Y) and m∈M(Y). We thus obtain a morphism ρM:M⟶M⊗C≅M⊗D(h⊗C) of right D-modules given by ρM(X):=ρM(X) for each X∈Ob(D).
The conditions in (4.3) now follow from the fact that ρM(X) is a C-coaction for each X∈Ob(D). Therefore, M is a right (h⊗C)-comodule.
Conversely, let N∈Comod-(h⊗C). Then, N is a right D-module with a given morphism ρN:N⟶N⊗D(h⊗C)≅N⊗C of right D-modules satisfying the conditions in (4.3). Thus, for each Y∈Ob(D), we have a morphism ρN(Y):N(Y)⟶N(Y)⊗C which satisfies
[TABLE]
In (4.7), we have identified idN⊗ΔC=idN⊗Δh⊗C and idN⊗εC=idN⊗ε(h⊗C) as in (4.5) and (4.6) respectively. Therefore, ρN(Y) defines a right C-comodule structure on N(Y) for every Y∈Ob(D). Since ρN is a morphism of right D-modules, we also have
[TABLE]
for any f∈HomD(X,Y) and n∈N(Y). Therefore, N∈M(ψ)DC.
∎
Lemma 4.4**.**
Let i:E⟶D be an inclusion of small K-linear categories. Then, the functor h⊗Eh:Eop⊗E⟶VectK is a D-coring, where h is the D-D-bimodule as in (4.1).
Proof.
It is immediate that the functor h⊗Eh is a D-D-bimodule. We need to show that h⊗Eh is a coalgebra
object in DMD. We now define Δ:h⊗Eh⟶(h⊗Eh)⊗D(h⊗Eh)≅(h⊗Eh)⊗Eh as follows: for (X,Y)∈Ob(Dop⊗D), we set
[TABLE]
for any f⊗f′∈hY(Z)⊗Xh(Z) and Z∈Ob(E). It is easy to check that Δ(X,Y) is well-defined. Also, it can be verified that for any morphism (ϕ′,ϕ′′):(X,Y)⟶(X′,Y′) in Dop⊗D, the following diagram commutes:
[TABLE]
Thus, Δ is a morphism of D-D-bimodules.
The map ε:h⊗Eh⟶h is defined by composition. It may be verified that Δ and ε satisfy the coassociativity and counit axioms respectively.
∎
Let D be a small K-linear category and let C be a K-coalgebra. We consider the category DMC of left-right Doi-Hopf modules (compare Example 2.3). Explicitly, an object in DMC consists of a left D-module M with a given right C-comodule structure on M(X) for each X∈Ob(D) such that the following compatibility condition holds:
[TABLE]
for each f∈HomD(X,Y) and m∈M(X).
A morphism η:M⟶N in DMC is a left D-module morphism such that each η(X):M(X)⟶N(X) is right C-colinear.
By definition, (h⊗C)(X,−)=Xh⊗C is a left D-module for each X∈Ob(D). The map id⊗ΔC:HomD(X,Y)⊗C⟶HomD(X,Y)⊗C⊗C gives a right C-comodule structure on (Xh⊗C)(Y) for each Y∈Ob(D). Clearly, Xh⊗C∈DMC.
From this point onwards, we suppose additionally that each HomD(X,Y) has a given right C-comodule structure denoted by
[TABLE]
Definition 4.5**.**
Let E⊆D be the subcategory with Ob(E)=Ob(D) and
[TABLE]
We will say that E is the subcategory of C-coinvariants of D.
Example 4.6**.**
Let H be a Hopf algebra over K and let D be a right co-H-category. In this case, the subcategory E of H-coinvariants of D is given by setting Ob(E)=Ob(D) and HomE(X,Y)=HomD(X,Y)coH.
It follows that the right C-comodule structures ρXY:HomD(X,Y)⟶HomD(X,Y)⊗C induce a morphism Xh⟶Xh⊗C of left E-modules for each X∈Ob(D). Further, for every Y∈Ob(D), this induces a morphism
[TABLE]
where f∈HomD(Z,Y),f′∈HomD(X,Z) and Z∈Ob(E). It may be easily verified that the coaction in (4.10) makes h⊗EXh an object of DMC.
We obtain therefore canonical morphisms of K-vector spaces given by the following composition
[TABLE]
For each X∈Ob(D), this induces a morphism in DMC as follows
[TABLE]
Definition 4.7**.**
Let C be a K-coalgebra and D be a small K-linear category such that HomD(X,Y) has a right C-comodule structure for every X,Y∈Ob(D). Let E be a K-linear subcategory of D. Then, D is called a C-Galois extension of E if
(i)
Ob(E)=Ob(D)* and HomE(X,Y)=HomMod-DC(hX,hY).*
(ii)
The induced canonical morphism canX:h⊗EXh⟶Xh⊗C is an isomorphism in DMC for each X∈Ob(D).
Let D be a C-Galois extension of E. For each X∈Ob(D), we define
[TABLE]
We refer to these as the translation maps of the Galois extension.
Lemma 4.8**.**
*Let D be a C-Galois extension of E. Let {τX:C⟶hX⊗EXh}X∈Ob(D) be the associated translation maps. We use the notation τX(c)=c(1)⊗c(2) (summation omitted). Then,
(i) τX is right C-colinear i.e., c(1)⊗c(2)0⊗c(2)1=(c1)(1)⊗(c1)(2)⊗c2.
(ii) For any f∈HomD(X,Y), we have f0(f1)(1)⊗(f1)(2)=idY⊗f∈hY⊗EXh.
(iii) c(1)c(2)=εC(c)⋅idX.*
Proof.
The C-colinearity of τX follows from the C-colinearity of canXX−1. Explicitly, for any c∈C, we have
[TABLE]
This proves (i). Since canX−1:Xh⊗C⟶h⊗EXh is a morphism of left D-modules for each X∈Ob(D), we also have
[TABLE]
This proves (ii). Again using the definition of canXX and τX, we have (canXX∘τX)(c)=idX⊗c. Thus,
[TABLE]
Now, by applying the map id⊗εC to both sides, we get (iii).
∎
Theorem 4.9**.**
Let D be a C-Galois extension of E. We denote by
ρXY:HomD(X,Y)⟶HomD(X,Y)⊗C the right C-comodule structure maps.
Then,
there exists a unique right-right entwining structure (D,C,ψ) which makes hY an object in M(ψ)DC for every Y∈Ob(D) with its canonical D-module structure and right C-coactions {ρXY}X∈Ob(D).
This entwining structure (D,C,ψ) is given by
[TABLE]
Proof.
Using Lemma 4.8, the proof will follow essentially in the same way as that of [BH, Theorem 2.7].
∎
Lemma 4.10**.**
Let D be a C-Galois extension of E. Then, h⊗Eh≅h⊗C as D-corings.
Proof.
We define can:h⊗Eh⟶h⊗C by setting can(X,Y):=canXY for each (X,Y)∈Ob(Dop⊗D). We first verify that can is a morphism of D-D-bimodules. Clearly, can(X,−)=canX which, by definition, is a morphism of left D-modules. Therefore, it suffices to show that can(−,Y) is a morphism of right D-modules, i.e., the following diagram commutes for any g∈HomD(Z,Z′):
[TABLE]
By Theorem 4.9, we know that hW is an object in M(ψ)DC for each W∈Ob(D). Thus, for any f∈hW(Z′), we have
[TABLE]
Therefore, for any f′⊗f∈hY(W)⊗EZ′h(W), we obtain
[TABLE]
It remains to verify that can is also a coalgebra morphism. First, we show that the following diagram commutes:
[TABLE]
For any (X,Y)∈Ob(Dop⊗D) and w⊗w′∈hY(W)⊗Xh(W), we have
[TABLE]
It may be verified easily that can is compatible with counits.
Since can is a morphism in the category of D-D-bimodules and can(X,Y)=canXY is an isomorphism for each (X,Y)∈Ob(Dop⊗D), it follows that can is an isomorphism with inverse given by can−1(X,Y):=canXY−1. This proves the result.
∎
Definition 4.11**.**
Let D be a small K-linear category such that HomD(X,Y) is a right C-comodule for every X,Y∈Ob(D). Let ΦXY:C⟶HomD(X,Y) and ΦYZ:C⟶HomD(Y,Z) be two C-comodule maps. Then, their convolution product is given by
[TABLE]
A collection of right C-comodule maps Φ={ΦXY:C⟶HomD(X,Y)}X,Y∈Ob(D) is said to be convolution invertible if there exists a collection Φ′={ΦXY′:C⟶HomD(X,Y)}X,Y∈Ob(D) of C-comodule maps such that
[TABLE]
for every c∈C.
Theorem 4.12**.**
Let C be a K-coalgebra and D be a small K-linear category such that HomD(X,Y) has a right C-comodule structure ρXY for every X,Y∈Ob(D). Let E be the subcategory of C-coinvariants of D. If there exists a convolution invertible collection Φ={ΦXY:C⟶HomD(X,Y)}X,Y∈Ob(D) of right
C-comodule maps, then the following are equivalent:
(i) D is a C-Galois extension of E.
(ii) There exists a right-right entwining structure (D,C,ψ) such that hY is an object in M(ψ)DC for every Y∈Ob(D) with its canonical D-module structure and right C-coactions {ρXY}X∈Ob(D).
(iii) For any f∈HomD(X,Y), the morphism f0∘ΦZX′(f1)∈HomE(Z,Y) for every Z∈Ob(D), where Φ′ is the convolution inverse of Φ.
Proof.
By Theorem 4.9, we have (i)⇒(ii). To prove (ii)⇒(iii), we will use the equality
[TABLE]
for any c∈C. We first give a proof of this. Since hY∈M(ψ)DC, we have
For any f∈HomD(X,Y), consider the morphism f0∘ΦZX′(f1):Z⟶Y in D. Then f0∘ΦZX′(f1) induces a morphism of right D-modules hZ⟶hY which we denote by f~.
We now verify that the map f~(X′):hZ(X′)⟶hY(X′) is right C-colinear for each X′∈Ob(D). Since
hY is an object in M(ψ)DC for every Y∈Ob(D), the following diagram commutes for any g∈HomD(X′,Z):
[TABLE]
Thus, we have
[TABLE]
Therefore, f~∈HomMod-DC(hZ,hY)=HomE(Z,Y).
For (iii)⇒(i), we start by showing that canXY:hY⊗EXh⟶HomD(X,Y)⊗C is an isomorphism for each X,Y∈Ob(D). We define canXY−1:HomD(X,Y)⊗C⟶hY⊗EXh by
[TABLE]
for any f∈HomD(X,Y) and c∈C. Then, using the C-colinearity of ΦXY, we have
[TABLE]
On the other hand, by assumption, we obtain
[TABLE]
for any g⊗Eg′∈hY⊗EXh. From the definition in (4.16), it is clear that setting canX−1(Y):=canXY−1
for each Y∈Ob(D) determines a morphism in DMC which is inverse to canX. This completes the proof.
∎
Example 4.13**.**
Let H be a Hopf algebra over K. If C is a left H-module category, then the smash product category C#H (see [CiSo]) is a right co-H-category with the right H-coaction determined by f#h↦f#h1⊗h2 on each HomC#H(X,Y)=HomC(X,Y)⊗H. By definition, we know that Ob(C)=Ob(C#H). It is easy to see that HomC(X,Y)=HomC#H(X,Y)coH.
We claim that
C#H is an H-Galois extension of C. We first observe that for any f#h∈HomC#H(Z,Y) and f′#h′∈HomC#H(X,Z), we have
[TABLE]
Thus, canXY:hY⊗CXh⟶HomC#H(X,Y)⊗H has the following form
[TABLE]
for each X,Y∈Ob(C#H),
Then, it may be verified that for each X,Y∈Ob(C#H), canXY is an isomorphism with inverse canXY−1:HomC#H(X,Y)⊗H⟶hY⊗CXh determined by
[TABLE]
Proposition 4.14**.**
Let D be a C-Galois extension of E. If there exists a convolution invertible collection Φ={ΦXY:C⟶HomD(X,Y)}X,Y∈Ob(D) of right
C-comodule maps, then
[TABLE]
for each X∈Ob(E)=Ob(D).
Proof.
Let Φ′ be the convolution inverse of Φ. Given f∈HomD(X,Y), it follows from Theorem 4.12 that f0∘ΦZX′(f1)∈HomE(Z,Y) for every Z∈Ob(D). We define
[TABLE]
Using Definition 4.5, we see that ρXY′(gf)=gf0⊗f1 for any g∈HomE(Y,Y′). Hence, we have
[TABLE]
Using (4.17), it may be easily seen that η is a morphism of left E-modules. Using the coassociativity of the C-coactions {ρXY}X,Y∈Ob(D), it is also clear that η is objectwise C-colinear. Therefore, η is a morphism in EMC.
Conversely, we define ζ:HomE(X,−)⊗C⟶HomD(X,−) given by ζ(Y)(f′⊗c):=f′∘ΦXX(c) for Y∈Ob(E). It is immediate that ζ is a morphism of left E-modules. Moreover,
[TABLE]
where the last equality follows from the fact that ΦXX is C-colinear.
It follows that ζ(Y) is C-colinear for each Y∈Ob(E) and hence ζ is a morphism in EMC. It may be verified that ζ is the inverse of η.
∎
Definition 4.15**.**
Let D be a small K-linear category and E be a K-subcategory. Let (C,ΔC,εC) be a D-coring. Then, a collection
[TABLE]
is said to be group-like for C with respect to E if
(i)
ΔC(X,X)(sX)=sX⊗sX* and εC(sX)=idX for any X∈Ob(E),*
(ii)
For any f∈HomE(X,Y), we have
[TABLE]
Example 4.16**.**
(i) If E is a subcategory of D, then the collection {idX⊗idX∈hX⊗EXh}X∈Ob(E) is group-like for h⊗Eh with respect to E.
(ii) Let D be a C-Galois extension of E. Then h⊗C is a D-coring (by Theorem 4.9 and Lemma 4.2) and the collection {idX0⊗idX1∈HomD(X,X)⊗C}X∈Ob(E) is group-like for h⊗C with respect to E. Since hY∈M(ψ)DC for each Y∈Ob(D), we have
[TABLE]
for any f∈HomD(X,Y). But, if f∈HomE(X,Y), then we also have
[TABLE]
Proposition 4.17**.**
Let E⊆D be a subcategory and C be a D-coring. Let {sX}X∈Ob(E) be a group-like collection for C with respect to E. For a right C-comodule (N,ρN), the
E-submodule NcoC:Eop⟶VectK of coinvariants of N is given by:
[TABLE]
for any X∈Ob(E), f∈HomE(X,Y) and n′∈NcoC(Y).
Proof.
We will show that for any f∈HomE(X,Y), the morphism NcoC(f):NcoC(Y)⟶NcoC(X) is well-defined. Since ρN:N⟶N⊗DC is a morphism of right D-modules, we have the following commutative diagram:
[TABLE]
Let n′∈NcoC(Y) so that ρN(Y)(n′)=n′⊗sY. Since f∈HomE(X,Y), using (4.18) we have
[TABLE]
This shows that N(f)(n′)=NcoC(f)(n′)∈NcoC(X). The result follows.
∎
The next result shows that in the case of a C-Galois extension E⊆D, we recover the notion of coinvariants as in Definition (4.5).
Lemma 4.18**.**
Let D be a C-Galois extension of E. Consider the collection {idX0⊗idX1∈HomD(X,X)⊗C}X∈Ob(D) which is group-like for h⊗C with respect to E. Then, (HomD(−,Y))co(h⊗C)(X)=HomE(X,Y) for any X, Y∈Ob(D)=Ob(E).
Proof.
Since D is a C-Galois extension of E, we know that there is a canonical entwining (D,C,ψ) such that hY∈M(ψ)DC. Using Proposition 4.3, hY may be treated as an object of Comod-(h⊗C).
Let g∈(HomD(−,Y))co(h⊗C)(X). Then, ρXY(g)=g∘idX0⊗idX1. Using the fact that hY∈M(ψ)DC we have
[TABLE]
for any f∈HomD(Z,X). Therefore, g∈HomE(X,Y). The converse follows directly using the Definition (4.5).
∎
Lemma 4.19**.**
Let D be a C-Galois extension of E and let (D,C,ψ) be the canonical entwining structure associated to it. We denote by
ρXY:HomD(X,Y)⟶HomD(X,Y)⊗C the right C-comodule structure maps. Then, for any M∈Mod-E, we may obtain an object M⊗Eh∈M(ψ)DC by setting
[TABLE]
for f∈HomD(X,Y) and m⊗g∈M(Z)⊗Yh(Z). In fact, this determines a functor
from Mod-E to M(ψ)DC.
Proof.
Clearly, M⊗Eh∈Mod-D. For each Y∈Ob(D), it may be verified that M⊗EYh has a right C-comodule structure given by
[TABLE]
for any g∈HomD(Y,Z) and m∈M(Z). By Theorem 4.9, hZ is an object in M(ψ)DC for every Z∈Ob(D) with its canonical D-module structure and right C-coactions {ρXZ}X∈Ob(D). Therefore, we have
ρXZ(hZ(f)(g)))=(gf)0⊗(gf)1=g0fψ⊗g1ψ for any f∈HomD(X,Y). Consequently, we have
[TABLE]
This shows that M⊗Eh∈M(ψ)DC.
∎
Lemma 4.20**.**
Let D be a C-Galois extension of E. If there exists a convolution invertible collection Φ={ΦXY:C⟶HomD(X,Y)}X,Y∈Ob(D) of right
C-comodule maps, then
(i) HomD(X,−) is flat as a left E-module.
(ii) X∈Ob(D)⨁HomD(X,−) is faithfully flat as a left E-module.
(iii) For any M∈Mod-E, there is a monomorphism M↪M⊗Eh in Mod-E given by m↦m⊗idX for any m∈M(X).
Proof.
(i) Let i:M1↪M2 be a monomorphism of right E-modules. By Proposition 4.14, it follows that the induced map M1⊗EHomD(X,−)⟶M2⊗EHomD(X,−) coincides with the map M1(X)⊗Ci(X)⊗idCM2(X)⊗C for each X∈Ob(E)=Ob(D). Since i(X)⊗idC is clearly a monomorphism, it follows that HomD(X,−) is flat as a left E-module.
(ii) This is clear from the fact that M(X)⊗C=M⊗EHomD(X,−)=0⇒M(X)=0.
(iii) Since Y∈Ob(D)⨁HomD(Y,−) is faithfully flat as a left E-module, it is enough to prove that for each Y∈Ob(D), we have a monomorphism
[TABLE]
This is true because the morphism in (4.20) has a section
[TABLE]
for any m′∈M(Z) and g′⊗f′∈Xh(Z)⊗Yh(X).
∎
Theorem 4.21**.**
Let D be a C-Galois extension of E and let (D,C,ψ) be the canonical entwining structure associated to it.
Suppose there exists a convolution invertible collection Φ={ΦXY:C⟶HomD(X,Y)}X,Y∈Ob(D) of right
C-comodule maps. Then, the categories M(ψ)DC and Mod-E are equivalent.
Proof.
We consider the collection {idX0⊗idX1∈HomD(X,X)⊗C}X∈Ob(E) which is group-like for the coring h⊗C with respect to E.
We define
[TABLE]
Using Lemma 4.19 and Proposition 4.17, we see that the functors F and G are well-defined. We now verify that G∘F≅idMod-E i.e., (M⊗Eh)co(h⊗C)≅M for any M∈Mod-E.
From Lemma 4.10, we know that h⊗C≅h⊗Eh as D-corings. Under this isomorphism, the collection {idX0⊗idX1∈HomD(X,X)⊗C}X∈Ob(E) maps to the collection {idX⊗idX∈hX⊗Xh}X∈Ob(E) which is group-like for h⊗Eh with respect to E. Therefore, it suffices to show that M≅(M⊗Eh)co(h⊗Eh).
By Lemma (4.20)(iii), we have an inclusion i:M⟶M⊗Eh of right E-modules. It is clear that i(M)⊆(M⊗Eh)co(h⊗Eh).
By definition, ρ~=ρM⊗Eh:M⊗Eh⟶(M⊗Eh)⊗D(h⊗Eh) is determined by
[TABLE]
for each X∈Ob(D).
The coinvariants (M⊗Eh)co(h⊗Eh):Eop⟶VectK are given by
[TABLE]
For ∑mY⊗fY∈(M⊗Eh)co(h⊗Eh)(X), we now have
[TABLE]
We set P:=(M⊗Eh)/M∈Mod-E and consider the following short exact sequence:
[TABLE]
Then η induces the morphism η⊗idh:(M⊗Eh)⊗Eh⟶P⊗Eh of right E-modules which for each X∈Ob(D) is given by
[TABLE]
where m′∈M(Z), f′∈HomD(Y,Z), g′∈HomD(X,Y) and Y,Z∈Ob(E).
Applying (η⊗idh)(X) to (4.22), we obtain
[TABLE]
Applying Lemma (4.20)(iii) to the inclusion P↪P⊗Eh, it follows from (4.23) that ∑η(X)(mY⊗EfY)=0 for every X∈Ob(E). Therefore, ∑mY⊗fY∈i(M)(X). This proves that M≅(M⊗Eh)co(h⊗Eh).
It remains to show that F∘G≅idM(ψ)DC. Let N∈M(ψ)DC≅Comod-(h⊗C). Then, N is a right D-module with a given morphism
[TABLE]
in M(ψ)DC. By definition, Nco(h⊗C) is the equalizer of the following morphisms
[TABLE]
where j is given by
[TABLE]
for every X∈Ob(D). By Lemma 4.20(i), it follows that Nco(h⊗C)⊗EXh is the equalizer of the following morphisms
[TABLE]
Comparing with (4.9), we observe that j⊗id=idN⊗EΔh⊗Eh(X,−).
Using the coassociativity of ρN:N⟶N⊗Eh, it follows from (4.25) that ρN(X) factorises through Nco(h⊗C)⊗EXh, which is denoted by ρN′(X):N(X)⟶Nco(h⊗C)⊗EXh⊆N⊗EXh.
We claim that ρN′:N⟶Nco(h⊗C)⊗Eh is an isomorphism in M(ψ)DC. From the counit property, we know that (idN⊗Dεh⊗Eh)∘ρN=idN. Hence, ρN is a monomorphism and so is ρN′. It remains to show that ρN′(X) is an epimorphism for each X∈Ob(D). For each X∈Ob(D), we define
[TABLE]
Since ρN′ is a morphism of right D-modules, we now have
[TABLE]
This shows that F∘G≅idM(ψ)DC.
∎
Bibliography1
The reference list from the paper itself. Each links out to its DOI / PubMed record.
1AUTHOR = Brzeziński, T., author = Hajac, Piotr M., TITLE = Coalgebra extensions and algebra coextensions of Galois type, JOURNAL = Comm. Algebra, VOLUME = 27, YEAR = 1999, NUMBER = 3, PAGES = 1347–1367,