Blog Archive

Thursday, September 17, 2026

The Duality Behind the Equivalence of Modules and Comodules

From Right Duality to Actions and Coactions

Duality in a monoidal category reverses both the direction of morphisms and the order of tensor factors. Consequently, if a monoid admits a right dual, its right dual naturally becomes a comonoid.

Duality also transforms actions of the monoid into coactions of the corresponding comonoid, while leaving the underlying objects unchanged.

In the monoidal category of endofunctors, these statements become:

  • A right adjoint of a monad naturally carries a comonad structure.

  • The category of algebras for the monad is isomorphic to the category of coalgebras for the corresponding comonad.

We first explain the general monoidal construction, then verify the correspondence between algebras and coalgebras directly.

The terminology requires some care. A monoid is an algebra object in a monoidal category. An algebra for a monad is an object equipped with an action of that monad. For example, algebras for the monad A are left A-modules.

Right Duals and Conventions

Let (M,,1) be a monoidal category. We suppress associativity and unit constraints, interpreting all expressions using monoidal coherence.

A right dual of an object L is an object R equipped with morphisms

ev:LR1,coev:1RL

such that

(evidL)(idLcoev)=idL

and

(idRev)(coevidR)=idR.

These are the snake identities. We use this convention for right duals throughout.

Suppose X and Y have chosen right duals. For a morphism f:XY, its right dual

fr:YrXr

is the composite

YrcoevXidXrXYridfidXrYYridevYXr.

The snake identities imply

(gf)r=frgr,(idX)r=idXr.

There are also canonical identifications

(XY)rYrXr,1r1.

For example, evaluation between XY and YrXr first evaluates the middle pair YYr, then evaluates XXr. Thus, reversing the tensor order is intrinsic to duality; no braiding is required.

There are therefore two reversals to keep track of: duality reverses morphisms and reverses the order of tensor factors.

The Right Dual of a Monoid

Let L be a monoid with multiplication and unit

μ:LLL,u:1L.

Suppose L has a right dual R. Taking right duals of μ and u, define

Δ:Rμr(LL)rRR

and

e:Rur1r1.

We verify that these maps make R a comonoid.

Associativity of μ says

μ(μidL)=μ(idLμ).

Under the canonical duality identifications,

(μidL)r=idRΔ,

and

(idLμ)r=ΔidR.

Taking right duals of the associativity equation therefore gives

(idRΔ)Δ=(ΔidR)Δ.

This is coassociativity.

Similarly, the unit identities

μ(uidL)=idL=μ(idLu)

become

(idRe)Δ=idR=(eidR)Δ.

Hence (R,Δ,e) is a comonoid.

Only the existence of a right dual of L is required. The ambient monoidal category need not be rigid, and R itself need not admit a further right dual.

Duality in the Category of Endofunctors

Consider the monoidal category

(End(C),,IdC),

whose tensor product is composition of functors.

A right dual R of an endofunctor L consists of natural transformations

ε:LRIdC,η:IdCRL

satisfying

(εL)(Lη)=idL,(Rε)(ηR)=idR.

These are exactly the data and triangle identities of an adjunction LR.

Thus we have the following dictionary:

Monoidal notionInterpretation in End(C)
Tensor productComposition of functors
MonoidMonad
ComonoidComonad
Right dualRight adjoint
Right dual of a morphismRight mate of a natural transformation

Consequently, if (L,μ,u) is a monad and LR, the preceding construction equips R with a comonad structure

(R,δ,e),δ:RR2,e:RIdC.

We now verify that this construction also gives an isomorphism

Alg(L)Coalg(R)

preserving underlying objects and morphisms.

Algebras and Coalgebras: A Direct Verification

Fix the adjunction LR, with unit η and counit ε. We continue to write u for the unit of the monad.

These are different natural transformations:

u:IdCL,η:IdCRL.

For f:LXY, its adjoint transpose is

ΦX,Y(f)=R(f)ηX:XRY.

The inverse transpose sends g:XRY to

εYL(g):LXY.

Applying the adjunction twice gives a bijection

ΨX,Y:C(L2X,Y)C(X,R2Y),hR2(h)R(ηLX)ηX.

The induced comonad structure can be written explicitly as

eY=εYuRY,

and

δY=ΨRY,Y(εYμRY).

The latter is the double transpose of

L2RYμRYLRYεYY.

Structure maps. Given a morphism a:LXX, define

b=R(a)ηX:XRX.

Conversely, given b:XRX, define

a=εXL(b):LXX.

These assignments are mutually inverse. Indeed,

εXL(R(a)ηX)=εXLR(a)L(ηX)=aεLXL(ηX)=a,

where the second equality uses naturality of ε, and the last uses a triangle identity.

Similarly,

R(εXL(b))ηX=R(εX)RL(b)ηX=R(εX)ηRXb=b.

Here we use naturality of η and the other triangle identity.

The unit axiom. For corresponding structure maps a and b, we have

eXb=eXR(a)ηX=aeLXηX=aεLXuRLXηX=aεLXL(ηX)uX=auX.

The second line uses naturality of e, the fourth uses naturality of u, and the final line uses a triangle identity.

Therefore,

auX=idXeXb=idX.

The associativity axiom. The algebra associativity axiom compares two morphisms L2XX:

aμX=aL(a).

We transpose each side separately.

Since a=εXL(b), naturality of μ gives

aμX=εXL(b)μX=εXμRXL2(b).

By naturality of Ψ in its source variable and the definition of δ,

ΨX,X(aμX)=ΨRX,X(εXμRX)b=δXb.

For the other side, substitute directly into the formula for Ψ:

ΨX,X(aL(a))=R2(a)R2L(a)R(ηLX)ηX=R2(a)R(ηX)R(a)ηX=R(R(a)ηX)(R(a)ηX)=R(b)b.

The second equality follows from naturality of η applied to a:LXX:

RL(a)ηLX=ηXa.

Since ΨX,X is a bijection,

aμX=aL(a)δXb=R(b)b.

The equation on the right is precisely the coalgebra coassociativity axiom.

Together with the unit calculation, this proves that a defines an L-algebra structure if and only if b defines an R-coalgebra structure.

Morphisms. Suppose (X,a) and (Y,a) correspond to (X,b) and (Y,b). For a morphism f:XY, naturality of the adjunction gives

ΦX,Y(fa)=R(f)b,

and

ΦX,Y(aL(f))=bf.

Therefore,

fa=aL(f)R(f)b=bf.

Thus the same underlying map is an algebra morphism exactly when it is a morphism between the corresponding coalgebras.

The mutually inverse assignments

(X,a)(X,R(a)ηX)

and

(X,b)(X,εXL(b))

act identically on underlying morphisms. They therefore define an isomorphism

Alg(L)Coalg(R).
Modules and Comodules in a Monoidal Category

Return to a general monoidal category M. Let L be a monoid with right dual R, equipped with the comonoid structure constructed above.

The duality data give an adjunction

LR.

Algebras for the monad L are left L-modules, while coalgebras for the corresponding comonad R are left R-comodules. Hence

LMod(M)RComod(M).

Explicitly, a left action a:LXX corresponds to the left coaction

ρ:XcoevidXRLXidRaRX.

The inverse construction is

a:LXidLρLRXevidXX.

More generally, let D be a left M-module category, with action denoted by . The same duality data give

LR,

and consequently identify left L-module objects in D with left R-comodule objects in D.

The object X on which the action is defined does not need to have a dual.

An Example and a Check of Conventions

Let A be an algebra over a field k. On Vectk, there is an adjunction

AHomk(A,).

The action of a left A-module M corresponds to

β:MHomk(A,M),β(m)(a)=am.

The corresponding comonad structure is

eV(f)=f(1A),δV(f)(a)(b)=f(ba).

The counit axiom for β is therefore

1Am=m.

Evaluating its coassociativity axiom at a,bA gives

(ba)m=b(am).

These formulas hold for arbitrary A, with no finite-dimensionality assumption. Here the relevant right dual is the right adjoint of the endofunctor A.

If A is finite-dimensional, then A itself also has a right dual A in Vectk, and

Homk(A,V)AV.

At this point, the order of tensor factors matters.

Using the order-reversing right-dual identification

(AA)rArAr,

the resulting comultiplication satisfies

φ(1)(a)φ(2)(b)=φ(ba).

By contrast, the usual linear-dual coalgebra structure Δstd is characterized by

φ[1](a)φ[2](b)=φ(ab).

The two differ by the symmetry:

Δ=τΔstd.

Thus the construction above directly identifies left A-modules with left comodules over the opposite coalgebra

(A,Δstd)cop.

Applying the symmetry to the two factors of the coaction yields the familiar formulation

AModComod(A,Δstd).

The correspondence between left modules and right comodules is therefore consistent with the left-action/left-coaction formulation above. The difference is accounted for by the reversal of tensor order in right duality.

References
  • P. Etingof, S. Gelaki, D. Nikshych, and V. Ostrik, Tensor Categories, §2.10, especially Definition 2.10.2, Exercises 2.10.4 and 2.10.7, and Proposition 2.10.8. Author's version.

  • G. Böhm, T. Brzeziński, and R. Wisbauer, Monads and comonads in module categories, §§2.2 and 2.6. arXiv:0804.1460.

Popular Posts