Blog Archive

Wednesday, August 26, 2026

Rigid Monoidal Category via Walking Adjunction.

 

Rigid Monoidal Categories: Duality as Adjunction

A useful way to understand rigid monoidal categories is to regard duality as a special case of adjunction.

The central idea is

duality in a monoidal category=adjunction in its one-object bicategory.

The passage from dual objects to ordinary adjoint functors is then explained by the canonical action

F:BC→Cat,X⟼X⊗−.

Thus there are two levels:

X⊣X∨in BC,

and

X⊗−⊣X∨⊗−in Cat.

The second adjunction is obtained from the first by applying the pseudofunctor F.


1. A Monoidal Category as a One-Object Bicategory

Let

(C,⊗,1)

be a monoidal category.

Its delooping

BC

is the bicategory with a single 0-cell

∗.

The correspondence is

CBCobject X1-cell X:∗→∗morphism f:X→Y2-cell f:X⇒YX⊗Ycomposition of 1-cells11∗

Thus an object X∈C becomes a 1-morphism

X:∗→∗

in BC.

Therefore, when we speak of an adjoint of an object X in a monoidal category, what we really mean is an adjoint of the corresponding 1-morphism in BC.


2. The Walking Adjunction

The walking adjunction Adj is the 2-category freely generated by an adjunction

L⊣R.

It has two objects

0,1,

generating 1-morphisms

L:0→1,R:1→0,

and generating 2-morphisms

η:10⇒RL,
ε:LR⇒11,

subject to the triangle identities

(εL)∘(Lη)=1L,
(Rε)∘(ηR)=1R.

Since Adj is freely generated under composition, it also contains composites such as

RL,LR,LRL,RLR,…

together with the 2-morphisms generated from η and ε.

Its universal property is

2-functors Adj→B⟷adjunctions in B.

Thus a map out of the walking adjunction is precisely a realization of an adjunction.


3. Duality as an Adjunction in BC

Suppose X∨ is a right dual of X.

Then there are morphisms

coevR:1→X∨⊗X,

and

evR:X⊗X∨→1,

satisfying the snake identities.

Inside BC, these are precisely the unit and counit of an adjunction

X⊣X∨.

Equivalently, there is a realization of the walking adjunction

D:Adj→BC

such that

D(0)=D(1)=∗,
D(L)=X,D(R)=X∨,

and

D(η)=coevR,D(ε)=evR.

The triangle identities in Adj become the snake identities in C.

Thus

right dual of X=right adjoint of X in BC.

4. Left Duals and Right Duals

In a general monoidal category, left and right duals must be distinguished.

A left dual of X is denoted

∨X

and comes with

evL:∨X⊗X→1,

and

coevL:1→X⊗∨X.

In BC, this is the adjunction

∨X⊣X.

A right dual X∨ comes with

evR:X⊗X∨→1,

and

coevR:1→X∨⊗X,

corresponding to

X⊣X∨.

Hence a rigid object sits in an adjoint triple

∨X⊣X⊣X∨.

5. Left Rigid, Right Rigid, and Rigid

A monoidal category C is left rigid if every object has a left dual:

C is left rigid⟺∀X,∨X⊣X.

Equivalently, every 1-cell in BC has a left adjoint.

Similarly,

C is right rigid⟺∀X,X⊣X∨.

Equivalently, every 1-cell in BC has a right adjoint.

A rigid monoidal category has both:

C is rigid⟺∀X,∨X⊣X⊣X∨.

Thus rigidity can be expressed entirely in bicategorical language:

every 1-cell in BC has both a left and a right adjoint.

6. The Walking Adjoint Triple

For a single right dual, the walking adjunction

L⊣R

is sufficient.

To describe both the left and right dual of a fixed object X, it is more natural to consider the 2-category freely generated by an adjoint triple

L⊣M⊣R.

A realization

Wtri→BC

may send

L↦∨X,
M↦X,
R↦X∨.

Thus the universal adjoint triple

L⊣M⊣R

becomes

∨X⊣X⊣X∨.

For a fixed object, this walking adjoint triple captures both sides of rigidity.

Rigidity of the whole category means that every X∈C can occur as the middle term of such an adjoint triple.


7. The Pseudofunctor F:BC→Cat

The monoidal structure is originally a bifunctor

⊗:C×C→C.

Fixing the first variable gives, for every X∈C, an endofunctor

X⊗−:C→C.

These endofunctors assemble into a pseudofunctor

F:BC→Cat.

It is defined by

F(∗)=C,

and for a 1-cell X:∗→∗,

F(X)=X⊗−.

For a 2-cell

f:X⇒Y,

that is, a morphism f:X→Y in C, we obtain the natural transformation

F(f)=f⊗1:X⊗−⇒Y⊗−.

Compatibility with composition is supplied by the associator:

F(X)∘F(Y)=X⊗(Y⊗−)≅(X⊗Y)⊗−=F(X⊗Y).

Thus

X⟼X⊗−

is the action on 1-cells of the pseudofunctor

F:BC→Cat.

It is induced by the tensor bifunctor

⊗:C×C→C.

This is the left regular action of C on itself.


8. The Walking Adjunction Produces Adjoint Functors

Suppose X∨ is a right dual of X.

The dual pair gives a realization of the walking adjunction

D:Adj→BC

with

L↦X,R↦X∨.

Now compose this with

F:BC→Cat.

We obtain

Adj→DBC→FCat.

Since Adj is the walking adjunction, the composite

F∘D:Adj→Cat

is precisely the data of an ordinary adjunction of functors.

The generating 1-morphisms are sent to

F(D(L))=F(X)=X⊗−,

and

F(D(R))=F(X∨)=X∨⊗−.

Hence

X⊗−⊣X∨⊗−.

Schematically,

Adj⟶BC⟶Cat

gives

L⊣R
⇓
X⊣X∨
⇓
X⊗−⊣X∨⊗−.

Thus the adjunction of tensor functors is not an independent fact. It is the image of the bicategorical adjunction

X⊣X∨

under the pseudofunctor

F:BC→Cat.

The walking-adjunction principle is exactly what turns the dual pair into an ordinary pair of adjoint functors.


9. The Tensor-Hom Adjunction

The adjunction

X⊗−⊣X∨⊗−

means that there is a natural isomorphism

HomC(X⊗A,B)≅HomC(A,X∨⊗B).

Given

f:X⊗A→B,

the corresponding morphism is

A≅1⊗A→coevR⊗1X∨⊗X⊗A→1⊗fX∨⊗B.

Conversely, given

g:A→X∨⊗B,

we obtain

X⊗A→1⊗gX⊗X∨⊗B→evR⊗1B.

These constructions are mutually inverse because the evaluation and coevaluation satisfy the snake identities.

Thus the snake identities are precisely the triangle identities of the corresponding adjunction.


10. Left Closed and Right Closed Monoidal Categories

There are one-sided notions of closedness.

Using one common convention, C is called left closed if, for every X,

X⊗−

has a right adjoint:

X⊗−⊣[X,−]L.

Thus

Hom(X⊗A,B)≅Hom(A,[X,B]L).

Similarly, C is called right closed if

−⊗X

has a right adjoint:

−⊗X⊣[X,−]R,

so that

Hom(A⊗X,B)≅Hom(A,[X,B]R).

If both exist, the monoidal category is biclosed.

The terminology left closed'' andright closed'' varies between authors, so it is often safer to specify explicitly which tensor functor is being considered.


11. Why Rigidity Implies Closedness

Suppose X has a right dual

X⊣X∨

in BC.

Applying

F:BC→Cat

gives

X⊗−⊣X∨⊗−.

Therefore the required internal Hom is explicitly

[X,B]L≅X∨⊗B.

Hence

right rigid⟹left closed.

Similarly, if

∨X⊣X,

then using the right regular action gives

−⊗X⊣−⊗∨X,

and hence

[X,B]R≅B⊗∨X.

Therefore

left rigid⟹right closed.

Combining the two,

rigid⟹biclosed.

12. Finite-Dimensional Vector Spaces

The basic example is

Vectkfd.

For a finite-dimensional vector space X,

X∨=X∗=Homk(X,k).

Since X is dualizable,

X⊗−⊣X∨⊗−.

Therefore

[X,Y]≅X∨⊗Y.

In ordinary linear algebra notation,

Homk(X,Y)≅X∨⊗Y.

The natural map

X∨⊗Y→Homk(X,Y)

is

φ⊗y⟼(x↦φ(x)y).

If

e1,…,en

is a basis of X, with dual basis

e1,…,en,

then a linear map

f:X→Y

corresponds to

∑iei⊗f(ei).

In particular,

End(X)≅X∨⊗X,

and

X∨≅[X,1].

13. Left and Right Duals as an Adjoint Chain

In a rigid category we have

∨X⊣X⊣X∨.

Since adjoints are unique up to canonical isomorphism,

∨(X∨)≅X

and

(∨X)∨≅X.

This corresponds to taking an adjoint and then taking the adjoint of the opposite kind:

X→right dualX∨→left dualX.

By contrast,

X⟼X∨⟼X∨∨

takes the right dual twice.

There is no general reason for this same-sided double dual to return X.

Hence

∨(X∨)≅X

and

X∨∨≅X

express different phenomena.

 

No comments:

Post a Comment

Popular Posts