Penneys Math 8110, Higher Linear Algebra 2-categories

TODO: lead in
5.1. 2-categories.
Definition 5.1.1. A 2-category C consists of
• A collection of objects, a.k.a. 0-morphisms; we write c ∈ C to denote c is an object
of C;
• For each a, b ∈ C, a hom category C(a → b). Objects of C(a → b) are called
1-morphisms. We write a Xb ∈ C(a → b) or X : a → b to denote that X is a 1-
morphism from a to b. Morphisms in C(a → b) are called 2-morphisms. We write
f ∈ C(a Xb ⇒ a Yb ) or f : X ⇒ Y to denote that f is a 2-morphism from X to Y .
• For each a, b, c ∈ C, a 1-composition functor. Based on the 2-category, the typical
convention for this 1-composition may vary between left-to-right and right-to-left.
We use two different notations depending on our choice of convention, so that the
reader may infer the direction of composition directly from the notation.
⊗ = ⊗b : C(a → b) × C(b → c) → C(a → c) (left-to-right)
◦ = ◦b : C(b → c) × C(a → b) → C(a → c) (right-to-left)
We will typically choose the first, which is usually used for algebras, bimodules, and
intertwiners, and we will use the second for 2-categories of categories, functors, and
natural transformations.
This functor necessarily satisfies the exchange relation
(f ⊗idZ )◦(idW ⊗g) = (idX ⊗g)◦(f ⊗idY ) ∀ f ∈ C(a Wb → a Xb ), ∀ g ∈ C(b Yc → b Zc ).
• For each a Xb , b Yc , c Zd , an associator isomorphism
αX,Y,Z : X ⊗b (Y ⊗c Z) ⇒ (X ⊗b Y ) ⊗c Z.
These associator isomorphisms must be natural in each variable and satisfy the ob-
vious pentagon axiom.
• For each c ∈ C, there is a unit 1-morphism 1c ∈ C(c → c), along with unitor
isomorphisms ρcY : Y ⊗c 1c ⇒ Y for all Y ∈ C(b → c) for all b ∈ C, and λcZ : 1c ⊗c Z ⇒
Z for all Z ∈ C(c → d) for all d ∈ C. Again, these unitors must be natural in each
variable and satisfy the obvious triangle axiom.
A 2-category is called strict if all associators and unitors are identity 2-morphisms. A 2-
category is called linear if all 2-morphism spaces C(a Xb ⇒ a Yb ) are finite dimensional complex
vector spaces, and all composition functors are bilinear.
Warning 5.1.2. Sometimes in the literature, 2-category means strict 2-category, and the
fully weak notion is called a bicategory. However, 2-category is clearly the better name, and
the fully weak notion is clearly the better notion, so the better notion should get the better
name. We can then add adjectives for more strict notions.
Example 5.1.3. The 2-category Cat, the strict 2-category of categories, functors, and nat-
ural transformations. In these notes, we will usually write Cat for the strict 2-category of
linear categories and linear functors.
1
Example 5.1.4. The 2-category 2Vec is the strict 2-category of finite semisimple categories,
linear functors, and natural transformations.
Example 5.1.5. There is a 2-category MonCat of monoidal categories, monoidal functors,
and monoidal natural transformations.
Example 5.1.6. There is a 2-category of topological spaces, continuous maps, and homotopy
classes of homotopies between continuous maps.
Example 5.1.7. There is a 2-category Alg of complex algebras, bimodules, and intertwin-
ers. Usually, we will write Alg for the 2-category of semisimple finite dimensional complex
algebras.
Example 5.1.8. Given a monoidal category C, we get a 2-category BC with exactly one
object ∗ called the delooping of C. We simply define the hom category BC(∗ → ∗) := C with
the obvious 1-composition functor, associator, unit, and unitors.
Conversely, given a 2-category C, picking any object c ∈ C, the loop space Ωc C := EndC (c)
is a monoidal category with the obvious tensor product functor, associator, unit 1c , and
unitors.
Corollary 5.1.9. If C is a 2-category, the monoid EndC (1C ) is always commutative.
Definition 5.1.10. Suppose C is a 2-category and a, b ∈ C. The linking category is the
monoidal category
   
C(a → a) C(a → b) C(a → a) C(b → a)
L⊗ (a, b) := L◦ (a, b) :=
C(b → a) C(b → b) C(a → b) C(b → b)
whose objects are formal matrices of objects and whose morphisms are formal matrices
of morphisms between objects respectively. When 1-composition is left-to-right, the first
definition must be used, and when 1-composition is right-to-left, the second definition must
be used. The unit is the formal matrix
 
1a
,
1b
and we leave it to the reader to write down formulas for the associator and unitors. Similarly,
we can define the n-fold linking category L(a1 , . . . , an ) for any a1 , . . . , an ∈ C.
Definition 5.1.11. A 2-category C is called linear if every 2-morphism space C(a Xb ⇒ a Yb )
is a finite dimensional vector space, i.e., an object in Vec.
Generally speaking, if a certain property ‘P’ holds for every hom 1-category of C, we call
C locally ‘P.’ For example, we call a linear 2-category locally Cauchy complete (or locally
(finitely) semisimple) if all hom 1-categories are Cauchy complete (respectively (finitely)
semisimple).
Construction 5.1.12. Suppose C is a linear 2-category. We define a new locally Cauchy
complete linear 2-category called ¢1 (C) by replacing all hom 1-categories by their Cauchy
completions. In more detail, ¢1 (C) has the same objects as C, and we define the hom 1-
categories by
¢1 (C)(a → b) := ¢(C(a → b)).
2
To define the composition functor, first observe that we may view 1-composition as a linear
functor on the Delinge product (as opposed to a bilinear functor on the product)
−⊗−
C(a → b) ⊠ C(b → c) −−−→ C(a → c).
Now observe that we have an equivalence of categories from the Deligne product of Cauchy
complete categories (which is the Cauchy completion of the usual Deligne product) to the
Cauchy completion of the Deligne product of linear categories
¢1 (C)(a → b) ⊠ ¢1 (C)(b → c) ∼
= ¢(C(a → b) ⊠ C(b → c)).
By the universal property of Cauchy completion, 1-composition thus extends to ¢1 (C):
∼
¢(C(a → b) ⊠ C(b → c)) =
¢1 (C)(a → b) ⊠ ¢1 (C)(b → c)
∃¢(−⊗−)
−⊗−
C(a → b) ⊠ C(b → c) C(a → c) ¢(C(a → c)).

The associators and unitors for ¢1 (C) are defined similarly to those built for Add(C) and
Idem(C) [[]], as ¢ = Idem ◦ Add on linear categories.

Exercise 5.1.13. Define analogous notions of Add1 (C) for a linear 2-category which replaces
each hom 1-category with its additive envelope and Idem1 (C) for a 2-category which replaces
each hom 1-category with its idempotent completion.
5.2. Graphical calculus for 2-categories and adjoints. Similar to monoidal categories,
2-categories admit a graphical calculus of string diagrams which are dual to pasting diagrams.
In a pasting diagram, one represents objects as vertices, 1-morphisms as arrows, and 2-
morphisms as 2-cells. In the string diagram calculus, we represent objects by shaded regions,
1-morphisms by (oriented) strands between these regions, and 2-morphisms by coupons.
Z
Z

f : a X ⊗b Yc ⇒ a Zc ⇝ a f c ⇝ a f c
X b Y b
X Y

As before, we suppress all associators and unitors; 1-composition is denoted by horizontal

juxtaposition, and 2-composition is denoted by stacking of diagrams.
As in a monoidal category, we can define the notion of dual and predual for 1-morphisms.
If a, b ∈ C and a Xb is a 1-morphism, a dual is a 1-morphism b Xa∨ together with maps

coevX = X X∨ and ϵ= X∨ X where X∨ = X ∨, X =X

satisfying the zig-zag/snake equations. There is a similar notion of predual. However, as we

continue our exploration into higher categories, we will reserve the word ‘dual’ for objects,
and we will use the term adjoint for a dual of a 1-morphism. This term is borrowed from
the 2-category Cat of categories.
3
Definition 5.2.1. Suppose C, D are categories and F : C → D and G : D → C are functors.
We say that F is left adjoint to G, equivalently G is right adjoint to F , denoted F ⊣ G, if
there is a family of isomorphisms
D(F (c) → d) ∼
= C(c → G(d)). (5.2.2)
which is natural in c ∈ C and d ∈ D.
We will show in Proposition 5.2.8 below that the unit and counit witness that G is the
dual of F in Cat, and F is the predual of G. (Since 1-composition in Cat is left-to-right, the
convention for duals switches!)
Example 5.2.3. There are many free/forget adjoints across mathematics. Examples in-
clude:
• Forget : Vec → Set which forgets the vector space structure and Free : Set → Vec
defined by S 7→ C[S]
• TODO: more examples
Example 5.2.4. When b ∈ C is dualizable, b∨ ⊗ − : C → C is left adjoint to b ⊗ − : C → C
by Frobenius reciprocity [[]]. Similarly, − ⊗ b is left adjoint to − ⊗ b∨ .1
Notation 5.2.5. Let ∗ denote the terminal category, which has one object ∗ and one 1-
morphism id∗ . When considering linear categories, ∗ denotes the category with one object
∗ and End(∗) = C id∗ , which is no longer terminal (the terminal linear category satisfies
End(∗) = 0).
There is a canonical equivalence C ∼
= Fun(∗ → C) given by a 7→ (∗ 7→ a) and (f : a → b) 7→
(f∗ : (∗ 7→ a) ⇒ (∗ 7→ b)). Denoting ∗ by the empty shaded region in the graphical calculus
for Cat, we see that we can represent objects c ∈ C by strings and morphisms by coupons
with a C-shading on the right (recall we compose left-to-right in Cat). We then apply the
functor F : C → D by adding an F strand to the left of the c-strand.
b
a ↔c∈C f ↔ (f : a → b) F a ↔ F (a) ∈ D
a

Natural transformations ρ : F ⇒ G are the represented by coupons between the F, G strings,

and naturality is represented by the exchange relation:
G b
ρ f
F b = ρb ◦ F (f ) = G(f ) ◦ ρa = G a .
f ρ
a F

Definition 5.2.6. Suppose C, D are categories and F : C → D and G : D → C are functors

with F ⊣ G. We say f : F (c) → d and g : c → G(d) are mates if they map to each

1Some people call a dual a left (or right) dual and a predual a right (or left) dual (respectively). However,
it is not clear whether a dual should be a left or a right dual, as tensoring on the left with b∨ is a left adjoint
and tensoring on the right with b∨ is a right adjoint.
4
 other under the natural isomorphism (5.2.2). The unit of the adjunction is the natural
transformation η : idD ⇒ GF given by
ηc := mate(idF (c) ) ∈ C(c → GF (c)) ∼
= D(F (c) → F (c)),
and the counit of the adjunction is the natural transformation ϵ : F G ⇒ idC given by
ϵd := mate(idG(d) ) ∈ D(F G(d) → d) ∼
= C(G(d) → G(d)).
Lemma 5.2.7. The operations of taking mate are natural with respect to pre-composition
and post-composition by another morphism:
(mate1) mate(f2 ◦ f1 ) = G(f2 ) ◦ mate(f1 ) for all f1 : F (c) → d1 and f2 : d1 → d2 .
(mate2) mate(g2 ◦ g1 ) = mate(g2 ) ◦ F (g1 ) for all g1 : c1 → c2 and g2 : c2 → G(d).
Proof. We prove the first, and the second is similar. By naturality of the adjunction isomor-
phism (5.2.2), the following diagram commutes.
∼
=
D(F (c) → d1 ) C(c → G(d1 ))
f2 ◦− G(f2 )◦−
∼
=
D(F (c) → d2 ) C(c → G(d2 )).

Going down and then left gives mate(f2 ◦ f1 ), and goin right and then down gives G(f2 ) ◦
mate(f1 ). □

The next proposition shows that adjoints are the same thing as duals for 1-morphisms
(functors) in Cat.
Proposition 5.2.8. Suppose C, D are categories and F : C → D and G : D → C are
functors.
(1) If there exist natural transformations evF : F ◦ G ⇒ idD and coevF : idC ⇒ G ◦ F
satisfying the snake equations (so G ∼ = F ∨ ), then F ⊣ G with adjunction natural
isomorphism (5.2.2) given by the Frobenius reciprocity isomorphisms from (??) for
2-categories.
D(F (c) → d) ∼
= C(c → F ∨ (d))
d d
F∨
f 7→ f
F c c

d F∨ d
g ←[ g
F
c c

(2) If F ⊣ G, then the unit and counit satisfy the snake equations and thus can be
represented by

η= G F and ϵ= F G where F = F, G = G.
5
Proof. Statement (1) is immediate from naturality of evF and coevF . To prove statement
(2), we must prove that
G
F d = G d ⇐⇒ G(ϵd ) ◦ ηG(d) = idG(d) ∀d ∈ D
G

F
G c = F c ⇐⇒ ϵF (c) ◦ F (ηc ) = idF (c) ∀ c ∈ C.
F

The first is (mate1) with f1 = idF G(d) and f2 = ϵd , and the second is (mate2) with g1 = ηc
and g2 = idGF (c) . □
Exercise 5.2.9. Modify [[]] to prove that every equivalence a ≃ b in a 2-category can be
modified to an adjoint equivalence, i.e., there are 1-morphisms a Xb and b Ya and isomorphisms
η : 1a ⇒ X ⊗b Y and ϵ : Y ⊗b X → 1b which satisfy the snake equations.
Deduce that every equivalence of categories F : C → D can be augmented to an adjoint
equivalence, i.e., there is an equivalence G : D → C such that F ⊣ G.
Definition 5.2.10. A linear 2-category C is called pre-semisimple if it is rigid and locally
semisimple, equivalently every n-fold linking category is a (semisimple) multitensor category.2
5.3. Higher morphisms between 2-categories.
Definition 5.3.1. A 2-functor F : C → D between 2-categories consists of
• an assignment of an object F (c) to each object c ∈ C,
• a functor Fa,b : C(a → b) → D(F (a) → F (b)),
• for all objects c ∈ C, a unitor 2-isomorphism Fc0 ∈ D(1F (c) → F (1c )), and
2
• for all 1-morphisms a Xb , b Yc ∈ C, a compositor/tensorator 2-isomorphism FX,Y ∈
D(F (X) ⊗F (b) F (Y ) ⇒ F (X ⊗b Y ))
subject to the following axioms:
2
• (naturality) FX,Y is natural in X and Y ,
• (associativity) For all a Xb , b Yc , and c Zd in C, the following diagram commutes:
2
idF (X) ⊗FY,Z 2
FX,Y ⊗Z
F (X) ⊗ (F (Y ) ⊗ F (Z)) F (X) ⊗ F (Y ⊗ Z) F (X ⊗ (Y ⊗ Z))
αD
F (X),F (Y ),F (Z) F (αC
X,Y,Z )
2
FX,Y ⊗idF (Z) 2
FX⊗Y,Z
(F (X) ⊗ F (Y )) ⊗ F (Z) F (X ⊗ Y ) ⊗ F (Z) F ((X ⊗ Y ) ⊗ Z)
• (unitality) for all a, b ∈ C and a Xb ∈ C(a → b),
F (a)
λF (X) ρD
F (X)
1F (a) ⊗ F (X) F (X) F (X) ⊗ 1F (b) F (X)
Fa1 ⊗idF (X) F (λa
X) idF (X) ⊗Fb1 F (ρbX )
2
F12a ,X FX,1
b
F (1a ) ⊗ F (X) F (1a ⊗ X) F (X) ⊗ F (1b ) F (X ⊗ 1b )
A 2-functor is called:
2Our definition of a pre-semisimple 2-category differs slightly from that of presemisimple 2-category in
[DR18, Def. 1.2.7], but our easier definition will still complete to a semisimple 2-category later on in §[[]].
6
 • fully faithful if each functor Fa,b is an equivalence, and
• essentially surjective if every object d ∈ D is equivalent to an object of the form F (c)
for some c ∈ C.
• an equivalence if it is fully faithful and essentially surjective (cf. [JY21, Thm. 7.4.1]).
A 2-functor F : C → D between linear 2-categories is called linear if each functor Fa,b is
linear.
Example 5.3.2. The additive envelope Add is a 2-functor on the 2-category of linear cate-
gories. TODO:
Example 5.3.3. Idempotent completion Idem is a 2-functor on the 2-category Cat categories
which preserves the 2-subcategory of linear 2-categories. TODO:
Example 5.3.4. Composing the last two examples, Cauchy completion ¢ is a 2-functor on
the 2-category of linear categories.
Exercise 5.3.5. A 2-functor is called strict if the unitors and tensorators are identities.
Show that strict 2-categories and strict 2-functors form a 1-category.
Exercise 5.3.6 (⋆). Prove that every 2-category is equivalent to a strict 2-category.
Hint: Find a fully faithful 2-functor よ : C → Hom(C mp → Cat). (See [JY21, §8] for more
details.)
Notation 5.3.7. For 2-categories C, D we have a strict 2-category of 2-functors Hom(C →
D). We represent objects (2-functors) in this 2-category by unshaded regions with textures,
e.g.:
=F =G =H
We represent 2-transformations (see [[]] below) by textured strands between these textured
regions:
=ρ:F ⇒G =σ:G⇒H
We represent 2-modifications (see [[]] below) by coupons between textured strands.

m = m : F ρG ⇛ F σ G

Remark 5.3.8. We use the ⊗ convention for 1-composition in the 2-category Fun(C → D),
as the 1-morphisms are 2-transformations. Note that we do not compose functors in Fun(C →
D) as they are objects, so no confusion can arise with the ◦ convention for 1-composition.
Notation 5.3.9. We represent a 2-morphism in D in the image of a 2-functor F : C → D
using an overlay graphical calculus, which was described in [CP22, §2.1]. We apply a 2-
morphism m : F ρG ⇛ F σG from Fun(C → D) to an object a ∈ C to obtain the 2-morphism
ma : ρa ⇒ σa in D:
 
σ σa
!
 
= ma
 m 
  = F (a), = G(a).
  a
ρ ρa
7
Given a 1-morphism a Xb in C, there are 4 basic diagrams which one could obtain from
overlaying m:
 
F (X) ρb
 
 
    
 
F (X) ρb
 ρa G(X)

   
could represent one of (5.3.10)

    
   ρa G(X) 
 
ρ X  
 
 
 
F (X) ρb ρa G(X)

These crossings on the right hand side are data of a 2-transformation ρ : F ⇒ G, which
must satisfy various coherence conditions. For example, the two crossings on the off-diagonal
should be each others’ inverses under vertical composition, making these 4 diagrams into a
system of matrix units. We refer the reader to [[]] below for further details.
Remark 5.3.11. We will see in [[]] that 2-categories form a 3-category called 2Cat. While
this 3-category is not strict, it does have the nice property that 1-composition is strictly
associative. There is a semistrict notion of 3-category called a Gray-category, and every
3-category is equivalent to a Gray-category by [Gur13].
Now there is a 3D graphical calculus for Gray-categories [BMS12, Bar14], which can be
applied to any 3-category by [Gut19]. Our overlay graphical calculus is an example of this
3D graphical calculus in 2Cat. We sketch this below, and we leave a rigorous proof to the
interested reader.
Indeed, given a 2-category C ∈ 2Cat, similar to Notation 5.2.5 above, we may identify
C = Fun(∗ → C) where ∗ is the terminal 2-category with one object ∗, one 1-morphism 1∗ ,
and one 2-morphism id1∗ (In the linear setting, Hom(1∗ → 1∗ ) = C id1∗ ). This identification
allows us to identify the internal 2D string diagrammatic calculus for C with the external
2D string diagrammatic calculus for Fun(∗ → C) as a hom 2-category of 2Cat. Since we
may identify a 2-functor F : C → C with the 2-functor F ◦ − : Fun(∗ → C) → Fun(∗ → D)
given by post-composition with F , and similarly for transformations and modifications, our
overlay graphical calculus is exactly stacking of 2D sheets in the 3D graphical calculus for
2Cat.
 
σ
f
 
 
 m 
f = m
⇛

 
  ∗
ρ C
→

D
⇒
In order to interpret each overlay diagram as a 2-morphism in D, one must require the
strings and coupons of our C-diagram and our Fun(C → D) diagram not overlap, except
at finitely many points where strings can cross transversely. The axioms of 2-functor, 2-
transformation, and 2-modification then ensure than any two ways of resolving non-generic
intersections agree. We saw one such example in (5.3.10) above. If we also included a
8
modification m : ρ ⇛ σ there, we would
  σa G(X) σa G(X)
σ 
   ma

 m 
  could represent or ,
mb
  
ρ X

F (X) ρb F (X) ρb

and requiring these diagrams to be equal produces the modification coherence axiom in [[]]
below. The data of the crossing ρX for the 2-transformation ρ : F ⇒ G may be interpreted
as an interchanger in 2Cat, which arise from resolving two stacked 2D diagrams in 2Cat.
Definition 5.3.12. Suppose C, D are 2-categories and F, G : C → D are 2-functors. A
2-transformation ρ : F ⇒ G consists of
• for every c ∈ C, a 1-morphism ρc : F (c) → G(c), and
• for every 1-morphism a Xb ∈ C, an invertible 2-morphism ρX : F (X) ⊗F (b) ρb ⇒
ρa ⊗G(a) G(X)
ρa ρa G(X)
F (a) G(a)
ρX
F (X) G(X) = ρX
ρb
F (b) G(b) F (X) ρb

such that the following coherence axioms holds.

ρa G(Y ) ρa G(X)

G(f )
• (naturality) for all f ∈ C(a Xb ⇒ a Yb ), =
F (f )

F (X) ρb F (X) ρb

ρa G(1a ) ρa G(1a )

G0a
• (unitality) for all a ∈ C, =
Fa0
1F (a) ρa 1F (a) ρa

ρa G(X⊗Y )
ρa G(X⊗Y )
G2X,Y

• (monoidality) for all a Xb , b Yc in C, =

2
FX,Y

F (X)F (Y ) ρc
F (X)F (Y ) ρc
9
Definition 5.3.13. Suppose C, D are 2-categories, F, G : C → D are 2-functors, and ρ, σ :
F ⇒ G are 2-transformations. A 2-modification m : ρ ⇛ σ consists of a 2-morphism
mc : ρc ⇒ σc for all c ∈ C such that for all 1-morphisms X ∈ C(a → b),
σa
σa G(X) σa G(X)

⇒
ma
ρa σz
F (a) G(a) F (a) G(a) ma
ρX σX
F (X) G(X) = F (X) G(X) =
ρb σb
mb
F (b) G(b) F (b) G(b)

⇒
mb
ρb F (X) ρb F (X) ρb

5.4. Direct sums.

Definition 5.4.1. Suppose C is a locally additive linear 2-category. A zero object in C is
an object 0 ∈ C such that EndC (0) is the terminal category, i.e., it has exactly one object 10
and one morphism id10 .
Given a, b ∈ C, a direct sum is an object a ⊞ b ∈ C and 1-morphisms Ia : a → a ⊞ b,
Ib : b → a ⊞ b, Pa : a ⊞ b → a, and Pb : a ⊞ b → b, satisfying the following axioms (in
right-to-left convention):
(⊞1) Pa ◦ Ia ∼= 1a and Pb ◦ Ib ∼= 1b ,
(⊞2) Ia ◦ Pa ⊕ Ib ◦ Pb ∼
= 1a⊞b , and
(⊞3) Pb ◦ Ia and Pa ◦ Ib are both zero objects in their respective hom 1-categories.
Note that when we write 1-composition from left-to-right, these relations will appear trans-
posed.
Remark 5.4.2. When C is locally semisimple, (⊞3) is superfluous. Indeed, when S is a
semisimple category, s ∼
= s ⊕ t implies t ∼
= 0. Hence
Pb ∼
= Pb ◦ Ia ◦ Pa ⊕ Pb ◦ Ib ◦ Pb ∼
= Pb ◦ Ia ◦ Pa ⊕ Pb
implies Pb ◦ Ia ◦ Pa ∼
= 0. Now pre-compose with Ia to see that Pb ◦ Ia ∼
= 0. Similarly,
∼
Pa ◦ Ib = 0.
Note here that the data of the isomorphisms at height 2 are not part of the data of the
direct sum. Direct sums of a, b ∈ C form a contractible space. Indeed, given another direct
sum a ⊞′ b, we have a canonical equivalence 1-morphism Ia ◦ Pa′ ⊕ Ib ◦ Pb′ : a ⊞′ b → a ⊞ b
with inverse Ia′ ◦ Pa ⊕ Ib′ ◦ Pb : a ⊞ b → a ⊞′ b. This equivalence is unique up to canonical
natural isomorphism, so there is a contractible 2-groupoid (2-category with all morphisms
invertible) of direct sums.
Definition 5.4.3. We call a locally additive linear 2-category additive if it admits a zero
object and all direct sums.
Exercise 5.4.4. Show that linear 2-functors preserve direct sums (when they exist).
We now construct the additive envelope of a 2-category.
Construction 5.4.5. Suppose C is a linear 2-category. By replacing C with ¢1 (C) via Con-
struction 5.1.12, we may assume C is locally Cauchy complete (or at least locally additive).
We define a 2-category Add(C) as follows.
10
 • Objects are formal tuples (aj )nj=1 of objects in C.
• Hom 1-categories are formal matrix categories which are defined similar to linking
categories.

C(a1 → b1 ) · · · C(a1 → bn )
 

Add(C)((aj )nj=1 → (bi )m .. ..

i=1 ) = . .
 
C(am → b1 ) · · · C(an → bm )

• 2-composition happens component-wise.

• 1-composition is given by the usual matrix product formula, only taking direct sums.
If X = (Xij ∈ C(ai → bj )) and Y = (Yjk ∈ C(bj → ck )) where i = 1, . . . , m,
j = 1, . . . , n, and k = 1, . . . , p, then
n
M
(X ⊗(bj ) Y )ik := Xij ◦ Yjk ∈ C(ai → ck ).
j=1

A similar formula is used for 2-morphisms, and one can see − ⊗ − is a functor.
• The associator αX,Y,Z : X ⊗(bj ) (Y ⊗(ck ) Z) ⇒ (X ⊗(bj ) Y ) ⊗(ck ) Z is a formal matrix
of associators whose i, ℓ-entry is given by
p
n M p
n M
M M
αXij ,Yjk ,Zkℓ : Xij ⊗bj (Yjk ⊗ck Zkℓ ) =⇒ (Xij ⊗bj Yjk ) ⊗ck Zkℓ .
j=1 k=1 j=1 k=1

Unitors are defined similarly.

Observe that Add(C) comes equipped with a fully faithful 2-functor ι : C ,→ Add(C) given by
a 7→ (a), a Xb 7→ (X), and (f : a Xb ⇒ a Yb ) 7→ (f ).

Lemma 5.4.6. Add(C) is additive.

Proof. Since each C(aj → bi ) is additive, so is

M
Add(C)((aj )nj=1 → (bi )m
i=1 ) = C(aj → bi ),
i,j

so Add(C) is locally additive.

Now consider objects c1 = (aj )m n
j=1 and c2 = (aj )j=m+1 in Add(C). We define their direct
sum c1 ⊞ c2 as the tuple (aj )nj=1 . For i = 1, 2, we define Ii : ci → c1 ⊞ c2 and Pi : c1 ⊞ c2 → ci
as matrices of objects whose entries are either 1aj or 0 depending on the indices similar to
Lemma [[]] for 1-categories. The rest of the verification is left to the reader. □

Our next task is to prove the universal property of the additive envelope.

Proposition 5.4.7. For every additive 2-category D, precomposition with the canonical
inclusion C ,→ Add(C) gives an equivalence
∼
=
ι∗ : Fun(Add(C) → D) −
→ Fun(C → D).
11
Proof. Suppose F : C → D is a linear 2-functor. We define a linear 2-functor Add(C) → D
by setting
Add(F )((aj )nj=1 ) := ⊞nj=1 F (aj )
M
Add(F )((Xij )) := Pi ⊗ai F (Xij ) ⊗aj Ij
i,j
X
Add(F )((fij )) := ιij · F (fij ) · πij
i,j

where the Pi , Ij witness ⊞nj=1 F (aj ) as a 2-direct sum and ιij , πij witness Add(F )((Xij ))
as an ordinary 1-direct sum. The compositor Add(F )2(Xij ),(Yjk ) is made from the structure
isomorphisms Ij ⊗ Pj ∼ = idF (aj ) and associators and unitors. Clearly F = Add(F ) ◦ ι, so ι∗ is
essentially surjective.
TODO: 2-transformations and 2-modifications □

5.5. Algebras: higher idempotents. For this section, we fix a 2-category C.

Definition 5.5.1. Suppose a ∈ C. A pair (a Aa , m : A ⊗a A ⇒ A) is called an algebra if the
following associativity axiom is satisfied:
idA ⊗m
A ⊗ (A ⊗ A) A⊗A
m
αA,A,A
m A ⇝ = ; = m.
m⊗idA
(A ⊗ A) ⊗ A A⊗A

An algebra (a Aa , m) is called
• unital if there is a 2-morphism i : 1a ⇒ A such that the following unitality axioms
are satisfied:
(λcA )−1 (ρcA )−1
1a ⊗ A A A ⊗ 1a
i⊗idA idA idA ⊗i ⇝ = = ; = i.
m m
A⊗A A A⊗A
A unital algebra is called connected if dim(C(1a ⇒ A)) = 1.
• separable if the multiplication map splits as an A − A bimodule map, i.e., there is a
map ∆ : A → A ⊗ A such that
– (m splits) = where =∆

– (as an A − A bimodule) = =

A triple (A, m, ∆) consisting of a separable algebra (A, m) equipped with a separator ∆ is

called a condensation algebra [GJF19].
A quintuple (A, m, i, ∆, ε) is called a Frobenius algebra if
• (A, m, i) is a unital algebra,
12
 • (A, ∆, ε) is a co-unital coalgebra. Coassociativity is the folllowing axiom:
A ⊗ (A ⊗ A) idA ⊗∆
A⊗A
αA,A,A ∆ A ⇝ = ; = ∆.
∆
(A ⊗ A) ⊗ A ∆⊗idA
A⊗A
We leave the co-unital axiom to the reader.
• ∆ is an A − A bimodule map.
There is also a notion of algebra object in a monoidal category C; it is equivalent to an
algebra in BC.
Remark 5.5.2. The most natural settings in which to work are condensation algebras,
Frobenius algebras, and unital separable algebras. Working with separable algebras which
are non-unital, but not equipped with a particular splitting, is not a well-behaved notion.
Facts 5.5.3. We now give a list of basic facts about algebras.
(A1) If an algebra (A, m) is unital, then its unit is unique.
(A2) If (A, m, ∆) is a condensation algebra, then ∆ is automatically co-associative. TODO:
 
(A3) If (Aa , m, i, ∆, ε) is a Frobenius algebra, then A, , is a dual for A.
(A4) Suppose (A, m, i) is an algebra in a multitensor category and ε : A → 1C is a fixed
non-zero map. The following are equivalent:
• The pairing is non-degenerate, i.e., TODO: is invertible
• There is a ∆ : A → A ⊗ A giving a Frobenius algebra structure
Indeed, TODO:
The proof above proves that ∆ : A → A ⊗ A above is unique as it can be expressed
in terms of TODO:
(A5) When C is a pivotal multitensor category, we call a Frobenius algebra symmetric if
TODO:
(A6) If C is a pivotal tensor category and (A, m, i) is a connected separable algebra such
that dimL (A) ̸= 0 ̸= dimR (A), then A can be canonically equipped with the structure
of a symmetric Frobenius algebra. TODO:
(A7) Given an algebra (A, m), (A∨ , m∨ ) is also an algebra. Moreover, (A, m) is unital if
and only if (A∨ , m∨ ) is with unit i∨ . If ∆ makes (A, m) into a condensation algebra,
then ∆∨ makes (A∨ , m∨ ) into a condensation algebra. An analogous statement holds
for Frobenius structure as well.
Exercise 5.5.4. Find a complete characterization of unital algebras in Set and in Cat.
Exercise 5.5.5. Show that a unital algebra in Vec is separable if and only if it is semisimple.
Example
L 5.5.6. We classify connected unital separable algebras in Vec(G, ω). Suppose
A = ∼
Ag ∈ Vec(G, ω) is a connected unital algebra with unit i : C = Ae ⊂ A and
multiplication m : A ⊗ A → A. We will use the spherical structure on Vec(G, ω) in which all
quantum dimensions of simples are equal to 1.
First, we claim that each non-zero f ∈ Ag is left invertible. Indeed, we may view Ag ∼ =
−1
Hom(Cg → A) by Yoneda, and f non-zero means it has a left-inverse f : A → Cg such
13
that f −1 · f = idCg . Since A is connected and separable and dA ̸= 0, there is a canonical
symmetric Frobenius algebra structure on A by (A6) which satisfies ε · i = dA . We calculate
g
−1 g
f
−1 f
1= g = A = f A
g
f
g

The diagram on the right is the product of the element (f −1 )∨ ∈ Ag−1 times the element
f ∈ Ag in A viewed as an element in Ae ∼ = Hom(Ce → A), post-composed with ε. Since it
is non-zero, we conclude that a scalar multiple of (f −1 )∨ is a left-inverse for f in A.
We now see that each Ag is either 0 or 1-dimensional, and dim(Ag ) = dim(Ag−1 ) for all
g ∈ G. Indeed, if a, b ∈ Ag are non-zero and a−1 ∈ Ag−1 is the left inverse of a ∈ A which
exists by the above argument, then ba−1 ∈ Ae ∼ = C, so ba−1 = λi for some λ ∈ C. Right
multiplying by a, we have b = ba−1 a = λa as claimed.
We have just seen that to each connected separable algebra, there is a subgroup H ≤ G
on which A is supported, i.e., H = {h ∈ G|Ah ̸= 0}. Pick a basis element ah ∈ Ah for each
h ∈ H. The multiplication map m is then the same data as a scalar µg,h ∈ C× satisfying
ag ah = µg,h agh for each g, h ∈ H. Associativity of the algebra implies that
µgh,k µh,k aghk = ag (ah ak ) = ω(g, h, k)(ag ah )ak = ω(g, h, k)µg,h µgh,k
which implies that
ω(g, h, k) = µh,k µ−1 −1
gh,k µg,hk µg,h = (dµ)(g, h, k).
Thus a connected separable algebra is the data of a subgroup H ≤ G and a 2-cochain
µ ∈ C 2 (G, C× ) which trivializes the 3-cocycle ω, i.e., A is exactly the twisted group algebra
C[H, µ].
If we chose different bases (bh )h∈H , then for each h ∈ H, we have a scalar λh ∈ C× such
that bh = λh ah . This scalar changes µ by a 1-coboundary, so connected separable algebras
up to algebra isomorphism correspond to subgroups H ≤ G on which ω trivializes together
with a 2-cocycle [µ] ∈ H 2 (H, C× ).
Example 5.5.7. Suppose a Xb ∈ C(a → b). A separable dual for a Xb is a dual b Xa∨ ∈ C(b →
a) with maps coevX ∈ C(1a → a X ⊗b Xa∨ ) and evX ∈ C(b X ∨ ⊗a Xb ⇒ 1b ) such that evX
admits a right inverse ϵX ∈ C(1b → b X ∨ ⊗a Xb ).
Given a separable dual for a Xb , we can canonically endow X ⊗b X ∨ with the structure of
a unital condensation algebra. Indeed, we define

m := = idX ⊗ evX ⊗ idX ∨ i := = coevX ∆ := = idX ⊗ϵX ⊗idX ∨

We leave the rest of the straightforward verification to the reader.

Definition 5.5.8. Suppose (a Aa , mA ), (a Ba , mB ) are algebras and θ : A ⇒ B. We call θ an
algebra map if
B B

θ
θ θ
=

A A A A
14
If A, B are unital, we call θ a unital algebra map if in addition
B
B

θ = iB
.
iA

Observe that algebra objects in Ωa C and algebra maps form a 1-category.

Definition 5.5.9. A unital separable algebra a Aa splits if it is isomorphic via an algebra
map to a unital separable algebra of the form a X ⊗b Xa∨ from Example 5.5.7 where b Xa∨ is
a separable dual of a Xb .
Remark 5.5.10. A condensation algebra is the 2-categorical analog of an idempotent. An
idempotent in a 1-category can replicate freely on a line, and replicating arbitrarily many
times leads to the notion of splitting for an idempotent.
e e e e e ··· e e e
a a
= a a a a a
= ··· = a r s a

Similarly, a condensation algebra can replicate freely in a 2D mesh, and replicating arbitrarily
many times leads to the notion of splitting for a separable algebra.

= = =

Definition 5.5.11. A locally Cauchy complete 2-category is called 2-idempotent complete

if every unital separable algebra splits.
5.6. Bimodules and intertwiners. One gets the notion of a module M for an algebra A
by taking the axioms for a module and changing the appropriate instance of A to M .
Definition 5.6.1. Suppose (a Aa , m) is an algebra in C. A left A-module is a pair (a Mb , λ :
A ⊗a M → M ) for some b ∈ C such that the following associativity axiom holds:
idA ⊗λ
A ⊗ (A ⊗ M ) A⊗M
λ
αA,A,M
λ M ⇝ = ; = λ.
m⊗idM
(A ⊗ A) ⊗ M A⊗M
We leave the definition of right module to the reader.
If (a Aa , m) is unital, we call (a Mb , λ) unital if the following unitality axiom is saitsfied:
(λcA )−1
1a ⊗ M M
i⊗idM idM ⇝ = .
λ
A⊗M M
If (a Aa , m) is separable, we call (a Mb , λ) separable if λ splits as a left A-module map, and λ
is also a co-module map with respect to the splitting, i.e.,

• (λ splits) there is a map = δ : M → A ⊗ M such that λ · δ = = = idM ,

15
 • (as an A-(co)module map) = =

Exercise 5.6.2. Suppose (a Aa , m, i) is a unital separable algebra and (a Mb , λ) is a left

A-module. Prove that the following are equivalent: TODO: fix this
(1) (M, λ) is unital, and
(2) For any choice of separator ∆ : A → A ⊗ A, the map δ : M → A ⊗a M given by

δ :=

witnesses the separability of M .

Repeat the above exercise for right B-modules and A − B bimodules.
Exercise 5.6.3. Prove that for every c ∈ C, 1c is canonically a unital condensation algebra
in C. Then prove that for every a Xb ∈ C(a → b), the only separable 1a − 1b bimodule
structure on X is given by the unitors.
Definition 5.6.4. Suppose now (a Aa , mA ) and (b Bb , mB ) are algebras. an A − B bimodule
is a triple (a Mb , λ : A ⊗a M → M, ρ : M ⊗b B → M ) such that (M, λ) is a left A-module,
(M, ρ) is a right B-module, and the additional associativity axiom holds:
idA ⊗ρ
A ⊗ (M ⊗ A) A⊗M
λ
αA,M,A ρ M ⇝ = ; = ρ.
λ⊗idA
(A ⊗ M ) ⊗ A M ⊗A
As before, there is a notion of a (bi)module object in a multitensor category C for an
algebra object; it is a (bi)module for the corresponding algebra in BC.
Definition 5.6.5. Suppose (a Aa , m) is an algebra. Given two left A-modules (a Mb , λM ) and
(a Nb , λN ), a 2-morphism θ ∈ C(a Mb ⇒ a Nb ) is called a left A-module map if the following
diagram commutes:
λM
A⊗M M
⇝ θ = ; = λM , = λN
idA ⊗θ θ
θ
λN
A⊗N N
We leave the definition of a right B-module map and an A − B bimodule map to the reader.
Construction 5.6.6. Given an algebra object (A, m) in C, we can take the category CA
whose:
• objects are unital right A-module objects (M, ρ) in C, and
• 1-morphisms are right A-module maps.
Given c ∈ C, we get a free A-module given by c ⊗ A with right action map idc ⊗ρ. Observe
that we get a functor C → CA given by c 7→ c ⊗ A and (f : c1 → c2 ) 7→ f ⊗ idA ; we call this
the free module functor.
There is also a similar category A C of left A-modules in C and a category of left free
A-modules.
16
Exercise 5.6.7. Suppose (a Aa , m, i) is a unital algebra in C, (b Ma , ρ) is a right A-module,
and b Xa ∈ C(b → a). Find an isomorphism
CA (b X ⊗a Aa ⇒ b Ma ) ∼ = C(b Xa ⇒ b Ma )
which is natural in both M and X. Deduce that the forgetful functor CA → C which forgets
the right A-module structure is right-adjoint to the free module functor C → CA .
Exercise 5.6.8. Suppose (A, m) is an algebra object in C. Show that every idempotent
right A-module map e : MA → NA splits in CA . Deduce that CA is Cauchy complete, as is
A CA , the category of A − A bimodules in C with A − A bimodule maps.

Exercise 5.6.9. Suppose (A, m) is an algebra in a locally semisimple 2-category C. Show

that any right A-module sits in a co-equalizer diagram
ρM ⊗A ρM
M ⊗A⊗A M ⊗A M.
idM ⊗m

Deduce the following facts.

(1) Every free module is projective.
(2) If (A, m, ∆) is a condensation algebra in C, then every right condensation module
MA is a summand of the free module M ⊗ A.
(3) CA is the Cauchy completion of FreeModC (A).
Definition 5.6.10. Given a locally semisimple 2-category C, we have a 2-category Alg(C)
where
• objects are unital separable algebras in C,
• 1-morphisms A → B are separable A − B bimodules in C, and
• 2-morphisms A MB ⇒ A NB are intertwiners.
We now define 1-composition. Suppose A, B, C ∈ Alg(C) and A MB and B NC are separable
bimodules. Observe that M ⊗b N is organically an A − C bimodule, and by Exercise 5.6.8,
the category BimC (A → C) of A − C bimodules in C is Cauchy complete. We define the
relative tensor product M ⊗B N by splitting the idempotent

pM,N := := (idM ⊗λN ) ◦ (∆M ⊗ idN ) (5.6.11)

in BimC (A → C). Observe that M ⊗B N is only defined up to unique isomorphism.

Given f : A KB ⇒ A MB and g : B LC ⇒ B NC , the map f ⊗ g : K ⊗b L ⇒ M ⊗b N is an
A − C bimodule map such that
pM,N · (f ⊗ g) = (f ⊗ g) · pK,L .
We thus define f ⊗B g : A L ⊗B KC ⇒ A M ⊗B NC as the above composite morphism.
It is straightforward to verify the interchange axiom, so we have a 1-composition functor
− ⊗B − : BimC (A → B) × BimC (B → C) → BimC (A → C).
TODO: associator

Exercise 5.6.12. Verify (5.6.11) is an idempotent. Then show = .

Lemma 5.6.13. Alg(C) is 2-idempotent complete.

17
Proof. TODO: □
Definition 5.6.14. Two algebras A, B ∈ C are called Morita equivalent if their categories
of right modules in C are equivalent, i.e., CA ∼
= CB .
Proposition 5.6.15. Two unital separable algebras are Morita equivalent if and only if they
are equivalent in the 2-category Alg(C).
Proof. First, suppose A ∼ B in Alg(C), i.e., there is an A − B bimodule A MB and a B − A
bimodule B NA and bimodule isomorphisms A M ⊗B NA → A AA and B N ⊗A MB → B BB .
TODO: □
5.7. Module categories. In this section, C denotes a multitensor category unless stated
otherwise.
For (M, ρ) ∈ CA , observe that c ⊗ M also has a right A-module structure with action map
idc ⊗ρ. Thus the category ModC (A) has the structure of a left C-module category. (Below,
we require module categories to be semisimple, but this is not necessary in the definition.)
Definition 5.7.1. A left C-module category for a multitensor category C consists of a
semisimple category M together with a left C-action functor  : C × M → M a family
of natural unitor isomorphisms λm : 1C  m → m, and a family of natural associator iso-
morphisms αa,b,m : a  (b  m) → (a ⊗ b)  m which satisfy the pentagon and triangle
axioms.
Proposition 5.7.2. The following are equivalent for a unital algebra (A, m, i) in a semisim-
ple tensor category C.
(1) A is separable,
(2) CA is semisimple,
(3) A C is semisimple, and
(4) A CA is semisimple.
Proof.
(1) ⇒ (2): TODO: add abelian category section in earlier chapter If A is separable,
then CA is abelian by [[]]. We show every object MA ∈ CA is projective. TODO:
(1) ⇒ (3): Similar to (1) ⇒ (2) and omitted.
(2) ⇒ (4): TODO:
(3) ⇒ (4): Similar to (2) ⇒ (4) and omitted.
(4) ⇔ (1): TODO: □
Definition 5.7.3. Suppose M, N are two left C-module categories. A C-module functor
2
F : M → N is a functor equipped with a family of natural actionator isomorphisms Fc,m :
c  F(m) → F(c  m) satisfying an associative condition. Given two C-module functors
F, G : M → N , a C-module natural transformation θ : F ⇒ G is a natural transformation
F ⇒ G such that the following compatibility axiom is satisfied with the actionators:
2
Fc,m
c  F(m) G(c  m)
idc θm θm (5.7.4)
2
Gc,m
c  G(m) G(c  m).
18
Exercise 5.7.5. Show how to endow the left C-modules, C-module functors, and C-module
natural transformations with the structure of a 2-category. We call this 2-category Mod(C).
Construction 5.7.6. Suppose C is a multitensor category, A, B ∈ C are separable unital
algebra objects, and M ∈ C is an A − B bimodule object. We get a C-module functor
− ⊗A MB : CA → CB . Indeed, suppose NA ∈ CA , and let LB be an object obtained from
splitting pN,M . TODO:
TODO:
TODO: finitely many equivalence classes of indecomposable module cats over
a multifusion. This is actually too hard for right now...

5.8. Enriched categories and Ostrik’s Theorem. For this section, V denotes a monoidal
category. Typically, we will take V = Vec in applications, but sometimes we take V to be
sVec or another multifusion category.
Definition 5.8.1 ([Kel05]). Given a monoidal category V, a V-(enriched) category A consists
of the following data:
• a collection of objects a ∈ A,
• for each a, b ∈ A, a hom object A(a → b) ∈ V,
• a unit map jc ∈ V(1V → A(c → c)) for every c ∈ A, and
• a composition morphism − ◦A − ∈ V(A(b → c) ⊗ A(a → b) → A(a → c)) for all
a, b, c, ∈ A.
This data must satisfy the following axioms:
A(a→d) A(a→d)

− ◦A − − ◦A −
• (associativity) =
− ◦A − − ◦A −

A(c→d) A(b→c) A(a→b) A(c→d) A(b→c) A(a→b)

A(a→b) A(a→b) A(a→b)

− ◦A − − ◦A −
• (identity) = =
jb ja

A(a→b) A(a→b) A(a→b)

Exercise 5.8.2. Given a V-category A, the underlying category AV has the same objects as
A, but AV (a → b) := V(1V → A(a → b)). Show how to endow AV with the structure of an
ordinary category.
Remark 5.8.3. While we will not need this here at this time, there are notions of V-functor
and natural transformation so that V-categories, V-functors, and natural transformations
forms a 2-category. Taking the underlying category, functor, and natural transformation
gives a 2-functor VCat → Cat. We refer the reader to [Kel05] for more details.
19
 For the remained of this section, C denotes a multifusion category.
Definition 5.8.4. Suppose M is a finitely semisimple left C-module category. For each
m, n ∈ M, M(−  m → n) : C op → Vec is a linear functor and it is thus representable by
semisimplicity of C. We define the internal hom object M(m
c → n) ∈ C as the representing
object for the functor M(−  m → n). Moreover, internal hom is natural in each argument,
giving an adjunction
c → n)) ∼
C(c → M(m = M(c  m → n). (5.8.5)
Construction 5.8.6. The assignment M(m
c → n) ∈ C to m, n ∈ M defines a C-enriched
category with the same objects as M. The identity element jm ∈ C(1C → M(m
c → m)) is
the mate of λm : 1C  m → m under Adjunction (5.8.5). For each m, n ∈ M, we have an
evaluation map
c → n) ⊗ m → n) ∼
εm→n ∈ M(M(m = C(M(mc → n) → M(m c → n))
(5.8.5)

which is the mate of the identity idM(m→n)

c . We define the composition morphism − ◦M
c− ∈

C(M(n
c → p) ⊗ M(m c → n) → M(m c → p)) as the mate of

εn→p · (idM(n→p)
c ⊗εm→n ) ∈ M(M(n c → n)  m → p)
c → p) ⊗ M(m

under Adjunction (5.8.5).

Exercise 5.8.7. Verify the coherence axioms for the C-enriched category M
c constructed
above.
Remark 5.8.8. Construction 5.8.6 builds a C-enriched category M c from a left C-module
category. Conversely, given a particularly nice C-enriched category which is tensored [Lin81,
MPP18], we can build a left C-module category. These two constructions are mutually in-
verse; indeed the 2-category of these nice tensored C-enriched categories is equivalent to the
2-category of left C-module categories, C-module functors, and C-module natural transfor-
mations. We refer the reader to [Lin81, MPP18, KYZZ21, Del21] for more details.
Exercise 5.8.9. Show that for every m ∈ M, M(mc → m) has the structure of a unital
algebra object in C, and for every n ∈ M, M(m → n) is a right M(m
c c → m)-module.

Exercise 5.8.10. Suppose M is a finitely semisimple left C-module category.

c 1 → c  m2 ) ∼
(1) Find a canonical isomorphism M(m = c ⊗ M(m
c 1 → m2 ).
(2) Use (1) to prove that for any m ∈ M, M(m c → −) : M → C is a left C-module
functor.
c  m1 → m2 ) ∼
(3) Find a canonical isomorphism M(c c 1 → m2 ) ⊗ c∨ .
= M(m
Lemma 5.8.11. Suppose M is an indecomposable left C-module category, i.e., M is not the
direct sum of two non-zero C-module categories. For every m ∈ M, A := M(m
c → m) is a
unital separable algebra such that M(m → −) : M → CA is an equivalence of left C-module
c
categories.
Proof. Exercise 5.8.9 builds the unital algebra structure. Separability will follow showing
M(m
c → −) : M → CA is an equivalence by Proposition 5.7.2.
20
 First, M(m
c → −) : M → CA is well-defined by Exercise 5.8.9.
Second, since M is indecomposable, m generates M as a C-module category, i.e., every
object of M is a summand of a direct sum of objects of the form c  m. We claim this means
n ̸= 0 in M implies M(m
c → n) ̸= 0 in CA . Indeed, by indecomposability, there is a c ∈ C
such that M(c  m → n) ̸= 0, which immediately implies that
c → n)) ∼
C(c → M(m = M(c  m → n) ̸= 0.
Third, since M(m
c → −) is a linear functor which is non-zero on every simple object of
M, it is automatically faithful by [[]]. To see it is also full, we show it is full on objects of
the form c  m. Indeed,
c → a  m) → M(m
CA (M(m c → b  m)) ∼
= CA (a ⊗ M(m c → b  m))
c → m) → M(m
(Exer. 5.8.10)
∼
= c → b  m))
C(a → M(m
(Exer. 5.6.7)
∼
= M(a  m → b  m).
(5.8.5)

We thus see that M(mc → −) is full on the full subcategory M0 of M with objects of the
form c  m, so it is full on all of M as M is the Cauchy completion of M0 .
TODO: rewrite below: get a fully faithful linear functor from a semisimple
category M to the Cauchy complete category CA whose essential image contains
enough projectives (every object of CA is a quotient of a projective in the essential
image). This implies the latter is semisimple. Put this in an abelian category
section earlier on.
Finally, to see M(m
c → −) is essentially surjective, pick MA ∈ CA . By Exercise 5.6.9, MA
fits in a co-equalizer diagram
ρM ⊗A ρM
M ⊗A⊗A M ⊗A M. (5.8.12)
idM ⊗m

c → a  m) ∼
Since M(m = a ⊗ M(m
c → m) = a ⊗ A for all a ∈ C, by the last step, we have

M((M ⊗ A)  m → M  m) ∼ c → (M ⊗ A)  m) → M(m
= CA (M(m c → M  m))
∼
= CA (M ⊗ A ⊗ M(m
c → m) → M ⊗ M(m
c → m))
= CA (M ⊗ A ⊗ A → M ⊗ A)

Since M(m
c → −) is linear, it preserves kernels and cokernels by [[]]. Hence by transporting
the co-equalizer (5.8.12) across the fully faithful functor M(m
c → −), we see that MA fits in
a coequalizer diagram of free modules, and is thus in the essential image of M(m
c → −). □

We now state Ostrik’s Theorem and give a pedestrian proof following [Ost03]. (One can
use the Barr-Beck Theorem to prove this as well [BZBJ18, §4].)
Theorem 5.8.13 (Ostrik’s Theorem for multifusion categories). Let C be a multifusion
category. The map A 7→ CA and A MB 7→ − ⊗A MB is a 2-equivalence Alg(C) → Mod(C).
Proof of Theorem 5.8.13. It is straightforward to verify the above map gives a 2-functor.
21
 First, we check that for all unital A − B bimodules A MB , A NB ,

A CB (M ⇒ N ) ∋ θ 7→ − ⊗ θ ∈ FunC− (− ⊗A MB ⇒ − ⊗A NB )
is an isomorphism. Indeed, every C-module natural transformation ζ : − ⊗A MB ⇒ − ⊗A NB
is completely determined by ζA using (5.7.4) as CA is the Cauchy completion of FreeModC (A)
by Exercise 5.6.9.
Thus to show our 2-functor is fully faithful, we need to prove the hom functors are essen-
tially surjective. Suppose F : CA → CB is a left C-module functor. Then F(A) ∈ CB carries
2
both a right B-action and a left A-action using the modulator: λA := F(mA ) ◦ FA,A . Setting
∼
A MB := F(A), it is straightforward to check that F = − ⊗A M .
It remains to show the 2-functor is essentially surjective. To do this, we must show that
every semisimple left C-module category M is equivalent to CA for some separable unital
algebra A. We decompose M into indecomposable summands and apply Lemma 5.8.11 for
each summand to conclude the proof. □

5.9. 2-idempotent completion. We now prove the universal property of 2-idempotent

completion. The following is an adaptation of the proof from [CP22].
Definition 5.9.1. Suppose C is a 2-category and a, b ∈ C. A condensation or split surjection
X : a→ - b consists of 1-morphisms a Xb and b Ya and 2-morphisms ϵ : Y ⊗a X → 1b and
δ : 1b → Y ⊗a X such that ϵδ = id1b . In diagrams: TODO: bullets

δ= Y X and ϵ= Y X such that Y X = = id1b .

Definition 5.9.2. A 2-functor F : C → E is called

• locally dominant if every hom functor Fa→b : C(a → b) → E(F (a) → F (b)) is
dominant as an ordinary functor,
• 0-dominant if for all e ∈ E, there is a c ∈ C and a condensation F (c) →
- e, and
• dominant if F is both locally dominant and 0-dominant.
Proposition 5.9.3. Suppose G : C → E is a 2-functor and consider the 2-functor
G∗ = − ◦ G : Fun(E → D) → Fun(C → D).
If G is 0-dominant, then G∗ is faithful on 2-morphisms. If G dominant, then G∗ is fully
faithful on 2-morphisms.
Proof. TODO: □
Lemma 5.9.4. The inclusion functor ιC : C ,→ Alg(C) is dominant.
Proof. TODO: □
Theorem 5.9.5. Suppose D is a 2-idempotent complete 2-category. The 2-functor ι∗ :
Fun(Alg(C) → D) → Fun(C → D) is an equivalence.
Proof. TODO: □
22
5.10. Yoneda for (linear) 2-categories. Suppose C is a 2-category such that all hom
1-categories are finite semisimple categories. For each c ∈ C, we get two representable
2-functors:
C(c → −) : C → 2Vec and C(− → c) : C 1 op → 2Vec.
Here, C 1 op means the 2-category with the same objects as C, but C 1 op (a → b) := C(b → a).
The associators and unitors are defined similar to the definition of Dmp for a monoidal
category D.
Lemma 5.10.1 (Yoneda). Given a 2-functor F : C 1 op → 2Vec, evaluation at 1c gives an
equivalence of categories
よ(c, F ) : Hom(C(− → c) ⇒ F ) ∼
= F (c).
Moreover, よ(c, F ) is natural in both c and F .
Proof. TODO: □
TODO: Yoneda embedding, when is it an equivalence...
Corollary 5.10.2. The Yoneda embedding 2-functor
よ : C ,→ Fun(C 1 op → 2Vec)
c 7→ C(− → c)
is fully faithful.
Proof. TODO: □
Definition 5.10.3. A 2-functor F : C 1 op → 2Vec is called representable if there is an a ∈ C
and an invertible 2-transformation α : F ⇒ C(− → a). We call (a, α) a representing pair for
F.
Proposition 5.10.4. Representing pairs for F form a contractible space (when they exist).
Proof. First, note that if (a, α) is a representing pair for F , then α−1 is uniquely defined up
to a unique invertible 2-modification. As in the case for representable 1-functors, we thus
get canonical invertible 2-transformations
α−1 β β −1 α
C(− ⇒ a) =⇒ F =⇒ C(− → b) and C(− ⇒ b) =⇒ F =⇒ C(− → a).
By the Yoneda embedding, these 2-transformations must come from C(a → b) and C(b → a),
i.e., there is an a Xb ∈ C(a → b) and an invertible 2-modification m : β ◦ α−1 ⇛ − ⊗a Xb ,
and similarly a b Ya ∈ C(b → a) and an invertible 2-modification n : α ◦ β −1 ⇛ − ⊗b Ya .
TODO: □

5.11. Semisimple 2-categories. For 1-categories, semisimplicity is a purely algebraic prop-

erty, but for 2-categories, semisimplicity also includes dualizability, which is more of a topo-
logical property.
TODO:
Definition 5.11.1. A linear 2-category is called semisimple if it is locally finite semisimple,
additive, and idempotent complete such that all 1-morphisms admit left and right adjoints.
23
Proposition 5.11.2. An additive, 2-idempotent complete linear 2-category is semisimple if
and only if every endomorphism multifusion category EndC (c) is a multifusion category for
each c ∈ C.
Proof. TODO: □
TODO: bound on centers not what we know how to do now
Definition 5.11.3. A semisimple 2-category C is called finite if there is a global bound K > 0
such that for every c ∈ C, the multifusion category EndC (c) has at most K indecomposable
summands.
Theorem 5.11.4. A 2-category C is finite semisimple if and only it is equivalent to Mod(F)
for a multifusion category F.
Proof. TODO: □
References
[Bar14] Bruce Bartlett. Quasistrict symmetric monoidal 2-categories via wire diagrams, 2014.
arXiv:1409.2148.
[BMS12] John W. Barrett, Catherine Meusburger, and Gregor Schaumann. Gray categories with duals and
their diagrams, 2012. arXiv:1211.0529.
[BZBJ18] David Ben-Zvi, Adrien Brochier, and David Jordan. Integrating quantum groups over surfaces.
J. Topol., 11(4):874–917, 2018. MR3847209 DOI:10.1112/topo.12072 arXiv:1501.04652.
[CP22] Quan Chen and David Penneys. Q-system completion is a 3-functor. Theory Appl. Categ., 38:Pa-
per No. 4, 101–134, 2022. MR4369356 DOI:10.1002/num.22828 arXiv:2106.12437.
[Del21] Zachary Dell. A characterization of braided enriched monoidal categories, 2021. arXiv:2104.
07747.
[DR18] Christopher L. Douglas and David J. Reutter. Fusion 2-categories and a state-sum invariant for
4-manifolds, 2018. arXiv:1812.11933.
[GJF19] Davide Gaiotto and Theo Johnson-Freyd. Condensations in higher categories, 2019. arXiv:1905.
09566.
[Gur13] Nick Gurski. Coherence in three-dimensional category theory, volume 201 of Cam-
bridge Tracts in Mathematics. Cambridge University Press, Cambridge, 2013. MR3076451
DOI:10.1017/CBO9781139542333.
[Gut19] Peter Guthmann. The tricategory of formal composites and its strictification, 2019. arXiv:1903.
05777.
[JY21] Niles Johnson and Donald Yau. 2-dimensional categories. Oxford University Press, Oxford, 2021.
2002.06055 DOI:10.1093/oso/9780198871378.001.0001 arXiv:MR4261588.
[Kel05] G. M. Kelly. Basic concepts of enriched category theory. Repr. Theory Appl. Categ., (10):vi+137,
2005. MR2177301, Reprint of the 1982 original [Cambridge Univ. Press, Cambridge; MR0651714].
[KYZZ21] Liang Kong, Wei Yuan, Zhi-Hao Zhang, and Hao Zheng. Enriched monoidal categories I: centers,
2021. arXiv:2104.03121.
[Lin81] Harald Lindner. Enriched categories and enriched modules. Cahiers Topologie Géom.
Différentielle, 22(2):161–174, 1981. Third Colloquium on Categories, Part III (Amiens, 1980)
MR649797.
[MPP18] Scott Morrison, David Penneys, and Julia Plavnik. Completion for braided enriched monoidal
categories, 2018. arXiv:1809.09782.
[Ost03] Victor Ostrik. Module categories, weak Hopf algebras and modular invariants. Transform. Groups,
8(2):177–206, 2003. MR1976459 arXiv:math/0111139.

24