The Linear-Non-Linear Substitution Monad OWLS, 15 July 2020 - - PowerPoint PPT Presentation

the linear non linear substitution monad
SMART_READER_LITE
LIVE PREVIEW

The Linear-Non-Linear Substitution Monad OWLS, 15 July 2020 - - PowerPoint PPT Presentation

The Linear-Non-Linear Substitution Monad OWLS, 15 July 2020 Christine Tasson (tasson@irif.fr) Joint work with Martin Hyland at https://arxiv.org/abs/2005.09559 Institut de Recherche en Informatique Fondamentale Roadmap: The linear-non-linear


slide-1
SLIDE 1

The Linear-Non-Linear Substitution Monad

OWLS, 15 July 2020

Christine Tasson (tasson@irif.fr) Joint work with Martin Hyland at https://arxiv.org/abs/2005.09559

Institut de Recherche en Informatique Fondamentale

slide-2
SLIDE 2

Roadmap: The linear-non-linear substitution monad

Motivation:

  • Differential λ-calculus

Goal:

  • Axiomatisation using generalized multicategories.

Tool:

  • A colimit construction applied to combine 2-monads on Cat

Results:

  • The colimit is a 2-monad.
  • Charaterization of its algebras.

1

slide-3
SLIDE 3

Linear-non-linear substitution

Substitutions in differential λ-calculus

slide-4
SLIDE 4

Differential λ-Calculus

(Ehrhard-Regnier 2003, Ehrhard 2018)

Semantical observation: in quantitative models of Linear Logic, programs are interpreted by smooth functions, hence differentiation. Programs Functions M, N f , g Variable x x Variable Abstraction λx.M f : x → f (x) Map Application (λx.M)N f ◦ g : x → f ( g(x) ) Composition Differentiation Dλx.M · N u, x → Dfx(u) Derivation

2

slide-5
SLIDE 5

Linear and Non-Linear substitutions in Differential λ-calculus

Substitution (λx.M)N → M[x\N] Dλx.M · N → λx. ∂M ∂x · N

  • Linear approximation

f(x) Df (u)

x x

3

slide-6
SLIDE 6

Linear and Non-Linear substitutions in Differential λ-calculus

Substitution (λx.M)N → M[x\N] Dλx.M · N → λx. ∂M ∂x · N

  • Linear approximation

f(x) Df (u)

x x

3

slide-7
SLIDE 7

Linear-non-linear substitution

Type system and term calculus

slide-8
SLIDE 8

A term calculus for Linear-non-linear Logic

(Benton-Bierman-de Paiva-Hyland 1993, Barber 1996)

x1 : a1, . . ., xℓ : aℓ

  • Linear

| y1 : b1, . . ., yn : bn

  • Non-Linear

⊢ t : c

Linear rules:

x : a | ∆ ⊢ x : a Γ, x : a | ∆ ⊢ t : b Γ | ∆ ⊢ λxa.t : a ⊸ b Γ | ∆ ⊢ s : a ⊸ b Γ′ | ∆ ⊢ t : a Γ, Γ′ | ∆ ⊢ st : b

Non-linear rules:

Γ | ∆, x : b ⊢ x : b Γ | ∆, x : a ⊢ t : b Γ | ∆ ⊢ λxa.t : a → b Γ | ∆ ⊢ s : a → b · | ∆ ⊢ t : a Γ | ∆ ⊢ (s)t : b Γ , x : a | ∆ ⊢ t : b Linear-non-linear rule: Γ | ∆ , x : a ⊢ t : b

4

slide-9
SLIDE 9

A term calculus for Linear-non-linear Logic

(Benton-Bierman-de Paiva-Hyland 1993, Barber 1996)

x1 : a1, . . ., xℓ : aℓ

  • Linear

| y1 : b1, . . ., yn : bn

  • Non-Linear

⊢ t : c

Linear rules:

x : a | ∆ ⊢ x : a Γ, x : a | ∆ ⊢ t : b Γ | ∆ ⊢ λxa.t : a ⊸ b Γ | ∆ ⊢ s : a ⊸ b Γ′ | ∆ ⊢ t : a Γ, Γ′ | ∆ ⊢ st : b

Non-linear rules:

Γ | ∆, x : b ⊢ x : b Γ | ∆, x : a ⊢ t : b Γ | ∆ ⊢ λxa.t : a → b Γ | ∆ ⊢ s : a → b · | ∆ ⊢ t : a Γ | ∆ ⊢ (s)t : b Γ , x : a | ∆ ⊢ t : b Linear-non-linear rule: Γ | ∆ , x : a ⊢ t : b

4

slide-10
SLIDE 10

A term calculus for Linear-non-linear Logic

(Benton-Bierman-de Paiva-Hyland 1993, Barber 1996)

4

slide-11
SLIDE 11

A term calculus for Linear-non-linear Logic

(Benton-Bierman-de Paiva-Hyland 1993, Barber 1996)

x1 : a1, . . ., xℓ : aℓ

  • Linear

| y1 : b1, . . ., yn : bn

  • Non-Linear

⊢ t : c

Linear rules:

x : a | ∆ ⊢ x : a Γ, x : a | ∆ ⊢ t : b Γ | ∆ ⊢ λxa.t : a ⊸ b Γ | ∆ ⊢ s : a ⊸ b Γ′ | ∆ ⊢ t : a Γ, Γ′ | ∆ ⊢ st : b

Non-linear rules:

Γ | ∆, x : b ⊢ x : b Γ | ∆, x : a ⊢ t : b Γ | ∆ ⊢ λxa.t : a → b Γ | ∆ ⊢ s : a → b · | ∆ ⊢ t : a Γ | ∆ ⊢ (s)t : b Γ , x : a | ∆ ⊢ t : b Linear-non-linear rule: Γ | ∆ , x : a ⊢ t : b

4

slide-12
SLIDE 12

A term calculus for Linear-non-linear Logic

(Benton-Bierman-de Paiva-Hyland 1993, Barber 1996)

4

slide-13
SLIDE 13

A term calculus for Linear-non-linear Logic

(Benton-Bierman-de Paiva-Hyland 1993, Barber 1996)

x1 : a1, . . ., xℓ : aℓ

  • Linear

| y1 : b1, . . ., yn : bn

  • Non-Linear

⊢ t : c

Linear rules:

x : a | ∆ ⊢ x : a Γ, x : a | ∆ ⊢ t : b Γ | ∆ ⊢ λxa.t : a ⊸ b Γ | ∆ ⊢ s : a ⊸ b Γ′ | ∆ ⊢ t : a Γ, Γ′ | ∆ ⊢ st : b

Non-linear rules:

Γ | ∆, x : b ⊢ x : b Γ | ∆, x : a ⊢ t : b Γ | ∆ ⊢ λxa.t : a → b Γ | ∆ ⊢ s : a → b · | ∆ ⊢ t : a Γ | ∆ ⊢ (s)t : b Γ , x : a | ∆ ⊢ t : b Linear-non-linear rule: Γ | ∆ , x : a ⊢ t : b

4

slide-14
SLIDE 14

A term calculus for Linear-non-linear Logic

(Benton-Bierman-de Paiva-Hyland 1993, Barber 1996)

x1 : a1, . . ., xℓ : aℓ

  • Linear

| y1 : b1, . . ., yn : bn

  • Non-Linear

⊢ t : c

MLL x : a ⊢ x : a Γ, x : a ⊢ t : b Γ ⊢ λxa.t : a ⊸ b Γ ⊢ s : a ⊸ b Γ′ ⊢ t : a Γ, Γ′ ⊢ st : b λ-calculus ∆, x : b ⊢ x : b ∆, x : a ⊢ t : b ∆ ⊢ λxa.t : a → b ∆ ⊢ s : a → b ∆ ⊢ t : a ∆ ⊢ (s)t : b

4

slide-15
SLIDE 15

What is a model of substitution ?

combining linearity and non-linearity

slide-16
SLIDE 16

Axiomatic using Categories

In a category X, equipped with the right structure (SMCC/ CCC) Types are interpreted as objects Contexts are interpreted as objects (products/tensors) Terms are interpreted as morphisms Substitution is interpreted as composition In Multiplicative Linear Logic, a proof is interpreted as a morphism x1 : a1, . . ., xℓ : aℓ ⊢ t : c as a1 ⊗ · · · ⊗ aℓ ⊸ c.

5

slide-17
SLIDE 17

Axiomatic using Categories

In a category X, equipped with the right structure (SMCC/ CCC) Types are interpreted as objects Contexts are interpreted as objects (products/tensors) Terms are interpreted as morphisms Substitution is interpreted as composition In Multiplicative Linear Logic, a proof is interpreted as a morphism x1 : a1, . . ., xℓ : aℓ ⊢ t : c as a1 ⊗ · · · ⊗ aℓ ⊸ c.

5

slide-18
SLIDE 18

Axiomatic using Categories

In a category X, equipped with the right structure (SMCC/ CCC) Types are interpreted as objects Contexts are interpreted as objects (products/tensors) Terms are interpreted as morphisms Substitution is interpreted as composition In λ-calculus, a term is interpreted as a morphism x1 : b1, . . ., xn : bn ⊢ t : c as b1 × · · · × bn → c .

5

slide-19
SLIDE 19

Axiomatic using Categories

In a category X, equipped with the right structure (SMCC/ CCC) Types are interpreted as objects Contexts are interpreted as objects (products/tensors) Terms are interpreted as morphisms Substitution is interpreted as composition In λ-calculus, a term is interpreted as a morphism x1 : b1, . . ., xn : bn ⊢ t : c as b1 × · · · × bn → c .

5

slide-20
SLIDE 20

Axiomatic using Categories

In a category X, equipped with the right structure (SMCC/ CCC) Types are interpreted as objects Contexts are interpreted as objects (products/tensors) Terms are interpreted as morphisms Substitution is interpreted as composition In λ-calculus, a term is interpreted as a morphism x1 : b1, . . ., xn : bn ⊢ t : c as b1 × · · · × bn → c

!b1 ⊗···⊗!bn⊸c

.

5

slide-21
SLIDE 21

Axiomatic using Categories

In a category X, equipped with the right structure (SMCC/ CCC) Types are interpreted as objects Contexts are interpreted as objects (products/tensors) Terms are interpreted as morphisms Substitution is interpreted as composition In λ-calculus, a term is interpreted as a morphism x1 : b1, . . ., xn : bn ⊢ t : c as b1 × · · · × bn → c

!b1 ⊗···⊗!bn⊸c

. In lnl λ-calculus, x1 : a1, . . ., xℓ : aℓ | y1 : b1, . . ., yn : bn ⊢ t : c as a1 ⊗ · · · ⊗ aℓ⊗!b1 ⊗ · · · ⊗!bn ⊸ c.

5

slide-22
SLIDE 22

Axiomatic using generalized multicategories

A multicategory is a set of operations: Together with identity and multicomposition: ⇒

6

slide-23
SLIDE 23

Axiomatic using generalized Multicategories

In a multicategory Types are interpreted as objects Terms are interpreted as multimorphisms Substitution is interpreted as multicomposition.

7

slide-24
SLIDE 24

Axiomatic using generalized Multicategories

In a multicategory Types are interpreted as objects Terms are interpreted as multimorphisms Substitution is interpreted as multicomposition. In Multiplicative Linear Logic, a term is interpreted as a multimorphism in a symmetric multicategory: x1 : a1, . . ., xℓ : aℓ ⊢ t : c denoted as a1, . . ., aℓ ⊸ c.

7

slide-25
SLIDE 25

Axiomatic using generalized Multicategories

In a multicategory Types are interpreted as objects Terms are interpreted as multimorphisms Substitution is interpreted as multicomposition. In λ-calculus, a term is interpreted as a multimorphism in a cartesian multicategory: y1 : b1, . . ., yn : bn ⊢ t : c denoted as b1, . . ., bn → c.

7

slide-26
SLIDE 26

Axiomatic using generalized Multicategories

In a multicategory Types are interpreted as objects Terms are interpreted as multimorphisms Substitution is interpreted as multicomposition. What are the operations for the Linear-non-linear calculus, a term is interpreted as a multimorphism in a generalized multicategory: x1 : a1, . . ., xℓ : aℓ | y1 : b1, . . ., yn : bn ⊢ t : c denoted as a1, . . ., aℓ, b1, . . ., bn → c.

7

slide-27
SLIDE 27

Axiomatic using generalized Multicategories

In a multicategory Types are interpreted as objects Terms are interpreted as multimorphisms Substitution is interpreted as multicomposition. What are the operations for the Linear-non-linear calculus, a term is interpreted as a multimorphism in a generalized multicategory: x1 : a1, . . ., xℓ : aℓ | y1 : b1, . . ., yn : bn ⊢ t : c denoted as a1, . . ., aℓ, b1, . . ., bn → c. Multicategories can be seen as profunctors combined with a monad.

(Fiore-Plotkin-Turi 1999, Tanaka-Power 2006)

7

slide-28
SLIDE 28

Generalized Multicategories and Context Monads

A multicategory is a set of operations: M : TX op × X → Set The context is represented as a sequence of objects via a monad T on Cat. Together with unit and multicomposition: M ◦ TM ⇒ M ⇒

8

slide-29
SLIDE 29

Axiomatization using Multicategories via Profunctors

In a multicategory M : TX op × X → Set Types are interpreted as objects in X Terms are interpreted as elements of M Substitution is interpreted by the monadic structure M ◦ TM ⇒ M

9

slide-30
SLIDE 30

Axiomatization using Multicategories via Profunctors

In a multicategory M : TX op × X → Set Types are interpreted as objects in X Terms are interpreted as elements of M Substitution is interpreted by the monadic structure M ◦ TM ⇒ M In Multiplicative Linear Logic, a term is interpreted in a symmetric multicategory: M : LX op × X → Set x1 : a1, . . ., xℓ : aℓ ⊢ t : c as a1, . . ., aℓ ⊸ c in M(a1, . . ., aℓ; c) Algebras

  • f

L are symmetric strict monoidal categories LX is the free one on X

(Fiore-Gambino-Hyland-Winskel 2007)

9

slide-31
SLIDE 31

Axiomatization using Multicategories via Profunctors

In a multicategory M : TX op × X → Set Types are interpreted as objects in X Terms are interpreted as elements of M Substitution is interpreted by the monadic structure M ◦ TM ⇒ M In λ-calculus, a term is interpreted in a cartesian multicategory M : MX op × X → Set y1 : b1, . . ., yn : bn ⊢ t : c as b1, . . ., bn ⊸ c in M(b1, . . ., bn; c) Algebras of M are the categories with product MX is the free one over X

(Tanaka-Power 2004, Hyland 2017)

9

slide-32
SLIDE 32

Axiomatization using Multicategories via Profunctors

In a multicategory M : TX op × X → Set Types are interpreted as objects in X Terms are interpreted as elements of M Substitution is interpreted by the monadic structure M ◦ TM ⇒ M What is Q for a Mixed Linear-Non-Linear calculus a term is interpreted in a generalized multicategory M : QX op ×X → Set: x1 : a1, . . ., xℓ : aℓ | y1 : b1, . . ., yn : bn ⊢ t : c in M(a1, . . ., aℓ | b1, . . ., bn; c). QX is the category whose objects are mixed LNL sequences.

9

slide-33
SLIDE 33

Axiomatization using Multicategories via Profunctors

In a multicategory M : TX op × X → Set Types are interpreted as objects in X Terms are interpreted as elements of M Substitution is interpreted by the monadic structure M ◦ TM ⇒ M What is Q for a Mixed Linear-Non-Linear calculus a term is interpreted in a generalized multicategory M : QX op ×X → Set: x1 : a1, . . ., xℓ : aℓ | y1 : b1, . . ., yn : bn ⊢ t : c in M(a1, . . ., aℓ | b1, . . ., bn; c). QX is the category whose objects are mixed LNL sequences. Is Q a monad ? What are Q-algebras ?

9

slide-34
SLIDE 34

A Colimit construction

To build the Linear-non-linear monad

slide-35
SLIDE 35

colimits induced by a map in a category K

If λ : A → B is a map in K, then the induced colimit is A B C

λ k ℓ

  • for any

A B D

λ f g

= A B C D

λ k ℓ ∃!r 10

slide-36
SLIDE 36

Colax colimits induced by a map in a 2-category K

If λ : A → B is a map in K, then the induced colax colimit is A B C

λ k ℓ α

  • for any

A B D

λ f g φ

= A B C D

λ k ℓ α ∃!r 10

slide-37
SLIDE 37

Colax colimits induced by a map in a 2-category K

If λ : A → B is a map in K, then the induced colax colimit is A B C

λ k ℓ α

There are two universal aspects for 1-cells and 2-cells

  • for any

A B D

λ f g φ

= A B C D

λ k ℓ α ∃!r

  • for any

A B D

λ f f ′ ρ g φ

= A B D

λ f ′ g g′ σ φ′

, ∃! r

τ

= ⇒r ′ s.t.

A D

f f ′ ρ

= A C D

k r r′ τ

B D

g g′ σ

= B C D

ℓ r r′ τ 10

slide-38
SLIDE 38

Mixing Linear and Non-Linear contexts via a colimit

Remark:

  • Every category with products is a symmetric strict monoidal category

λX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ ∈ MX

11

slide-39
SLIDE 39

Mixing Linear and Non-Linear contexts via a colimit

Remark:

  • Every category with products is a symmetric strict monoidal category

λX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ ∈ MX Wanted

  • QX is in SymStMonCat and objects are a1, . . ., aℓ | b1, . . ., bn

11

slide-40
SLIDE 40

Mixing Linear and Non-Linear contexts via a colimit

Remark:

  • Every category with products is a symmetric strict monoidal category

λX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ ∈ MX Wanted

  • QX is in SymStMonCat and objects are a1, . . ., aℓ | b1, . . ., bn
  • QX contains Linear objects

kX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ | · ∈ QX

11

slide-41
SLIDE 41

Mixing Linear and Non-Linear contexts via a colimit

Remark:

  • Every category with products is a symmetric strict monoidal category

λX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ ∈ MX Wanted

  • QX is in SymStMonCat and objects are a1, . . ., aℓ | b1, . . ., bn
  • QX contains Linear objects

kX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ | · ∈ QX

  • QX contains Non-Linear ones

ℓX : b1, . . ., bn ∈ MX → · | b1, . . ., bn ∈ QX

11

slide-42
SLIDE 42

Mixing Linear and Non-Linear contexts via a colimit

Remark:

  • Every category with products is a symmetric strict monoidal category

λX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ ∈ MX Wanted

  • QX is in SymStMonCat and objects are a1, . . ., aℓ | b1, . . ., bn
  • QX contains Linear objects

kX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ | · ∈ QX

  • QX contains Non-Linear ones

ℓX : b1, . . ., bn ∈ MX → · | b1, . . ., bn ∈ QX Solution: Colax Colimit over λ in the 2-category of SymStMonCat LX MX QX

λX kX ℓX αX

αX, a1,...,aℓ ∈ QX( · | a1, . . ., aℓ , a1, . . ., aℓ | · )

11

slide-43
SLIDE 43

Mixing Linear and Non-Linear contexts via a colimit

Remark:

  • Every category with products is a symmetric strict monoidal category

λX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ ∈ MX Wanted

  • QX is in SymStMonCat and objects are a1, . . ., aℓ | b1, . . ., bn
  • QX contains Linear objects

kX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ | · ∈ QX

  • QX contains Non-Linear ones

ℓX : b1, . . ., bn ∈ MX → · | b1, . . ., bn ∈ QX Solution: Colax Colimit over λ in the 2-category of SymStMonCat LX MX QX

λX kX ℓX αX

αX, a1,...,aℓ ∈ QX( · | a1, . . ., aℓ , a1, . . ., aℓ | · ) x1 : a1, . . ., xℓ : aℓ | · ⊢ t : b · | x1 : a1, . . ., xℓ : aℓ ⊢ t : b

11

slide-44
SLIDE 44

Mixing Linear and Non-Linear contexts via a colimit

Remark:

λ : L → M is a map of 2-monads

  • Every category with products is a symmetric strict monoidal category

λX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ ∈ MX Wanted

  • QX is in SymStMonCat and objects are a1, . . ., aℓ | b1, . . ., bn
  • QX contains Linear objects

k : L → Q is a map of 2-monad

kX : a1, . . ., aℓ ∈ LX → a1, . . ., aℓ | · ∈ QX

  • QX contains Non-Linear ones

ℓ : M → Q is almost a map of 2-monad

ℓX : b1, . . ., bn ∈ MX → · | b1, . . ., bn ∈ QX Solution: Colax Colimit over λ in the 2-category of SymStMonCat LX MX QX

λX kX ℓX αX

αX, a1,...,aℓ ∈ QX( · | a1, . . ., aℓ , a1, . . ., aℓ | · ) x1 : a1, . . ., xℓ : aℓ | · ⊢ t : b · | x1 : a1, . . ., xℓ : aℓ ⊢ t : b

11

slide-45
SLIDE 45

A Colimit construction

Q is a 2-monad on Cat and Q-algebras

slide-46
SLIDE 46

Properties of the QX from universality for 1-cell and 2-cell

Colimit in the 2-category of Symmetric Strict Monoidal Categories. LX MX QX

λX kX ℓX αX

  • LX the free symmetric st. monoidal category
  • MX the free category with products
  • QX objects are a1, . . ., aℓ | b1, . . ., bn
  • QX is a symmetric strict monoidal category, i.e. an L-algebra

12

slide-47
SLIDE 47

Properties of the QX from universality for 1-cell and 2-cell

Colimit in the 2-category of Symmetric Strict Monoidal Categories. LX MX QX

λX kX ℓX αX

  • LX the free symmetric st. monoidal category
  • MX the free category with products
  • QX objects are a1, . . ., aℓ | b1, . . ., bn
  • QX is a symmetric strict monoidal category, i.e. an L-algebra

Z LZ Z

ηL w

L2Z LZ LZ Z

µL Lw w w

and coherences

12

slide-48
SLIDE 48

Properties of the QX from universality for 1-cell and 2-cell

Colimit in the 2-category of Symmetric Strict Monoidal Categories. LX MX QX

λX kX ℓX αX

  • LX the free symmetric st. monoidal category
  • MX the free category with products
  • QX objects are a1, . . ., aℓ | b1, . . ., bn
  • QX is a symmetric strict monoidal category, i.e. an L-algebra
  • QX is equipped with a strictly idempotent comonad

f : QX

hX

− − → MX

kX

− − → QX

12

slide-49
SLIDE 49

Properties of the QX from universality for 1-cell and 2-cell

Colimit in the 2-category of Symmetric Strict Monoidal Categories. LX MX QX

λX kX ℓX αX

  • LX the free symmetric st. monoidal category
  • MX the free category with products
  • QX objects are a1, . . ., aℓ | b1, . . ., bn
  • QX is a symmetric strict monoidal category, i.e. an L-algebra
  • QX is equipped with a strictly idempotent comonad

f : QX

hX

− − → MX

kX

− − → QX a1, . . ., aℓ | b1, . . ., bn → a1, . . ., aℓ, b1, . . . bn → · | a1, . . ., aℓ, b1, . . ., bn

12

slide-50
SLIDE 50

Properties of the QX from universality for 1-cell and 2-cell

Colimit in the 2-category of Symmetric Strict Monoidal Categories. LX MX QX

λX kX ℓX αX

  • LX the free symmetric st. monoidal category
  • MX the free category with products
  • QX objects are a1, . . ., aℓ | b1, . . ., bn
  • QX is a symmetric strict monoidal category, i.e. an L-algebra
  • QX is equipped with a strictly idempotent comonad

f : QX

hX

− − → MX

kX

− − → QX a1, . . ., aℓ | b1, . . ., bn → a1, . . ., aℓ, b1, . . . bn → · | a1, . . ., aℓ, b1, . . ., bn f

β

= ⇒ idQX x1 : a1, . . ., xℓ : aℓ | ∆ ⊢ t : b · | x1 : a1, . . ., xℓ : aℓ, ∆ ⊢ t : b βa1,...,aℓ | b1,...,bn : · | a1, . . ., aℓ, b1, . . ., bn → a1, . . ., aℓ | b1, . . ., bn

12

slide-51
SLIDE 51

Properties of the QX from universality for 1-cell and 2-cell

Colimit in the 2-category of Symmetric Strict Monoidal Categories. LX MX QX

λX kX ℓX αX

  • LX the free symmetric st. monoidal category
  • MX the free category with products
  • QX objects are a1, . . ., aℓ | b1, . . ., bn
  • QX is a symmetric strict monoidal category, i.e. an L-algebra
  • QX is equipped with a strictly idempotent comonad

f : QX

hX

− − → MX

kX

− − → QX

  • QX is almost with products, i.e. a left-semi M-algebra:

MQX

z

− → QX

12

slide-52
SLIDE 52

Properties of the QX from universality for 1-cell and 2-cell

Colimit in the 2-category of Symmetric Strict Monoidal Categories. LX MX QX

λX kX ℓX αX

  • LX the free symmetric st. monoidal category
  • MX the free category with products
  • QX objects are a1, . . ., aℓ | b1, . . ., bn
  • QX is a symmetric strict monoidal category, i.e. an L-algebra
  • QX is equipped with a strictly idempotent comonad

f : QX

hX

− − → MX

kX

− − → QX

  • QX is almost with products, i.e. a left-semi M-algebra:

MQX

z

− → QX Z MZ Z

ηM z ǫ

M2Z MZ MZ Z

µM Mz z z

and coherences

12

slide-53
SLIDE 53

Properties of the QX from universality for 1-cell and 2-cell

Colimit in the 2-category of Symmetric Strict Monoidal Categories. LX MX QX

λX kX ℓX αX

  • LX the free symmetric st. monoidal category
  • MX the free category with products
  • QX objects are a1, . . ., aℓ | b1, . . ., bn
  • QX is a symmetric strict monoidal category, i.e. an L-algebra
  • QX is equipped with a strictly idempotent comonad

f : QX

hX

− − → MX

kX

− − → QX

  • QX is almost with products, i.e. a left-semi M-algebra:

MQX

z

− → QX

  • The induced left-semi L-algebras are equal:

LQX

λQX

− − − → MQX

z

− → QX LQX

w

− → QX

f

− → QX

12

slide-54
SLIDE 54

Structure category of Q

Colimit in the 2-category of Symmetric Strict Monoidal Categories. LX MX QX

λX kX ℓX αX

  • LX the free symmetric st. monoidal category
  • MX the free category with products
  • QX objects are a1, . . ., aℓ | b1, . . ., bn

Structure Category

  • Z is a symm. str. monoidal category, i.e. an L-algebra: LZ

w

− → Z

  • Z is almost with products, i.e. a left-semi M-algebra: MZ

z

− → Z

  • Z is equipped with a strictly idempotent comonad

f : Z

ηz

− − → MZ

z

− → Z

  • The induced left-semi L-algebras are equal:

LZ

λZ

− − → MZ

z

− → Z LZ

w

− → Z

f

− → Z

13

slide-55
SLIDE 55

Structure category of Q

Colimit in the 2-category of Symmetric Strict Monoidal Categories. LX MX QX

λX kX ℓX αX

  • LX the free symmetric st. monoidal category
  • MX the free category with products
  • QX objects are a1, . . ., aℓ | b1, . . ., bn

Structure Category QX has this structure !

  • Z is a symm. str. monoidal category, i.e. an L-algebra: LZ

w

− → Z

  • Z is almost with products, i.e. a left-semi M-algebra: MZ

z

− → Z

  • Z is equipped with a strictly idempotent comonad

f : Z

ηz

− − → MZ

z

− → Z

  • The induced left-semi L-algebras are equal:

LZ

λZ

− − → MZ

z

− → Z LZ

w

− → Z

f

− → Z

13

slide-56
SLIDE 56

A general colax colimit construction on 2-monads

Let λ : L → M a map of 2-monad on Cat. If L-algebras has colimits, then the colimit is LX MX QX

λX kX ℓX αX

Theorem

  • Q-algebras are objects in the Structure Category and
  • Q is a 2-monad on Cat.

The proof uses universality of the colimit.

14

slide-57
SLIDE 57

It is not the end of the story

Does Q lifts from Cat to Prof ?

slide-58
SLIDE 58

Multicategories as Profunctors with Context Monad

A multicategory can be seen as a profunctor in the Kleisli bicat of T: M : X −→ TX M : TX op × X → Set Together with unit and multicomposition: M ◦ TM ⇒ M ⇒ Q on Cat has to extend to Prof as L and M. Equivalently, Does the presheaf pseudomonad Psh lifts to pseudo Q-algebras ?

(Tanaka 2005, Fiore-Gambino-Hyland-Winskel 2016)

15

slide-59
SLIDE 59

Does Q extends from Cat to Prof ?

Wanted: The presheaf pseudomonad Psh lifts to pseudo Q-algebras.

16

slide-60
SLIDE 60

Does Q extends from Cat to Prof ?

Wanted: The presheaf pseudomonad Psh lifts to pseudo Q-algebras. Conjecture: Pseudo Q-algebras are pseudo Structure Categories.

16

slide-61
SLIDE 61

Does Q extends from Cat to Prof ?

Wanted: The presheaf pseudomonad Psh lifts to pseudo Q-algebras. Conjecture: Pseudo Q-algebras are pseudo Structure Categories. Problem: Pseudo Q-algebras are pseudo L-algebras, i.e. symmetric monoidal categories. There is NO COLIMIT in the 2-category of symmetric monoidal categories

16

slide-62
SLIDE 62

Does Q extends from Cat to Prof ?

Wanted: The presheaf pseudomonad Psh lifts to pseudo Q-algebras. Conjecture: Pseudo Q-algebras are pseudo Structure Categories. Problem: Pseudo Q-algebras are pseudo L-algebras, i.e. symmetric monoidal categories. There is NO COLIMIT in the 2-category of symmetric monoidal categories Work in progress: Use a strictification to recover colimits.

16

slide-63
SLIDE 63

It is not the end of the story

Categorical axiomatizations ?

slide-64
SLIDE 64

Can we bulid a Q-multicategory from a LNL adjunction ?

Linear-non-linear adjunction

(Benton 1994)

A monoidal adjunction X Y

s r

⊢ with X symmetric strict monoidal and Y with products

17

slide-65
SLIDE 65

Can we bulid a Q-multicategory from a LNL adjunction ?

Linear-non-linear adjunction

(Benton 1994)

A monoidal adjunction X Y

s r

⊢ with X symmetric strict monoidal and Y with products Structure Category from a ?

  • X is symmetric strict monoidal, so LX

w

− → X

  • Y has products, so MY

y

− → Y , we can build MX

r

− → MY

y

− → Y

s

− → X

  • f =! : X

r

− → Y

s

− → X is only lax monoidal

  • The two induced left-semi L-algebras are not equal

17

slide-66
SLIDE 66

Can we bulid a Q-multicategory from a LNL adjunction ?

Linear-non-linear adjunction

(Benton 1994)

A monoidal adjunction X Y

s r

⊢ with X symmetric strict monoidal and Y with products Structure Category from a ?

  • X is symmetric strict monoidal, so LX

w

− → X

  • Y has products, so MY

y

− → Y , we can build MX

r

− → MY

y

− → Y

s

− → X

  • f =! : X

r

− → Y

s

− → X is only lax monoidal

  • The two induced left-semi L-algebras are not equal

Work in progress: recover a structure category by going through multicategories.

17

slide-67
SLIDE 67

It is not the end of the story

Differential λ-calculus axiomatization

slide-68
SLIDE 68

Towards a multicategorical model of differential λ-calculus

Substitution: linear-non-linear multicategories Variable binding λx.s

(Fiore-Plotkin-Turi 1999, Hyland 2017)

  • Closed structure which allows to turn an operation with n + 1 inputs

to an operation with n inputs.

18

slide-69
SLIDE 69

Towards a multicategorical model of differential λ-calculus

Substitution: linear-non-linear multicategories Variable binding λx.s

(Fiore-Plotkin-Turi 1999, Hyland 2017)

  • Closed structure which allows to turn an operation with n + 1 inputs

to an operation with n inputs. Derivation u, x → Dxf (u)

  • Differential interaction nets

(Ehrhard-Regnier 2006)

  • Differential categories

(Blute-Cockett-Seely 2006)

18

slide-70
SLIDE 70

Towards a multicategorical model of differential λ-calculus

Substitution: linear-non-linear multicategories Variable binding λx.s

(Fiore-Plotkin-Turi 1999, Hyland 2017)

  • Closed structure which allows to turn an operation with n + 1 inputs

to an operation with n inputs. Derivation u, x → Dxf (u)

  • Differential interaction nets

(Ehrhard-Regnier 2006)

  • Differential categories

(Blute-Cockett-Seely 2006)

Chain rule

  • An additive structure due to

the derivation of the contraction.

18

slide-71
SLIDE 71

"The purpose of abstraction is not to be vague, but to create a new semantic level in which one can be absolutely precise".

(E. Dijkstra, The Humble Programmer, ACM Turing Lecture, 1972)

19