Lecture 3: Homotopical models of type theory Nicola Gambino School - - PowerPoint PPT Presentation

lecture 3 homotopical models of type theory
SMART_READER_LITE
LIVE PREVIEW

Lecture 3: Homotopical models of type theory Nicola Gambino School - - PowerPoint PPT Presentation

Lecture 3: Homotopical models of type theory Nicola Gambino School of Mathematics University of Leeds Young Set Theory Copenhagen June 14th, 2016 1 What happened yesterday? Type theory the type theory T Extensional vs intensional


slide-1
SLIDE 1

1

Lecture 3: Homotopical models of type theory

Nicola Gambino

School of Mathematics University of Leeds

Young Set Theory Copenhagen June 14th, 2016

slide-2
SLIDE 2

2

What happened yesterday?

Type theory

◮ the type theory T ◮ Extensional vs intensional type theories

Homotopical algebra

◮ Weak factorisation systems and model structures ◮ Groupoids ◮ Simplicial sets

Italy 2 – Belgium 0 Exercises

◮ x, y : A, u : IdA(x, y) ⊢ u−1 : IdA(y, x) ◮ x, y, z : A, u : IdA(x, y), v : IdA(y, z) ⊢ v ◦ u : IdA(x, z).

slide-3
SLIDE 3

3

Problems with intensionality

The axioms for identity types do not seem to capture fully what we want. Example

◮ we have

IdA×B( c, d ) ← → IdA

  • π1(c), π1(d)
  • × IdB
  • π2(c), π2(d)
  • ◮ but only

IdA→B(f , g) − → (Πx : A) IdB(fx, gx) Similar situation for Σ-types and Π-types. What about the type universe?

slide-4
SLIDE 4

4

Outline of Lecture 2

Part I: Models of type theory Part II: Identity types Part III: Π-types Part IV: Universes

slide-5
SLIDE 5

5

Part I: Models of type theory

slide-6
SLIDE 6

6

Models of type theories

Question: What structure on a category C do we need to have a model of T ? Idea: Look at the structure of the syntactic category of T.

◮ Objects: contexts

(x1 : A1) , (x1 : A1, x2 : A2, . . .) , . . .

◮ Maps: terms-in-context, e.g.

(a1) : Γ → (x1 : A1) if Γ ⊢ a1 : A1 (a1, a2) : Γ → (x1 : A1, x2 : A2) if Γ ⊢ a1 : A1 , Γ ⊢ a2 : A2[a1/x1] · · · Note. Γ ⊢ A : type = ⇒ (Γ, x : A)

pA

  • Γ
slide-7
SLIDE 7

7

Axiomatization

Fix

◮ category C with a terminal object 1 ◮ a class P ⊆ Map(C)

Type Theory Syntactic category Category Theory A : type (x : A)

pA

  • ( )

1.A

pA∈P

  • 1

Γ ⊢ A : type (Γ, x : A)

pA

  • Γ

Γ.A

pA∈P

  • Γ
slide-8
SLIDE 8

8

Terms as sections

Type Theory Category Theory Γ ⊢ a : A Γ

a

  • Γ.A

pA

  • Γ

a : A 1

a

A

slide-9
SLIDE 9

9

Substitution as pullback

Type Theory Category Theory Γ ⊢ a : A Γ, x : A ⊢ B : type Γ ⊢ B[a/x] : type Γ.B[a/x]

  • Γ.A.B

pB

  • Γ

a

Γ.A

Γ ⊢ a : A Γ, x : A ⊢ b : B Γ ⊢ b[a/x] : B[a/x] Γ

b[a/x]

  • b◦(1Γ,a)
  • Γ.B[a/x]
  • Γ.A.B

pB

  • Γ

a

Γ.A

slide-10
SLIDE 10

10

Weakening

For Γ ∈ Obj(C), let P/Γ be the category with

◮ objects: P-maps A : Γ.A → Γ ◮ maps: commutative triangles

Γ.A

f

  • pA
  • Γ.B

pB

  • Γ

Γ, x : A ⊢ f (x) : B Let pA : Γ.A → Γ be in P. Pullback Γ.A.E

  • Γ.E

pE

  • Γ.A

pA

Γ

Γ ⊢ E : type Γ, x : A ⊢ E : type This is the ‘weakening functor’ ∆A : P/Γ → P/Γ.A.

slide-11
SLIDE 11

11

General setting

Let

◮ C be a category ◮ P ⊆ Map(C)

and assume

◮ C has a terminal object. ◮ The pullback of a P-maps along any maps exists and is an P-map.

Question

◮ What additional structure on P do we need to interpret type-formers?

slide-12
SLIDE 12

12

Part II: Identity types

slide-13
SLIDE 13

13

Identity types (I)

For simplicity, let us assume Γ = 1 and work with pA : 1.A → 1. We need

  • 1. a P-map

qA : IdA → A × A

  • 2. a factorisation

A

reflA

  • ∆A
  • IdA

qA

  • A × A
  • 3. a diagonal filler for every commutative diagram

A

reflA

  • d

IdA.E

pE ∈P

  • IdA
  • IdA
  • Note. In fact, reflA ⋔ p for all p ∈ P.
slide-14
SLIDE 14

14

Identity types as path spaces

Idea p : IdA(a, b) ⇐ ⇒ p is a path in A from a to b Note

◮ This explains several aspects of the behaviour of identity types

  • Theorem. The syntactic category of the type theory T admits a weak

factorisation system (L, R), where L = {i | (∀p ∈ P) i ⋔ p} R = {p | (∀i ∈ L) i ⋔ p} where P is the set of projections pA : (Γ, x : A) → Γ.

slide-15
SLIDE 15

15

Homotopical models of type theory

Idea

◮ Take P = Fib in some model structure (Weq, Fib, Cof).

We get a ‘dictionary’ Type Theory Homotopical algebra A : type fibrant object A x : A ⊢ B(x) : type fibration p : B → A x, y : A ⊢ IdA(x, y) path space of A (Πx : A)B(x) space of sections of p type universe U a fibrant object U x ∈ U ⊢ El(x) : type a generic fibration π : ˜ U → U

slide-16
SLIDE 16

16

Example: Id-types in groupoids

Given by the (Weq ∩ Cof, Fib)-factorisation of diagonal A

r

  • ∆A

A × A

AJ

(s,t)

  • The groupoid AJ has

◮ objects: maps α : a0 → a1 in A ◮ maps: squares

a0

  • α
  • b0

β

  • a1

b1

Note Uniqueness of identity proofs fails in this model.

  • Warning. To define a model, one needs to take care of further aspects:

◮ mere existence vs structure ◮ coherence with respect to pullback (substitution)

slide-17
SLIDE 17

17

Part III: Π-types

slide-18
SLIDE 18

18

Π-types

Let A : Γ.A → Γ be in P.

◮ Recall the ‘weakening’ functor

∆A : P/Γ → P/Γ.A Γ ⊢ E : type Γ, x : A ⊢ E : type

◮ To interpret Π-types, we require a right adjoint

ΠA : P/Γ.A → P/Γ Γ, x : A ⊢ B : type Γ ⊢ (Πx : A)B : type Idea Γ.A

b

  • Γ.A.B
  • Γ.A

⇐ ⇒ Γ

λ(b)

  • Γ.ΠA(B)
  • Γ
slide-19
SLIDE 19

19

Π-types in groupoids

Theorem.

◮ For p : Γ.A → Γ an isofibration, the pullback functor

∆p : Gpd/Γ → Gpd/Γ.A has a right adjoint Πp : Gpd/Γ.A → Gpd/Γ

◮ Furthermore, the right adjoint preserves isofibrations, and thus gives us

Πp : Fib/Γ.A → Fib/Γ Γ.A.B

q

Γ.A

→ Γ.ΠA(B)

Πp(q) Γ

  • Example. When Γ = 1, the objects of the groupoid ΠA(B) are

A

b

  • 1A
  • A.B

q

  • A

i.e. sections of q.

slide-20
SLIDE 20

20

Π-types in simplicial sets

Theorem.

◮ For any map p : B → A, pullback along p has a right adjoint

Πp : SSet/B → SSet/A

◮ If p is a Kan fibration, then Πp preserves Kan fibrations, and hence gives

Πp : Fib/B → Fib/A

  • Proof. By duality, it suffices to show that

∆p : SSet/A → SSet/B preserves (Weq ∩ Cof)-maps. But

◮ Cof-maps are monomorphisms, so are always preserved. ◮ The preservation of weak equivalences by pullback along Kan fibrations is

the so-called right properness of the model structure.

  • Note. Constructivity issues.
slide-21
SLIDE 21

21

Part IV: Universes

slide-22
SLIDE 22

22

Generic P-maps

We need a notion of ‘smallness’ for P-maps, e.g. fibers having cardinality < κ. Then need a P-map π : ˜ U → U that weakly classifies ‘small’ P-maps, i.e. for every such p : B → A there exists a pullback diagram B

  • p
  • ˜

U

π

  • A

U .

  • Note. Given a : 1 → U, we can form a pullback

El(a)

  • ˜

U

π

  • 1

a

U

We think of a as the ‘name’ in U of the object El(a).

slide-23
SLIDE 23

23

The type universe in groupoids and simplicial sets

Fix an inaccessible cardinal κ.

◮ In Gpd, it is not difficult to construct a universe. For example, one can

consider the groupoid of all small (discrete) groupoids.

◮ In SSet, there exists a fibration

π : ˜ U → U that weakly classifies fibrations with fibers of cardinality < κ, i.e. for every such p : B → A there exists a pullback diagram B

  • p
  • ˜

U

π

  • A

U .

Here, Un = { p : B → ∆n | p Kan fibration }

  • Problem. But U needs to be fibrant!
slide-24
SLIDE 24

24

The type universe in simplicial sets

Theorem.

◮ The base U of the generic Kan fibration π : ˜

U → U is a Kan complex.

  • Proof. We need to show that U is a Kan complex. So show

Λn

k ∀b

  • hn

k

  • U

∆n

∃b′

  • This reduces to the problem of extending fibrations along horn inclusions:

B

  • p
  • B′
  • Λn

k hn

k

∆n

This can be done using the theory of minimal fibrations (AC).

slide-25
SLIDE 25

25

Minimal fibrations extend along (Weq ∩ Cof)-maps

Lemma 1. Let

◮ m : X → A be a minimal fibration ◮ i : A → A′ be a (Weq ∩ Cof)-map.

Then there exists m′ : M′ → A′ a minimal fibration and X

j

  • m
  • X ′

m′

  • A

i

A′

slide-26
SLIDE 26

26

(Weq ∩ Fib)-maps can be extended along cofibrations

Lemma 2. Let

◮ t : B → X be a (Weq ∩ Fib)-map ◮ j : X → X ′ a cofibration

Then there exists t′ : E ′ → X ′ a (Weq ∩ Fib)-map and B

  • t
  • B′

t′

  • X

j

X ′

slide-27
SLIDE 27

27

Proof of the theorem

Recall that we need to complete the diagram B

  • p
  • B′
  • Λn

k hn

k

∆n

It suffices to

◮ factor p as a (Weq ∩ Fib)-map t followed by a minimal fibration m ◮ apply Lemma 1 and Lemma 2 so as to get

B

  • t
  • B′

t′

  • X

j

  • m
  • X ′

m′

  • Λn

k hn

k

∆n

slide-28
SLIDE 28

28

Conclusions

The homotopical models of type theory suggest:

  • 1. To use type theory as a language for speaking about spaces
  • 2. To develop mathematics using this language; in particular, to define

sets =def discrete spaces

  • 3. To add axioms to type theory motivated by homotopy theory