lipn.univ-paris13.frmazza/Elica/Slides/Elica161006-07-Vial.pdf · INTRODUCTION GARDNER/DE...

INTRODUCTION GARDNER/DE CARVALHO’S ITS M0 THE INFINITARY CALCULUS Λ001 TRUNCATION AND APPROXIMABILITY SEQUENCES AS INTERSECTION TYPES CONCLUSION

Infinitary Intersection Typesas Sequences

(A New Answer to Klop’s Problem)

Pierre VIALIRIF, Paris 7

Elica meeting, Bologna

October 7, 2016

INTRODUCTION

GARDNER/DE CARVALHO’S ITS M0

THE INFINITARY CALCULUS Λ001

TRUNCATION AND APPROXIMABILITY

SEQUENCES AS INTERSECTION TYPES

CONCLUSION

HEREDITARY HEAD-NORMALIZATION

I I Head Normal Forms (HNF): terms t of the form:

λx1 . . . xp.x u1 . . . uq (p, q > 0)

head variable head arguments

λx1 . . . xp.x u1 . . . uq (p, q > 0)

I A term is head-normalizing (HN) if it can be reducedto a HNF (in a finite number of steps)

λx1 . . . xp.x u1 . . . uq (p, q > 0)

I I Normal Forms (NF): induction

t ::= λx1 . . . xp.x t1 . . . tq (p, q > 0)

λx1 . . . xp.x u1 . . . uq (p, q > 0)

t ::= λx1 . . . xp.x t1 . . . tq (p, q > 0)

I A term is weakly normalizing (WN) if it can be re-duced to a NF (in a finite number of steps)

λx1 . . . xp.x u1 . . . uq (p, q > 0)

t ::= λx1 . . . xp.x t1 . . . tq (p, q > 0)

I A term is weakly normalizing (WN) if it can be re-duced to a NF (in a finite number of steps)

I Inductively, a term is WN if it is HN and all the headarguments are themselves WN.

λx1 . . . xp.x u1 . . . uq (p, q > 0)

I Coinductively, a term is hereditary head-normalizing(HHN) if it can be reduced to a HNF and all the headarguments are themselves HHN.

KLOP’S PROBLEM

I The set of HN terms (resp. WN) terms have been staticallycharacterized by various intersection type assignement systems(ITS).

I Klop’s Problem [early 90s]: can the set of HHN terms can becharacterized by an ITS ?

I Tatsuta [07]: an inductive ITS cannot do it.

I Can a coinductive ITS characterize the set of HHN terms?

KLOP’S PROBLEM

PURPOSES OF THIS TALK

I Present the key notions of truncations and approximability(meant to avoid irrelevant derivations).

I Understand why commutative intersection (here, Gardner/deCarvalho’s multiset intersection) is unfit to express those keynotions.

I Present the coinductive type assignment system S: intersectiontypes are sequences of types, instead of sets of types(idempotent intersection fw.) or multisets of types (regularnon-idempotent fw.).

INTRODUCTION

CONCLUSION

TYPING RULES OF M0 (GARDNER/DE CARVALHO)Types (τ, σi): τ, σi := α ∈X | [σi]i∈I → τ .

Context (Γ,∆): assigns intersection types to variables.

axx : [τ ] ` x : τ

Γ, x : [σi]i∈I ` t : τabs

Γ ` λx.t : [σi]i∈I → τ

Γ ` t : [σi]i∈I → τ (∆i ` u : σi )i∈I

appΓ +

∑i∈I

∆i ` t(u) : τ

Remark

I Multiset equality: [σ, τ, σ] = [σ, σ, τ ] 6= [σ, τ ]

I Multiplicative rules: accumulation of typing information .

I Possibility to forget the argument (empty multiset).

ALTERNATIVE PRESENTATION

Standard presentation

axx : [[α, β, α]→ α] ` x : [α, β, α]→ α

axx : [α] ` x : α

axx : [β] ` x : β

axx : [α] ` x : α

appx : [α, β, α, [α, β, α]→ α] ` xx : α

abs` λx.xx : [α, β, α, [α, β, α]→ α]→ α

[α, β, α]→ α α

λx.xx[α, β, α, [α, β, α]→ α]→ α

Alternative presentation

I Indicate the arity ofapplication rules.

I Indicate the types given inaxiom leaves.

I Compute the type of theterm.

[α, β, α]→ α

λx.xx

[α, β, α, [α, β, α]→ α]→ α

[α, β, α]→ α

λx.xx

[α, β, α, [α, β, α]→ α]→ α

[α, β, α]→ α

λx.xx[α, β, α, [α, β, α]→ α]→ α

[α, β, α]→ α

λx.xx[α, β, α, [α, β, α]→ α]→ α

Where does this α come from?

[α, β, α]→ α

λx.xx[α, β, α, [α, β, α]→ α]→ α

From this axiom rule?

[α, β, α]→ α

λx.xx[α, β, α, [α, β, α]→ α]→ α

From this axiom rule? Or this one?

SUBJECT REDUCTION PROPERTY FOR M0

If Π B Γ ` t : τ and t→ t′, then ∃Π′ B Γ ` t′ : τ

(λx.r)s→ r[s/x]

Γ ` r : τ, x : [σi]i∈I

( )i ∈ IΠi

∆i ` s : σi

absΓ ` λx.r : [σi]i∈I → τ

appΓ +

∑i∈I

∆i ` (λx.r)s : τ

(λx.r)s→ r[s/x]

Γ ` r : τ, x : [σi]i∈I

Axiom leavestyping x inside Πr

ax( x : [σi] ` x : )i∈Iσi ( )i ∈ IΠi

∆i ` s : σi

appΓ +

∑i∈I

∆i ` (λx.r)s : τ

(λx.r)s→ r[s/x]

Γ ` r : τ, x : [σi]i∈I

ax( x : [σi] ` x : )i∈Iσi ( )i ∈ IΠi

∆i ` s : σi

appΓ +

∑i∈I

∆i ` (λx.r)s : τ

(λx.r)s→ r[s/x]

Γ ` r : τ, x : [σi]i∈I

ax( x : [σi] ` x : )i∈Iσi “association”( )i ∈ IΠi

∆i ` s : σi

appΓ +

∑i∈I

∆i ` (λx.r)s : τ

(λx.r)s→ r[s/x]

Γ ` r : τ, x : [σi]i∈I

ax( x : [σi] ` x : )i∈Iσi “association”( )i ∈ IΠi

∆i ` s : σi

appΓ +

∑i∈I

∆i ` (λx.r)s : τ

(λx.r)s→ r[s/x]

Γ ` r : τ+∑i∈I

∆i [s/x]

( )i ∈ IΠi

∆i ` s :

Vocabulary:We say each association (between x-axiom leaves and arg-derivations)yields a derivation reduct Π′ typing r[s/x].

(λx.r)s→ r[s/x]

Γ ` r : τ+∑i∈I

∆i [s/x]

( )i ∈ IΠi

∆i ` s :

Observation:If a type σ occurs several times in [σi]i∈I , there can be several associations,each one yielding a possibly different derivation reducts Π′.

(λx.r)s→ r[s/x]

Γ ` r : τ+∑i∈I

∆i [s/x]

( )i ∈ IΠi

∆i ` s :

NORMALIZABILITY RESULTS

PropositionA term is HN iff it is typable in M0.

PropositionA term is WN iff it is typable in M0 by using an unforgetfuljudgment.

DefinitionA judgement Γ ` t : τ is unforgetful if there is no negativeoccurrence of [ ] in Γ and no positive occurrence of [ ] in τ .

I [ ] occurs negatively in [ ]→ τ

I If [ ] occurs negatively in σ2 then [ ] occurs positively in [σ1, σ2, σ3]→ τand so on.

INTRODUCTION

CONCLUSION

∞-TERMS

Variable x

Abstraction λx.u

Application u v

I Position: finite sequence in {0, 1, 2}∗, e.g. 0 · 0 · 2 · 1 · 2.

I Applicative Depth (a.d.): number of↗-edges e.g.

ad(1 · 2 · 2 · 0 · 2 · 1 · 2) = 4

∞-TERMS

Variable x

Abstraction λx.u

Application u v

ad(1 · 2 · 2 · 0 · 2 · 1 · 2) = 4

∞-TERMS

Variable x

Abstraction λx.u

Application u v

ad(1 · 2 · 2 · 0 · 2 · 1 · 2) = 4

001-TERMS

Λ001: the set of∞-terms t s.t.:

b is an infinite branch of t⇒ ad(b) =∞.

001-TERMS

f ω := f (f (f (. . .)))

i.e. fω = f (fω) (fixpoint)

Infinite rightwardbranch

001-TERMS

•b0 (leaf)

I Start fromb ∈ supp(t)

I Move ↑ or↖I A leaf b0 must

be reached

001-TERMS

•b0 (leaf)

I Move ↑ or↖

I A leaf b0 mustbe reached

001-TERMS

•b0 (leaf)

I Move ↑ or↖

001-TERMS

•b0 (leaf)

I Move ↑ or↖

001-TERMS

•b0 (leaf)I Start from

b ∈ supp(t)

I Move ↑ or↖I A leaf b0 must

be reached

STRONG CONVERGENCE

DefinitionA reduction sequence t0

b0→ t1b1→ t2

b2→ . . .bn−1→ tn

bn→ . . . is stronglyconverging if it is of finite length or if limad(bn) =∞.

STRONG CONVERGENCE

∆f := λx.f (xx) ∆f ∆f : ”Curry”

∆f ∆f → f (∆f ∆f )→ f 2(∆f ∆f )→ f 3(∆f ∆f )→ f 4(∆f ∆f )→ . . .→∞ fω

∆f ∆f

STRONG CONVERGENCE

∆f ∆f

STRONG CONVERGENCE

∆f ∆f

STRONG CONVERGENCE

∆f ∆f

STRONG CONVERGENCE

∆f ∆f

STRONG CONVERGENCE

Increaseof a.d.

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

UnstablePart

STRONG CONVERGENCE

Increaseof a.d.

StabilizedPart

STRONG CONVERGENCE

Conclusion

A strongly converging reduction sequence (s.c.r.s) allows us todefine its limit.

STRONG CONVERGENCE

ConclusionA strongly converging reduction sequence (s.c.r.s) allows us todefine its limit.

INFINITARY NORMALIZATION

I The notions of redex and head-normalizability do not change.

I The NF of Λ001 are generated by the coinductive grammar:

t = λx1 . . . λxp.x t1 . . . tq (p, q > 0)

Definition (Infinitary WN)A 001-term is WN if it can be reduced to a NF through at least ones.c.r.s.

I Thus, a (finite) term is HHN iff it is 001-WN.

t = λx1 . . . λxp.x t1 . . . tq (p, q > 0)

INTRODUCTION

CONCLUSION

TRUNCATION (FIGURES)

Π′ B Γ ` f ω : α

Every Variableis Typed

Γ = f : [[α]→ α]ω (infinite multiplicity)

[α]→ α

We can use the same derivation frame Π∗1 to type f (. . .)

UntypedSubterm

Γ1 = f : [[ ]→ α]

[ ]→ α

Π11 B Γ1 ` f (∆f ∆f ) : α

UntypedSubterm

Γ1 = f : [[ ]→ α]

[ ]→ α

∆f ∆f

Π21 B Γ1 ` f (f (∆f ∆f )) : α

UntypedSubterm

Γ1 = f : [[ ]→ α]

[ ]→ α

∆f ∆f

Π31 B Γ1 ` f 3(∆f ∆f ) : α

UntypedSubterm

Γ1 = f : [[ ]→ α]

[ ]→ α

∆f ∆f

Π41 B Γ1 ` f 4(∆f ∆f ) : α

UntypedSubterm

Γ1 = f : [[ ]→ α]

[ ]→ α

∆f ∆f

Π′1 B Γ1 ` f ω : α

UntypedSubterm

Γ1 = f : [[ ]→ α]

[ ]→ α

We can use the same derivation frame Π∗2 to typef (f (. . .))

UntypedSubterm

Γ2 = f : [[α]→ α] + [[ ]→ α]

[ ]→ α

[α]→ α

Π22 B Γ2 ` f (f (∆f ∆f )) : α

UntypedSubterm

Γ2 = f : [[α]→ α] + [[ ]→ α]

[ ]→ α

[α]→ α

∆f ∆f

Π32 B Γ2 ` f 3(∆f ∆f ) : α

UntypedSubterm

Γ2 = f : [[α]→ α] + [[ ]→ α]

[ ]→ α

[α]→ α

∆f ∆f

Π42 B Γ2 ` f 4(∆f ∆f ) : α

UntypedSubterm

Γ2 = f : [[α]→ α] + [[ ]→ α]

[ ]→ α

[α]→ α

∆f ∆f

Π′2 B Γ2 ` f ω : α

UntypedSubterm

Γ2 = f : [[α]→ α] + [[ ]→ α]

[ ]→ α

[α]→ α

We can use the same derivation frame Π∗3 to type f 3(. . .)

UntypedSubterm

Γ3 = f : [[α]→ α]2 + [[ ]→ α]

[ ]→ α

[α]→ α

Π33 B Γ3 ` f 3(∆f ∆f ) : α

UntypedSubterm

Γ3 = f : [[α]→ α]2 + [[ ]→ α]

[ ]→ α

[α]→ α

∆f ∆f

Π43 B Γ3 ` f 4(∆f ∆f ) : α

UntypedSubterm

Γ3 = f : [[α]→ α]2 + [[ ]→ α]

[ ]→ α

[α]→ α

∆f ∆f

Π′3 B Γ3 ` f ω : α

UntypedSubterm

Γ3 = f : [[α]→ α]2 + [[ ]→ α]

[ ]→ α

[α]→ α

We can use the same derivation frame Π∗4 to type f 4(. . .)

UntypedSubterm

Γ4 = f : [[α]→ α]3 + [[ ]→ α]

[ ]→ α

[α]→ α

Π44 B Γ4 ` f 4(∆f ∆f ) : α

UntypedSubterm

Γ4 = f : [[α]→ α]3 + [[ ]→ α]

[ ]→ α

[α]→ α

∆f ∆f

Π′4 B Γ4 ` f ω : α

UntypedSubterm

Γ4 = f : [[α]→ α]3 + [[ ]→ α]

[ ]→ α

[α]→ α

Π′ B Γ ` f ω : α

Every Variableis Typed

Γ = f : [[α]→ α]ω (infinite multiplicity)

[α]→ α

[α]→ α[ ]→ α

[α]→ α

Π′ can be truncated into Π′4:[α]→ α

[α]→ α

Π′ can be truncated into Π′4:[α]→ α

[α]→ α

Π′ can be truncated into Π′4:

[ ]→ α

[α]→ α

[ ]→ α

[α]→ α

[ ]→ α

[α]→ α

INFINITARY SUBJECT EXPANSION

I How do we perform∞-subject expansion on Π′ (typing fω)?

I Π′, that types fω , cannot be expanded (yet).I Π′n, that also types fω , cannot be expanded (yet).I But Πk

n, that types f k(∆f ∆f ), can be expanded.I Πk

n yields a derivation Πn typing ∆f ∆f (after k exp-steps).I We can build a “join” of the Πn, thus producing an infinite unforgetful

derivation Π typing ∆f ∆f .

I Derivation Π features a type γ coinductively defined by thefixpoint equation γ = [γ]ω → α.

I Type γ allows to type ∆∆. Need for a validity criterion.

I Π′, that types fω , cannot be expanded (yet).

I Π′n, that also types fω , cannot be expanded (yet).I But Πk

I Π′, that types fω , cannot be expanded (yet).I Π′n, that also types fω , cannot be expanded (yet).

I But Πkn, that types f k(∆f ∆f ), can be expanded.

I Πkn yields a derivation Πn typing ∆f ∆f (after k exp-steps).

I We can build a “join” of the Πn, thus producing an infinite unforgetfulderivation Π typing ∆f ∆f .

n, that types f k(∆f ∆f ), can be expanded.

I Πkn yields a derivation Πn typing ∆f ∆f (after k exp-steps).

n yields a derivation Πn typing ∆f ∆f (after k exp-steps).

APPROXIMABILITY (HEURISTIC)

I Informally, see a derivation Π as a set of symbols (type variablesα or→ that we found inside each jugdment of P).

I A (finite) approximation fΠ of a derivation Π is a finite subset ofsymbols of Π which is itself a derivation. We write fΠ 6 Π.

I A derivation Π is said to be approximable if for all finite subsetB of symbols of Π, there is an approximation fΠ 6 Π thatcontains B.

APPROXIMABILITY (FIGURE)

••

• fΠ

NON-DETERMINISM AND TRUNCATION

(λx.r)s

σ1· x#1

σ2· x#2

Truncation possibily af-fects every type nestedinside Π.

Assume σ1 = σ2.

I Possible in Π:#1 7→ Π2,#2 7→ Π1

I If fσ1 6= fσ2, not in fP.

Assume σ1 6= σ2

I Not possible in Π:#1 7→ Π2,#2 7→ Π1

I If fσ1 = fσ2, possible in fP.

(λx.r)s

fσ1· x#1

fσ2· x#2

Assume σ1 = σ2.

Assume σ1 6= σ2

(λx.r)s

fσ1· x#1

fσ2· x#2

Assume σ1 = σ2.

Assume σ1 6= σ2

(λx.r)s

σ1· x#1

σ2· x#2

Assume σ1 = σ2.

Assume σ1 6= σ2

(λx.r)s

σ1· x#1

σ2· x#2

Assume σ1 = σ2.I Possible in Π:

#1 7→ Π2,#2 7→ Π1

Assume σ1 6= σ2

(λx.r)s

fσ1· x#1

fσ2· x#2

Assume σ1 = σ2.I Possible in Π:

#1 7→ Π2,#2 7→ Π1

Assume σ1 6= σ2

(λx.r)s

σ1· x#1

σ2· x#2

Assume σ1 = σ2.

Assume σ1 6= σ2

(λx.r)s

σ1· x#1

σ2· x#2

Assume σ1 = σ2.

Assume σ1 6= σ2

(λx.r)s

fσ1· x#1

fσ2· x#2

Assume σ1 = σ2.

Assume σ1 6= σ2

INTRODUCTION

CONCLUSION

TYPES OF S

I Type (metavariable S,T): coinductive grammar

Sk,T ::= α ∈X | (Sk)k∈K → T

I Sequence Type:

I Intersection type replacing multiset types.I F = (Tk)k∈K where Tk types and K ⊂ N− {0, 1}.

I Example (Sequence Type): (Tk)k=2,3,8 with T2 = T8 = S and T3 = S′.

F = S S′ S8 3 2 argument tracks (2, 3, 8)

I Example (Arrow Type): .

S S′ S

→root ε

TYPES OF S

Sk,T ::= α ∈X | (Sk)k∈K → T

I Sequence Type:

I Intersection type replacing multiset types.I F = (Tk)k∈K where Tk types and K ⊂ N− {0, 1}.

S S′ S

→root ε

TYPES OF S

Sk,T ::= α ∈X | (Sk)k∈K → T

I Sequence Type:I Intersection type replacing multiset types.

I F = (Tk)k∈K where Tk types and K ⊂ N− {0, 1}.I Example (Sequence Type): (Tk)k=2,3,8 with T2 = T8 = S and T3 = S′.

S S′ S

→root ε

TYPES OF S

Sk,T ::= α ∈X | (Sk)k∈K → T

I Sequence Type:I Intersection type replacing multiset types.I F = (Tk)k∈K where Tk types and K ⊂ N− {0, 1}.

S S′ S

→root ε

TYPES OF S

Sk,T ::= α ∈X | (Sk)k∈K → T

S S′ S

→root ε

TYPES OF S

Sk,T ::= α ∈X | (Sk)k∈K → T

S S′ S

→root ε

TYPES OF S

Sk,T ::= α ∈X | (Sk)k∈K → T

I Example (Arrow Type): . . .→ T.

S S′ S

→root ε

TYPES OF S

Sk,T ::= α ∈X | (Sk)k∈K → T

I Example (Arrow Type): . . .→ T.

S S′ S

→root ε

Track 1 dedicated tothe⊕ side of arrow

TYPES OF S

Sk,T ::= α ∈X | (Sk)k∈K → T

I Example (Arrow Type): F→ T.

S S′ S

→root ε

TYPES OF S

Sk,T ::= α ∈X | (Sk)k∈K → T

I Example (Arrow Type): F→ T.

8 3 2 1

S S′ S T

→root ε

DERIVATIONS OF S

The set Deriv of rigid derivations is coinductively generated by:

axx : (T)k ` x : T

C ` t : T absC− x ` λx.t : C(x)→ T

C ` t : (Sk)k∈K → T (Dk ` u : Sk )k∈Kapp

C ∪⋃

k∈KDk ` t u : T

I For the app-rule: syntactic equality.

I Warning !If Rt(C) and the Rt(Dk) are not pairwise disjoint, contexts areincompatible.

I Parsing: premise of abs on tr. 0, left premise of app on tr. 1.

DERIVATIONS OF S

axx : (T)k ` x : T

C ` t : (Sk)k∈K → T (Dk ` u : S′k )k∈K′ (Sk)k∈K = (S′k)k∈K′app

C ∪⋃

k∈KDk ` t u : T

DERIVATIONS OF S

axx : (T)k ` x : T

C ∪⋃

k∈KDk ` t u : T

DERIVATIONS OF S

axx : (T)k ` x : T

C ∪⋃

k∈KDk ` t u : T

RIGID DERIVATION (EXAMPLE)

[α, β, α]→ α

λx.xx[α, β, α, [α, β, α]→ α]→ α

(8 · β, 5 · α, 4 · (8 · α, 5 · α, 3 · α), 2 · α)→ α

λx.xx

(8 · β, 5 · α, 4 · (8 · α, 5 · α, 3 · α), 2 · α)→ α

λx.xx

(8 · β, 5 · α, 4 · (8 · α, 5 · α, 3 · α), 2 · α)→ α

13 5 8

λx.xx

(8 · β, 5 · α, 4 · (8 · α, 5 · α, 3 · α), 2 · α)→ α

13 5 8

(8 · α, 5 · β, 3 · α)→ α

λx.xx

(8 · β, 5 · α, 4 · (8 · α, 5 · α, 3 · α), 2 · α)→ α

13 5 8

((8 · α, 5 · β, 3 · α)→ α)4

λx.xx

(8 · β, 5 · α, 4 · (8 · α, 5 · α, 3 · α), 2 · α)→ α

13 5 8

((8 · α, 5 · β, 3 · α)→ α)4

λx.xx(8 · β, 5 · α, 4 · (8 · α, 5 · α, 3 · α), 2 · α)→ α

MAIN FEATURES

I Subject reduction is deterministic:

I Assume P types (λx.r)s. If there is an axiom rule typing x on track 5(#5-ax), by typing constraint, there will also be an argument derivation P5

typing s on track 5, concluded by exactly the same type S5I During reduction, #5-ax will be replaced by P5, even if there are other Pk

concluded by S = S5

I Trackability: every symbol used inside P can be pointed atunivocuously. Notion of biposition.

MAIN FEATURES

typing s on track 5, concluded by exactly the same type S5

I During reduction, #5-ax will be replaced by P5, even if there are other Pk

concluded by S = S5

MAIN FEATURES

concluded by S = S5

MAIN FEATURES

concluded by S = S5

BISUPPORT OF A DERIVATION

P A := supp(P) ⊂ N∗

C ` t : T•

pos. a

C ` t : T•

pos. a

Right biposition (a, c)

Right of `I c ∈ supp(T)

I (Tk)k∈K := C(x)I k ∈ K and c ∈ supp(Tk).

C ` t : T•

pos. a

C ` t : T•

pos. a

Left Biposition (a, x, k · c)Left of `

I x variable.

I (Tk)k∈K := C(x)I k ∈ K and c ∈ supp(Tk).

C ` t : T•

pos. a

I x variable.I (Tk)k∈K := C(x)

I k ∈ K and c ∈ supp(Tk).

C ` t : T•

pos. a

I x variable.I (Tk)k∈K := C(x)I k ∈ K and c ∈ supp(Tk).

C ` t : T•

pos. a

Bisupport of P: the set of (right or left) bipositions

APPROXIMABILITY

I Every symbol inside a rigid derivation P has a biposition (aposition pointing inside a type nested in a judgment of P).

I A finite part B of P is finite subset of bisupp(P).

I A finite approximation of P is a (finite) derivation induced by Pon a finite part of P.

I A rigid derivation P is said to be approximable if for all finitepart B of P, there is a finite approximation fP 6 P s.t. fP containsB.

APPROXIMABILITY

CHARACTERIZATION OF INFINITARY WN

TheoremA 001-term t is WN iff t is unforgetfully typable by means of anapproximable derivation.

Argument 1: If a term is typable by an approximable derivation, thenit is head normalizing. Unforgetfulness makes HN hereditary.Argument 2: Subject reduction holds for s.c.r.s. (with or withoutapproximability condition).Argument 3: Every NF can be typed by quantitative unforgetfulderivations and every quantitative derivation typing a NF isapproximable.Argument 4: Subject expansion property holds for s.c.r.s. (assumingapproximability only).

Argument 1: If a term is typable by an approximable derivation, thenit is head normalizing. Unforgetfulness makes HN hereditary.

Argument 2: Subject reduction holds for s.c.r.s. (with or withoutapproximability condition).Argument 3: Every NF can be typed by quantitative unforgetfulderivations and every quantitative derivation typing a NF isapproximable.Argument 4: Subject expansion property holds for s.c.r.s. (assumingapproximability only).

Argument 1: If a term is typable by an approximable derivation, thenit is head normalizing. Unforgetfulness makes HN hereditary.Argument 2: Subject reduction holds for s.c.r.s. (with or withoutapproximability condition).

Argument 3: Every NF can be typed by quantitative unforgetfulderivations and every quantitative derivation typing a NF isapproximable.Argument 4: Subject expansion property holds for s.c.r.s. (assumingapproximability only).

Argument 1: If a term is typable by an approximable derivation, thenit is head normalizing. Unforgetfulness makes HN hereditary.Argument 2: Subject reduction holds for s.c.r.s. (with or withoutapproximability condition).Argument 3: Every NF can be typed by quantitative unforgetfulderivations and every quantitative derivation typing a NF isapproximable.

Argument 4: Subject expansion property holds for s.c.r.s. (assumingapproximability only).

Argument 1: If a term is typable by an approximable derivation, thenit is head normalizing. Unforgetfulness makes HN hereditary.Argument 2: Subject reduction holds for s.c.r.s. (with or withoutapproximability condition).Argument 3: Every NF can be typed by quantitative unforgetfulderivations and every quantitative derivation typing a NF isapproximable.Argument 4: Subject expansion property holds for s.c.r.s. (assumingapproximability only).

INTRODUCTION

CONCLUSION

FUTURE WORK?

I Representation Theorem: every M -derivation is the collapse ofa S-derivation (already done, HOR 2016).

I Can we reformulate approximability ?

I Can infinitary Strong Normalization be characterized ?

I Is every term typable in S (without approximability) ?Yes ! S is completely unsound (difficult because of relevance).

I Categorical Adaptation of this framework (ongoing work with D.Mazza and L. Pellisier).

QUESTIONS

Thank you for your attention !

EXPANDED DERIVATIONS

[α]→ α([ ]→ α if n = 1)

γn−1 γn−2 γ1

γn−1 γ1

f @n−2

Ψn−1 Ψ1

I Ψn is the left subderivation.I γ1 = [ ]→ α and γn = [γi]16i6n−1 → α.I Πn is the n-expanded of Πn

n, the (n + 1)-expanded of Πn+1n ,..., the

∞-expanded of Π′n.

EXPANDED DERIVATIONS

[α]→ α

γ γ γ

I Ψ is the left subderivation.I γ = [γ]ω → α.

I Πn is the n-expanded of Πnn, the (n + 1)-expanded of Πn+1

n ,..., the∞-expanded of Π′n.

SYSTEM M

I The relation ≡ (between types or seq. types) is definedcoinductively:

I α ≡ α.I (Sk)k∈K → T ≡ (S′k)k∈K′ → T′ if (Sk)k∈K ≡ (S′k)k′∈K′ and T ≡ T′.I (Sk)k∈K ≡ (S′k)k∈K′ if there is a bijection ρ : K→ K′ s.t. ∀k ∈ K, Sk ≡ S′

σ(k).

I We set TypesM = Types / ≡.The set of multiset types is STypes / ≡.

I To obtain System M , take the rules of M0 coinductively (withthose types and multiset types).

SYSTEM M

I α ≡ α.

I (Sk)k∈K → T ≡ (S′k)k∈K′ → T′ if (Sk)k∈K ≡ (S′k)k′∈K′ and T ≡ T′.I (Sk)k∈K ≡ (S′k)k∈K′ if there is a bijection ρ : K→ K′ s.t. ∀k ∈ K, Sk ≡ S′

σ(k).

SYSTEM M

I α ≡ α.I (Sk)k∈K → T ≡ (S′k)k∈K′ → T′ if (Sk)k∈K ≡ (S′k)k′∈K′ and T ≡ T′.

I (Sk)k∈K ≡ (S′k)k∈K′ if there is a bijection ρ : K→ K′ s.t. ∀k ∈ K, Sk ≡ S′σ(k).

SYSTEM M

σ(k).

SYSTEM M

σ(k).

SYSTEM M

σ(k).

lipn.univ-paris13.frmazza/Elica/Slides/Elica161006-07-Vial.pdf · INTRODUCTION GARDNER/DE...

Documents

Transcript of lipn.univ-paris13.frmazza/Elica/Slides/Elica161006-07-Vial.pdf · INTRODUCTION GARDNER/DE...

NF-κB–inducing kinase plays an essential T cell ... · Susan E. Murray, 1 Fanny Polesso, Alexander M. Rowe,1 Soumen Basak,2 Yoshinobu Koguchi, Katelynne Gardner Toren,1 Alexander

Geometric Tomography - American Mathematical Society · Geometric Tomography R. J. Gardner 422 N OTICES OF THE AMS V OLUME 42, NUMBER 4 T omography, from the Greek ˝´o o& , a slice,

Percorsi di chimica organica - Soluzioni degli esercizi ... · 22. La prolina non è presente nei tratti ad α-elica delle proteine a causa dell’anello rigido di cui fa parte l’azoto,

Introduction to Plane-Wave Basis Sets and …Plane-Wave Basis Sets There is a further truncation of plane wave expansion in plane-wave calculations. Namely, only the reciprocal lattice

Modelling Dynamic Web Data · Modelling Dynamic Web Data Philippa Gardner Sergio Maﬀeis⁄ Department of Computing Imperial College London fpg,maffeisg@doc.ic.ac.uk October 2003

Θεωρίες Μάθησης Μάθηση είναι η απόκτηση και η διατήρηση ...Burrhus Frederic Skinner . Albert Bandura . Howard Gardner . 2 ... Skinner

Myoglobin and Iron in Lamb MeatMyoglobin and Iron in Lamb Meat …the influence of lamb growth Khama Kelman, L. Pannier, D.W. Pethick, G.E. Gardner School of Veterinary and Life Sciences,

Uriel Frisch Observatoire de la Côte d’Azur, Nice · Galerkin truncation, hyperviscosity and bottlenecks Uriel Frisch Observatoire de la Côte d’Azur, Nice with W. Pauls (Goettingen),

Φλαμίνγκο john gardner

STAT 7032 Probability CLT part - UC Homepagesbrycwz/classes/7032/Grad-Prob-2020-slides.pdfI Truncation makes variances nite: X k = X kI ... A basic remedy for sums of dependent random

Data Overview at ICBR We have Σ O( ) Aaron Gardner ( agardner@ufl.edu )

20/16-bit AGC Architecture · Web viewP01 Double precision multiply, MPAC * memory, preserving max truncation. R02 triple precision. ... 11 01244 RED OperAddr Go to big-end address.

Harmonic Coordinate Finite Element Method for Acoustic Waves · coordinate map to create new, regular grid elements. Sub-optimal convergence due to element truncation. Binford 2011:

Modeling stable One-types - Niles Johnson · Modeling stable One-types Ang elica Osorno University of Chicago January 7, 2012 joint with Niles Johnson Ang elica Osorno (UofC) Stable

krzakala slides partII · E. Gardner, B. Derrida (1989)] Teacher - Student Framework Simplest version

RNA non-codificanti indotti da danno al DNA regolano il ...amsdottorato.unibo.it/4746/1/bertozzi_davide_tesi.pdf · elica del DNA deve essere completamente svolta, e durante la trascrizione

Δημιοʑργικές εργασίεςusers.sch.gr/achrono/wordpress/wp-content/uploads/2018/...•Gardner: «Ένα δημιοργικό άομο λύνει προβλήμαα,

The somatosensory system codes ﬁve major …david/courses/sm12/Lectures/gardner/touch2012.pdf• The somatosensory system codes ﬁve major sensory modalities:! 1. ... free nerve

Neutron Decay Correlations - int. · PDF fileNeutron Decay Correlations 1 ... and Tulin(2007) Bhattacharya et al. (2012) ... ‘a’ vs. ‘A’ recoil terms: Gardner & Zhang (2001)

Geometric Tomography · 1999-04-22 · Geometric Tomography R. J. Gardner 422 N OTICES OF THE AMS V OLUME 42, NUMBER 4 T omography, from the Greek ˝´o o& , a slice, is by now an