• Aucun résultat trouvé

Approximations and η-rule

N/A
N/A
Protected

Academic year: 2022

Partager "Approximations and η-rule"

Copied!
5
0
0

Texte intégral

(1)

Lambda-Calculus (III-6)

[email protected] Tsinghua University, September 17, 2010

Plan

Bohm trees -- reminders

Morris extensional equivalence

Bohm trees and !-rule

Observational equivalences

Relation with Scott’s models

Bohm tree semantics - reminders

Theorem [continuity] For allb∈N such thatb�C[M], thenb�C[a]

for somea∈N such thata�M.

Theorem [monotony] M�NimpliesC[M]�C[N]

Theorem [!-theory] M≡NimpliesC[M]≡C[N]

Proofs: easy consequences of previous proofs.

Approximations and

η-rule

(2)

Bohm tree and !-rule

We need take !-rule into account ! How to mix !-rule and Bohm tree construction ?

Functional extensionality has not been considered since we can have:

MP≡NP for allP, butM�≡N.

(TakeM=xandN=λy.xy)

We take for granted that ηand β,ηare confluent.

Moreover η strongly normalizes.

The prefix ordering between approximants must be extended. For instance:

xΩ=ηλy.xΩy≤λy.xyy λx.yΩ≤λx.yx=ηy

Finite approximants

Lemma : We have following commutation properties:

We consider the set of !-normal forms of finite approximants with following relation:

Ne

a≤eb iff a(≤ ∪=η)b

η≤ ⊂ ≤ η

ηη

η ηη η

Corollary: a≤eb iff a ηa≤b η b

Examples:

λy.xΩ ≤λy.xy ηx xΩ ηλy.xΩy ≤ λy.xyy

Extensional Bohm trees

Definition : Let ωe(M) be theη-normal form ofω(M).

Definition : The extensional Bohm tree BTe(M) ofMis defined by:

BTe(M) ={a∈Ne | a ≤e ωe(N), M N}

M�eN iff BTe(M)⊂BTe(N) M≡eN iff BTe(M) = BTe(N)

Definition : Extensional Bohm tree semantics

Extensional Bohm trees

Theorem : eis a monotonic semantics and≡eforms aλ-theory.

Theorem [Hyland, 1975]:

M�eN iff for allC[ ],C[M] n.f. impliesC[N] n.f.

J.Morris extensional equivalence

MIT-1968

(3)

Closed Bohm trees

Finite Bohm trees revisited

We just added !-rule to the standard Bohm tree construction with completion by ideals.

However, we forgot slight difficulty: now, thanks to !-rule, finite Bohm tree now dominates an infinite number of other finite Bohm trees. See:

a0=Ω a1=λx1.xΩ a2=λx1.x(λx2.x1Ω) a3=λx1.x(λx2.x1(λx3.x2Ω))

I0={Ω}

I1={a∈Ne | a≤ea1} I2={a∈Ne | a≤ea2} I3={a∈Ne | a≤ea3} I={a∈Ne | a≤ex}

n=0In

a=x

new point

added by ideal completion

Finite Bohm trees revisited

There are two ways of completing finite Bohm trees.

standard completion by ideals (what we did)

completion with closed ideals (does not add new limit point) 1-

2-

BTec(M) = cl(BTec(M))

We define the closure of directed sets as being the set with its already existing limit in Ne

We therefore have:

I≡eclJ where J=Y(λfxy.x(fy))

and normal forms are no longer isolated points.

Finite Bohm trees revisited

The equality between I and J is not so unnatural since one may emulate infinite !- expansion with the "-rule:

I=λx.x η λxx1.xx1

η λxx1.x(λx2.x1x2)

η λxx1.x(λx2.x1(λx3.x2x3))

η ...

J=Y(λfxx1.x(fx1)) =β λxx1.x(Jx1)

=β λxx1.x(λx2.x1(Jx2))

=β λxx1.x(λx2.x1(λx3.x2(Jx3)))

=β ...

.

same phenomenon as for these two versions of identity on natural numbers:

I(n) ⇐ n

J(n) ⇐ ifn= 0then1else1 +J(n−1) .

(4)

Bohm trees and Scott’s models

We have following correspondances:

1- 2-

M�ecN iff M�DN M�N iff M�TωN

(Scott’s model) (Plotkin’s model)

One can also show that:

3- M�e+N iff M�N (Scott’s model)

where�e+ is Bohm tree construction from η≤ordering onNe.

finally, one may order Bohm trees with symmetrical:

4- M�eN

where�eis Bohm tree construction from≤ η ordering onNe.

Observational equivalences

To conclude we have the following results:

1- 2-

M�ecN iff for all contextsC[ ], C[M] hnf implies C[N] hnf M�eN iff for all contextsC[ ], C[M] nf implies C[N] nf

Making observational equivalences with other Bohm tree semantics is more difficult since one has to fight with !-equality. Take for instanceλx.xx andλx.x(λy.xy)

Homeworks

Exercices

1-Show that ifM has no hnf, thenMis totally undefined.

2- ShowΩM≡Ωandλx.Ω≡Ω. Show thatM ωN, thenM≡N.

4- ShowM�≡λx.Mxwhenx�∈var(M). What ifM≡λx.M1?

5- LetY0=Y,Yn+1=Yn(λxy.y(xy)). Show thatY ≡Ynfor alln. However allYn

are pairwise non interconvertible.

6 If M ≤ P andN ≤ P (M andN are prefix compatible), then BT(M�N) = BT(M)∩BT(N). (Thus BT is stable in Berry’s sense, 1978). What if not com- patible ?

3- FindM andN such thatMP ≡NP for allP, butM�≡N. (Meaning that≡is not extensional)

(5)

Exercices

7- [Barendregt 1971]

A closed expressionM (i.e. var(M) =∅) is solvable iff:

∀P,∃N1,N2, ...Nnsuch thatMN1N2· · ·Nn=βP (in short:

∀P,∃N,� MN� =βP )

Show that for every closed termM, the following are equivalent:

1. M has a hnf

2. ∃N,� MN� has a normal form 3. ∃N,� MN� =βI

4. M is solvable 8- [Barendregt 1974]

Show that, in theλI-calculus, a termM is solvable iffit has a normal form.

Exercices

9- LetRbe a preorder onN (reflexive + transitive) compatible with its structure:

a1Rb1, ...anRbn implies xa1a2· · ·an

aRb implies λx.aRλx.b

LetM�RN iff ∀a∈BT(M), ∃b∈BT(N), aRb Show that whenM is a closed term, one has:

∀P�,M�P�RNP� iff ∀C[ ], C[M]�RC[N]

10- (cont’d 1) LetMRNbe “ifM has a normal form, thenNhas a normal form”

Give examples ofM andNsuch thatM�RNbutM��N.

(cont’d 2) LetMRNbe “ifMhas a hnf, thenN has a hnf”

Give examples ofMandN such thatM�RNbutM��N.

11-

12- (cont’d 2) LetMRN be “ifMhas a hnf, thenNhas a similar hnf”

Give examples of M andN such that M �R N butM �� N. (Hint: consider M=λx.xxandN=λx.x(λy.xy)) [Compare withHyland 1975])

Références

Documents relatifs

L'effet Aharonov-Bohm est un phénomène quantique décrit en 1949 par Ehrenberg et Siday et redécouvert en 1959 par David Bohm et Yakir Aharonov.. Il décrit le paradoxe

Avant même la démonstration de la violation des inégalités de Bell qui interdisent les théories locales à variables cachées, cette interprétation non seulement supposait

In the second paragraph general relations of a certain kind are introduced; in the next paragraph these relations are used to define various notions of finiteness

The solver is given in the form of 16 rewriting rules which transform any first-order constraint ϕ into an equivalent disjunction φ of simple formulas such that φ is either the

Partie II : Tirages avec arrˆ et d` es qu’une boule blanche et une boule noire ont ´ et´ e obtenues Dans cette partie, on effectue des tirages successifs avec remise et on

This position paper discusses approaches and challenges for using those instances as input for machine learning algo- rithms, with a focus on rule learning algorithms, as a means

These functionalities are available as sub-tabs within the plug-in: (1) a Rule Graph tab that provides a graph structure to browse and explore SWRL rule base and rule

From the point of view of this workshop, the most important balance is be- tween giving freedom of action to the learner, which recognises the importance of students’ autonomy