prev

next

out of 26

View

212Download

0

Embed Size (px)

Acyclic Solos and Differential Interaction Nets

Thomas Ehrhard

Thomas.Ehrhard@pps.jussieu.fr

Preuves Programmes Systemes

CNRS Universite Paris 7

France

Olivier Laurent

Olivier.Laurent@pps.jussieu.fr

Preuves Programmes Systemes

CNRS Universite Paris 7

France

October 20, 2008

Abstract

We present a restriction of the solos calculus which is stable under reduction and expressiveenough to contain an encoding of the pi-calculus. As a consequence, it is shown that equalizingnames that are already equal is not required by the encoding of the pi-calculus. In particular,the induced solo diagrams bear an acyclicity property that induces a faithful encoding intodifferential interaction nets. This gives a (new) proof that differential interaction nets areexpressive enough to contain an encoding of the pi-calculus.

All this is worked out in the case of finitary (replication free) systems without sum, matchnor mismatch.

Keywords: solos calculus, pi-calculus, prefix, typing, differential interaction nets.

ACM classification: F.3.2 Semantics of Programming Languages Process models ; F.4.1Mathematical Logic Lambda calculus and related systems; F.1.2 Modes of Computation Parallelism and concurrency.

The question of extending the Curry-Howard correspondence (between the -calculus and intu-itionistic logic) to concurrency theory is a long-standing open problem. We developed in a previouspaper [EL08] a translation between the -calculus [MPW92] and differential interaction nets [ER06].This has shown that differential linear logic a logical system whose sequent calculus has beenobtained by the first author from a precise analysis of some denotational models of linear logicbased on vector spaces [Ehr02] is a reasonable candidate for a Curry-Howard correspondencewith concurrent computation, since differential interaction nets appear to be expressive enough torepresent key concurrency primitives as represented in the -calculus.

Let us tell a bit more about the genesis of this translation. We discovered the notion of com-munication areas as particular differential interaction nets able to represent some communicationprimitives. However, due to the very asynchronous flavour of differential interaction nets, it wasnot immediate that additional constructions such as prefixing could be easily encoded. This led usto have a look at the solos calculus.

The solos calculus [LV03] has allowed to prove how action prefixes (thus sequentiality con-straints) can be encoded in a calculus without prefixes. This is done by encoding the -calculus

Partially supported by the French ANR project CHOCO.

1

into the solos calculus. These two calculi differ on the way they handle name passing. As in thefusion calculus [PV98], the solos calculus defines communication by unification of names, whereasthe -calculus uses substitution of a bound name with another one.

Even if their behaviours are very similar, there is a mismatch between the solos calculus andcommunication areas with respect to the identification of a name x with itself during reduction.Note that such an identification never occurs in the -calculus. In the solos calculus, if x has tobe identified with itself, it is just considered as a dummy operation since we already have x = x.Communication areas on their side keep track of this identification through an explicit link (whichis not erased by reduction) connecting the communication area associated with x to itself.

We then had two possibilities: to follow the intuition coming from the translation of the -calculus into the solos calculus to define a translation of the -calculus into differential interactionnets (this is the methodology of [EL08]), or to find a fragment of the solos calculus which isexpressive enough to contain the image of the -calculus but which does not rely on behavioursrepresented differently in differential interaction nets (this is the goal of the present paper).

From this point of view, the main result of this work is to provide an alternative proof of theexpressive power of differential interaction nets by means of the solos calculus. The first half of thepaper will be devoted to the introduction of the required calculi (-calculus, solos calculus and solodiagrams) together with a simple translation of solo diagrams [LPV01] (the graphical syntax forthe solos calculus) into differential interaction nets extracted from the material presented in [EL08].In particular Section 3 is almost copy-pasted from [EL08]. We will finish the first part of the paperby giving a sufficient condition on solo diagrams for the translation into differential interaction netsto be a bisimulation.

The main technical contribution of the paper comes in the second half. Since the property thata name should not be unified with itself in the solos calculus is of course not preserved under thereduction of solos, we have to find a more clever property. By introducing a simple typing systemon solos and by deriving constraints on solos terms typed in this system, we introduce the acyclicsolos calculus. We prove this restriction to be well behaved with respect to the reduction of thesolos calculus. We show that the translation of a -term is always an acyclic solos term, showingthe expressiveness of the system. Finally we prove that the sufficient condition introduced at theend of the first part of the paper is fulfilled by solo diagrams corresponding to acyclic solos termsshowing that we obtain a bisimulation with respect to differential interaction nets.

Conventions. Since our goal is to focus on prefixing and sequentiality, we deal with calculiwithout replications nor recursive definitions, without match/mismatch and without sums.

We do not want to spend time to deal with arbitrary arities in the calculi we consider. This iswhy we only consider monadic -terms. There are three reasons for that: it makes the presentationsimpler, it does not lead to a loss of expressiveness, and finally the polyadic case has already beenconsidered in [EL08]. As a consequence (see the translation in Section 1.3), we are led to considera triadic solos calculus (all the names are of arity exactly 3) and triadic solo diagrams (all themultiedges are of arity exactly 3). The more general case of arbitrary arities could easily beobtained by introducing the appropriate sortings on the various calculi.

2

1 The -calculus and the solos calculus

In this section we recall the definition of the -calculus and of the solos calculus we are going touse. We also recall the translation from to solos given in [LV03].

1.1 The -calculus

The terms of the (monadic) (finitary) -calculus are given by:

P ::= 0 | u(x).P | ux.P | (P | P ) | x.P

where both u(x).P and x.P bind x.The structural congruence on -terms is the least congruence containing -equivalence and:

0 | P P

P | Q Q | P

(P | Q) | R P | (Q | R)

x.y.P y.x.P

x.0 0

(x.P ) | Q x.(P | Q) if x / fn(Q)

The reduction semantics of the -calculus is given by:

ux.P | u(y).Q P | Q [x/y]

P Q

P | R Q | R

P Qx.P x.Q

P P P Q Q QP Q

1.2 The solos calculus

Introduced in [LV03], the goal of the solos calculus is to prove the expressiveness of a calculuswithout prefix construction.

The terms of the (triadic) solos calculus are given by:

P ::= 0 | ux1x2x3 | ux1x2x3 | (P | P ) | (x)P

where (x)P binds x.The structural congruence is the least congruence containing -equivalence and:

0 | P P

P | Q Q | P

(P | Q) | R P | (Q | R)

(x) (y)P (y) (x)P

(x) 0 0

((x)P ) | Q (x) (P | Q) if x / fn(Q)

3

This equivalence allows us to present terms in the solos calculus in canonical forms: either the 0process or a bunch of scope constructions (x1) (x2) . . . followed by solos in parallel.

The reduction semantics of the solos calculus is given by (z stands for z1 . . . zn):

(z) (u x1x2x3 | u y1y2y3 | P ) P

where is a most general unifier of x1x2x3 and y1y2y3, suchthat exactly the names in z are modified (in particular, ineach equivalence class of names induced by unification, atmost one name is free).

P Q

P | R Q | R

P Q

(x)P (x)QP P P Q Q Q

P Q

An alternative (but equivalent) definition of this reduction semantics is given in [LV03] togetherwith additional explanations.

For example, in (x) (y) (z) (w) (u uxy | u zww | v zuy), the only possible reduction is betweenuuxy and u zww. It induces the identifications u = z, x = w and y = w, thus two equivalenceclasses {u, z} and {x, y,w}. In the first one, u is free and thus the only possibility is to map z tou. In the second one, all the elements are bound, we choose one of them: y for example (the otherchoices would lead to structurally congruent results). We consider the unifier containing z 7 u,x 7 y, w 7 y and which is the identity on the other names. We obtain the reduct (y) v uuy.

1.3 From -terms to solos

In [LV03], the authors give different translations of the fusion calculus [PV98] into solos. We aregoing to focus on one of them (the one which does not introduce matching). By pre-composingthis translation with the canonical embedding of the -calculus into the fusion calculus: u(x).P 7(x)ux.P , we obtain the translation of the -calculus into solos we present here.

A -term P is translated as [P ]:

Cv := (z) v zzv

[0]v := 0

[u(x).P ]v := (w) (y) (v uwy | Cy | (x) (v) (w xvv | [P ]v))

[ux.P ]v := (w) (y) (v uwy | Cy | (v) (w xvv | [P ]v))

[P | Q]v := [P ]v | [Q]v

[x.P ]v := (x) [P ]v

[P ] := (v) ([P ]v | Cv)

The -term x.(ux.0 | x(y).0) | u(z).zt.0 is translated as a solos term which is