S.GRAF C.LOISEAUX loiseaux@imag.fr. Keywords:abstractinterpretation,simulation,propertypreservation,model-checking. 1.Introduction



Similar documents
/,,7($78/267((7. 78/267(,'(1</(,6,b20,1$,688.6,$ 78/267(,'(1(6,77(/<

Cyclotomic Extensions

Integer Programming: Algorithms - 3

IP Subnetting Colin Weaver, ITdojo

3 1. Note that all cubes solve it; therefore, there are no more

How To Prove The Dirichlet Unit Theorem

Sample Configuration for H.323 Trunk between Avaya IP Office and Cisco Unified Communications Manager 7.0 Issue 1.0

26 Ideals and Quotient Rings

Factorization in Polynomial Rings

Application of Design of Experiments to an Automated Trading System

Lecture 13 Linear quadratic Lyapunov theory

Booth School of Business, University of Chicago Business 41202, Spring Quarter 2015, Mr. Ruey S. Tsay. Solutions to Midterm

Class Demonstration Problems on Inventory Errors:

Stat405. Simulation. Hadley Wickham. Thursday, September 20, 12

New Car $16,000 5 yr. payments Car note /month. New Car $30,000 5 yr. payments Car note $500./month Car insurance $250/month Gasoline $75/week

Topics in Chapter. Key features of bonds Bond valuation Measuring yield Assessing risk

To receive highlights of the information on this page delivered directly to your inbox, please register here.

Interest Rates and Bond Valuation

ยกระด บผลส มฤทธ ทางการเร ยน(O-NET) ป การศ กษา 2556 แผนงาน

Overview East Molloy Road East Syracuse, NY TEL: FAX:

SWAPTION PRICING OPENGAMMA QUANTITATIVE RESEARCH

Figure2:Themixtureoffactoranalysisgenerativemodel. j;j z

2 When is a 2-Digit Number the Sum of the Squares of its Digits?

9 Summary of California Law (10th), Partnership

CS103B Handout 17 Winter 2007 February 26, 2007 Languages and Regular Expressions

Credit Management. Why Credit Exists

FACTORING CERTAIN INFINITE ABELIAN GROUPS BY DISTORTED CYCLIC SUBSETS

Energy Storage and Distribution

Cisco Advanced Services for Network Security

ISDX Network Connectivity 26 October 2012

ON DEGREE OF APPROXIMATION ON A JORDAN CURVE TO A FUNCTION ANALYTIC INTERIOR TO THE CURVE BY FUNCTIONS NOT NECESSARILY ANALYTIC INTERIOR TO THE CURVE

IT:101 Cisco Networking Academy I Subnetting

4.6 Linear Programming duality

all degree Faculty of BMSS Summer semester all degree all degree all degree all degree all degree all degree Page 1 of 17

Beyond Market Research

Rigorous Software Development CSCI-GA

Credit & Risk Management. Lawrence Marsiello Vice Chairman and Chief Lending Officer

Lap Fillet Weld Calculations and FEA Techniques

E-procurement overview

IEOR 4404 Homework #2 Intro OR: Deterministic Models February 14, 2011 Prof. Jay Sethuraman Page 1 of 5. Homework #2

MTAT Cryptology II. Digital Signatures. Sven Laur University of Tartu

minimal polyonomial Example

End User Training and Documentation a capability of Solution Implementation. August 2011

The Role of Priorities in Assigning Indivisible Objects: A Characterization of Top Trading Cycles

Option Pricing. Chapter 4 Including dividends in the BS model. Stefan Ankirchner. University of Bonn. last update: 6th November 2013

A couple of things involving environments

Practical Guide to the Simplex Method of Linear Programming

*) Online Read The Conversion Code: Capture Internet Leads, Create Quality Appointments, Close More Sales download ebook free pdf format ID:gaucpe

E3: PROBABILITY AND STATISTICS lecture notes

Optimizing IT Deployment Issues

Rotation Rate of a Trajectory of an Algebraic Vector Field Around an Algebraic Curve

How To Test The Nms Adaptive Suite With An Ip Office On A Windows 2003 Server On A Nms Desktop On A Pnet 2.5 (Tapi) On A Blackberry 2.2 (Tapi) On An Ipo 2

Java Modules for Time Series Analysis

PivotStream Cloud Hosted PowerPivot for Excel 2010/2013 tool

Filings With the U.S. Securities and Exchange Commission Under the Securities Act of 1933

1.2 GRAPHS OF EQUATIONS. Copyright Cengage Learning. All rights reserved.

Analysis of Rhode Island Commercial Insurance Enrollment Trends by Line of Business and Carrier as of December 2010

What Benefits Can I Change Mid Year?

Deploying Intel Architecturebased Tablets with Windows* 8 at Intel Tiffany Pany, Intel IT April 2013

The default setting has been changed to be Direct Thermal on the Zebra 2746 and 2746e printers.

ALGEBRAIC EIGENVALUE PROBLEM

XML Export Interface. IPS Light. 2 April Contact

Testing Market Efficiency in a Fixed Odds Betting Market

SOME EXAMPLES OF INTEGRAL DEFINITE QUATERNARY QUADRATIC FORMS WITH PRIME DISCRIMINANT KI-ICHIRO HASHIMOTO

Mean squared error matrix comparison of least aquares and Stein-rule estimators for regression coefficients under non-normal disturbances

VI. Real Business Cycles Models

4.1 Modules, Homomorphisms, and Exact Sequences

How To Set up and Configure the WNA-100 Wireless Network Adapter

Corporate Payables via Virtual Credit Card

Application Notes for Multi-Tech FaxFinder IP with Avaya IP Office Issue 1.0

Duality of linear conic problems

System Optimizer Solution for resource planning, capacity expansion, and emissions compliance for portfolio optimization

Company Brief 4brands Reply GmbH & Co. KG

SETTING UP HRD ON THE HOST COMPUTER

Quotient Rings and Field Extensions

Genius in Salesforce.com Pre- Installation Setup

Retirement Plan Of CITGO Petroleum Corporation And Participating Subsidiary Companies. Summary Plan Description As in effect January 1, 2012

APPLICAZIONI PTO/POMPA PTO/PUMP ADAPTORS

ON CERTAIN DOUBLY INFINITE SYSTEMS OF CURVES ON A SURFACE

FOUNDATIONS OF ALGEBRAIC GEOMETRY CLASS 22

Transcription:

VericationofConcurrentSystems* PropertyPreservingAbstractionsforthe c1995kluweracademicpublishers,boston.manufacturedinthenetherlands. FormalMethodsinSystemDesign,6,1{36(1995) S.GRAF C.LOISEAUX loiseaux@imag.fr VERIMAG*,RueLavoisier,38330Monbonnot S.BENSALEM A.BOUAJJANI J.SIFAKIS bensalem@imag.fr bouajjan@imag.fr sifakis@imag.fr graf@imag.fr ReceivedOctober1,1992;RevisedFebruary1,1994 Editor:DavidProbst oftwosystems.weproposeandstudyanotionofpreservationofpropertiesexpressedbyformulas S'.Wegiveresultsonthepreservationofpropertiesexpressedinsublanguagesofthebranching toverifyapropertyforasystembyverifyingthesamepropertyonasimplersystemwhichisan Abstract.Westudypropertypreservingtransformationsforreactivesystems.Themainideais abstractionofit.weshowalsounderwhichconditionsabstractionofconcurrentsystemscanbe computedfromtheabstractionoftheircomponents.thisallowsacompositionalapplicationof theproposedvericationmethod. Thisisarevisedversionofthepapers[2]and[16];theresultsarefullydevelopedin[28]. ofalogic,byafunctionmappingsetsofstatesofasystemsintosetsofstatesofasystem time-calculuswhentwosystemssands'arerelatedviah;i-simulations.theycanbeused theuseofsimulationsparameterizedbygaloisconnections(;),relatingthelatticesofproperties Keywords:abstractinterpretation,simulation,propertypreservation,model-checking. 1.Introduction tobeveried,nda(simpler)abstractprogramsuchthatthesatisfactiononthe consistsinusingpropertypreservingabstractions:givenaprogramandaproperty modelsthathavetobeconstructedfortheirapplication.manytechniqueshave itationofautomaticvericationtechniquesistheirapplicabilityonlytorelatively smallnitestateprogramsbecauseoftheexponentialblow-upofthesizeofthe beendevelopedinordertopushfurtherthelimitsofmodel-checking.oneofthem Thegrowingcomplexityofdistributedandreactivesystemsrequiresrigorousde- *ThisworkwaspartiallysupportedbyESPRITBasicResearchAction\REACT" velopmentmethodologiesandautomaticvericationtechniques.awell-knownlim- J.FourierandVerilogSAassociatedwithIMAG *VerimagisajointlaboratoryofCNRS,InstitutNationalPolytechniquedeGrenoble,Universite

2abstractprogramimpliesthesatisfactionontheinitialprogram,calledconcrete actlythisproblem.programsarerepresentedbyfunctionsfonsomelatticeof oftheconsideredproperties. byavailabletools,andthatstillcontainsenoughrelevantdetailsforthesatisfaction constructanabstractprogramthatisboth,simpleenoughinordertobeveried programinthiscontext.animportantpointis,givenaconcreteprogram,howto tiongontheabstractlatticeisanabstractionoffiffgholds.this formingagaloisconnection[35]fromtheconcretetotheabstractlattice,afunc- properties.givensomeabstractlatticeofpropertiesandapairoffunctions(;), [40],[41],theideaofabstractinterpretationhasbeenappliedtoprogramsrepresentedbytransitionsystems,wherethelatticeofpropertiesisthepowersetofstates. forthevericationofinvariancepropertiesofsequentialprograms.however,in ofcorrespondingxpointsoff.untilrecently,thisapproachhasonlybeenapplied guaranteesthatgreatestandleastxpointsofgrepresentupperapproximations Theframeworkofabstractinterpretation(seeforexample[7],[8])addressesex- There,resultsshowingpreservationoffragmentsofCTL[9]fromtheabstractto theconcretesystemhavebeengiven. sion(respectivelyequality)ofobservablecomputationsequences(seeforexample in[25],[1],[30]).however,thisnotionofabstractiondoesnotdirectlyinduceaway tionsofabstractionsaregenerallydenedintermsofvariantsofsimulation[31] andbisimulation[32];theproblemoftheconstructionofabstractprogramshas onlybeenaddressedfornotionsofabstractionsdenedbyequivalences. ordersandequivalenceshasalsobeenwidelystudied.inthisframework,theno- Inthelinearsemanticsframework,theintuitivenotionofabstractionisinclu- Intheframeworkofprocessalgebras,theproblemofpropertypreservingpre- coincidesexactlywiththenotionofabstractiondenedbysimulationinthesenseof criterion. Milner[31],parameterizedbytherelationcorrespondingtotheGaloisconnection abstractionontransitionsystemsasasimulationparameterizedbygaloisconnections(;).weshowthatthenotionofabstractioninducedbyh;i-simulation ofcomputinganabstractprogramforagivenconcreteprogramandobservability Here,wetakeupagaintheapproachfollowedin[40],[41].Wedeneanotionof fifforanystateofs1whichsatisesf,allthestatesofs2initsimagealsosatisfy f.iftheconversealsoholds,thenwesaythatstronglypreservesf.apreservation systems1tothepowersetofthestatesofatransitionsystems2preservesaproperty resultofparticularpracticalinterest,saysthatiftwosystemsarerelatedviah;isimulation,thenallformulasofthe-calculususingnonegationandonlyuniversal thebranchingtime-calculusdenedin[24]forthefollowingnotionofproperty preservation:anarbitraryfunctionfromthepowersetofthestatesofatransition Then,wegivepreservationresultsforfragmentsofafutureandpastversionof

quanticationovercomputationsequences(called2l)arepreservedbyefrom theabstracttotheconcretesystem(whereeisthedualof). structurehomomorphismfromtheconcretetoabstractsystem. studiedintheparticularcasewherethepropertypreservingfunctiondenesa Ourpreservationresultstogetherwiththefactthat,givensomeconcretesystem Thesepreservationresultsgeneralizeresultsgivenin[10]wherethisproblemis 3 composition,whichisimportantfortheapplicationofthismethodinpractice. andsomeconnection(;),anabstractsystemcanbecomputed,allowtheuse gramofacomposedsystembycompositionofabstractionsofitscomponents.it powersetsofconcreteandabstractstates,computetheassociatedabstractsystem Fromapracticalpointofview,therearetworeasonsforbuildinganabstractpro- SAandverifyfonSA.IffholdsonSA,italsoholdsonS. ofthefollowingvericationmethod.inordertoverifyaproperty expressed asaformulafof2l onasystems,provideaconnection(;)betweenthe iseasiertodeneconnections(;)separatelyforeachcomponentthanforthe compoundsystem;proceedingthisway,allowsalsotoavoidbuildingarepresentationoftheglobaltransitionsystemassociatedwiththecomposedsystem.aswell betweencomponents),wegivecompositionalityresults,thatmeansrules,allowing forsynchronousasforasynchronousparallelcomposition(allowingsharedvariables Finally,wegivearesultconcerningcompositionalityofsimulationoverparallel todeduceh;i-simulationforacompoundsystemfromhi;ii-simulationsforits thedenitionofgaloisconnectionsandsomeinterestingpropertiesofthem.in components,whereh;iisexpressedintermsofhi;ii. Section3,thedenitionofh;i-simulationisgiven.Weshowthatthisnotion coincideswiththeusualnotionofsimulation.insection4,wedeneanotionof \abstractprogram"obtainedfromagivenfunctionoritsassociatedrelation. Section6givesresultsconcerningthepreservationoffragmentsofthe-calculus toprovethatafunctionpreservesthevalidityofformulasofagivenlanguage. Section5presentsthenotionofpropertypreservationandgeneralresultsallowing whentransitionsystemsarerelatedviah;i-simulation.section7,givesresults Thepaperisorganizedasfollows.InSection2,wegivesomenotationsandrecall Finally,AnnexAcontainssometechnicalproofs. concerningthecompositionalityofsimulationwithrespecttoparallelcomposition. 2.Preliminarydenitions InSection3,westudytherelationshipbetweenthenotionsofabstractioninthe gramsaremodeledastransitionsystems,thatmeansasbinaryrelationsontheset ofstates.intheframeworkofabstractinterpretation,programsarerepresentedby denethebasicnotions,necessaryforthiscomparison.inprocessalgebraspro- frameworksofprocessalgebrasandofabstractinterpretation.inthissection,we predicatetransformers,i.e.,functionstransformingsetsofstatesintosetsofstates.

4WithanytransitionrelationRcanbeassociateddierentpredicatetransformers, concerningthem,whichareusedintheproofslateron. 2.1.Transitionsystemsandpredicatetransformers WerecallherethedenitionofGaloisconnectionandsomewell-knownproperties theforwardandbackwardimagefunctions,whichwedenoteherebypre[r],respectivelypost[r].intheabstractinterpretationframework,thenotionofabstraction Denition1(Transitionsystems) isbasedontheexistenceofagaloisconnectionbetweenthelatticesofproperties. AtransitionsystemisapairS=(Q;R),whereQisasetofstatesandRisa transitionrelationonq(rqq). Notation1Weadoptthefollowingconventionsandnotations: WeidentifyaunarypredicateonQwithitscharacteristicsetsincethelattice WedenotebyIdQtheidentityfunctionon2Q. GiventworelationsRQQ0andSQ0Q00andtwofunctionsf:Q!Q0 stateq2q,thenotationsp(q)=true,p(q)andq2pareequivalent. ofunarypredicatesisisomorphicto2q.thus,foraunarypredicatepanda asasetofstates(oracorrespondingunarypredicate).therefore,inthesequel \propertylattice"isalwaysthesameas\powersetonthesetofstates". Denition2(Thepredicatetransformerspreandpost) Inthesequel,weconsideralwayspropertiestobestateproperties,i.e.,interpreted gfisappliedtosomeargumentq2q. andg:q0!q00,thendenotethecompositionoftherelationsrandsbyrs GivenarelationQ1Q2,wedenepre[]:2Q2!2Q1andpost[]:2Q1!2Q2 andthecompositionofthefunctionsfandgbygf,respectivelyg(f(q))if statesofq20viatherelationandforq10q1,post[](q10)representsthesetof \successors"ofthestatesofq10via.noticethatwehavepost[]=pre[?1]. by,pre[]def Thatmeans,forQ20Q2,pre[](Q20)representsthesetof\predecessors"ofthe post[]def =X:fq12Q1:9q22X:q1q2g formerspreandpostwhichcanforexamplebefoundin[41]. Thefollowingpropositionsgivesomeusefulresultsconcerningthepredicatetrans- =X:fq22Q2:9q12X:q1q2g Proposition1ForanyrelationfromasetQ1toasetQ2(Q1Q2),we have:

Notation2(Dualofafunction) 2.ForanyX1,X2subsetsofQ2,pre[](X1[X2)=pre[](X1)[pre[](X2), 1.pre[](;)=;, 5 Wedenotebyethedualofafunction:2Q1!2Q2thatis Proposition2LetbeQ1Q2andQ2Q3.Then, edef pre[]=pre[]pre[], post[]=post[]post[], =X:(X). 2.2.Galoisconnections WegivehereafterthedenitionofGaloisconnectionsandsomeusefulwell-known resultsaboutthem.moreinformationcan,e.g.,befoundin[35],[39]. gpost[]=gpost[]gpost[]. fpre[]=fpre[]fpre[], LetQ1andQ2betwosetsofstates.Aconnectionfrom2Q1to2Q2isapairof Denition3(Connections) IdQ1andIdQ2. Proposition3Foranyconnection(;)from2Q1to2Q2,wehave, monotonicfunctions(;),where:2q1!2q2and:2q2!2q1,suchthat =,and=, (;)=;, distributesover[anddistributesover\, (e;e)isaconnectionfrom2q2to2q1. Proposition5Foranyconnection(;)from2Q1to2Q2,wehave, Proposition4LetF:2Q1!2Q1andG:2Q2!2Q2betwofunctionsand(;) aconnectionfrom2q2to2q1.then, 8QQ1;Q0Q2:(Q)Q0iQ(Q0). =Y:SfX22Q1:(X)Yg, FGifandonlyifFG

betweentheconnectionsfrom2q1to2q2andthebinaryrelationsfromq1toq2. 6=X:TfY22Q2:X(Y)g. Proposition6(Connectionsgeneratedbyabinaryrelationonstates) characterizationsallowtodeducethefollowingtwopropositionsshowingthelinks IfQ1Q2,thenthepair(post[];fpre[])isaconnectionfrom2Q1to2Q2and (pre[];gpost[])isaconnectionfrom2q2to2q1. Proposition7(Relationsinducedbyconnections) Thatmeansthatanddetermineeachotherinauniquemanner.These If(;)isaconnectionfrom2Q1to2Q2,thenthereexistsauniquerelation by(q1;q2)2ifandonlyifq22(q1).since(;)=;anddistributesover[ (Proposition3),wehave=post[]. Q1Q2suchthat=post[]and=fpre[]. Proof:Let(;)beaconnectionfrom2Q1to2Q2.Considertherelationdened tionfrom2q2to2q1,thenwehave, Proposition8If(;)isaconnectionfrom2Q1to2Q2and(0;0)isaconnec- andasdistributesover[,wecanwrite=y:fq2q1:(fqg)yg.now, since=post[],itiseasytodeducethat=fpre[]. Furthermore,bytheProposition5,wehave=Y:SfX22Q1:(X)Yg, Proof:ConsidertherelationQ1Q2suchthat=post[]and=fpre[], whichexistsbyproposition7. totalonq1andidq2post[]pre[]foranyq1q2thatistotalonq2. Now,itiseasytoseethatIdQ1pre[]post[]foranyQ1Q2thatis 1.IdIm(e)eandIdIm()e, pre[0]pre[]pre[0]=pre[0]forsomeappropriaterelations;0.byproposition2,thisisequivalenttopre[00]=pre[0],thatis0=00. Symmetrically,0=00isequivalenttopost[0]=post[00],thatisto ByProposition7,theequatione0ee0=e0isequivalentto 2.e0ee0=e0ifandonlyif00=0. Inthissection,wedeneanotionofsimulationbasedonGaloisconnections(;), 3.Simulations calledh;i-simulation.itsdenitionisinspiredbythenotionofabstractinterpretationinthesenseofcousot[7],[8].there,aprogramisrepresentedbyafunction Fmappingpropertiesintoproperties.AfunctionG,mappingabstractproperties post[0]=post[0]post[]post[0],i.e.,0=00.

intoabstractproperties,isanabstractionoffifthereexistsaconnection(;) fromthetheconcretetoabstractlatticeofproperties,suchthatfg. blechoiceforthefunctionfistakingoneofthepredicatetransformersassociated stractionofs"and\ssimulatessa"areequivalent.weshowthatthenotionof withthetransitionrelationr.weconsiderthattheexpressions\saisanab- Inourframework,whereaprogramisatransitionsystemS1=(Q1;R1),apossi- 7 abstractioninducedbythechoicef=pre[r1]coincideswiththenotionofabstractioninducedbysimulationinthesenseofmilner[31]whichisusedinthe S2=(Q2;R2),i.e.,aconnectionfrom2Q1to2Q2. (;)relatingthepropertylatticesoftwotransitionsystemss1=(q1;r1)and frameworkofprocessalgebras. 3.1.Simulationsinducedbyconnections Denition4(vh;iand'h;i) LetS1=(Q1;R1)andS2=(Q2;R2)betwotransitionsystemsand(;)bea First,wedenesimulation(andbisimulations)parameterizedbyaconnection connectionfrom2q1to2q2.dene, fromproposition4. IfS1vh;iS2,wesaythatS1h;i-simulatesS2orS2isanh;i-abstractionof S1.Ausefuldualconditionforthedenitionofh;i-simulationcanbededuced S1'h;iS2ifandonlyifS1vh;iS2andS2vhe;eiS1. S1vh;iS2ifandonlyifpre[R1]pre[R2], Q1andQ2.InPropositions9and10weshowthatthesetwonotionsofsimulation senseofmilnerwhicharebasedonabinaryrelationbetweenthesetsofstates 3.2.Relatingh;i-simulationandbehaviouralsimulation coincide. Denition5(vand') Werecallrstthedenitionsofbehaviouralsimulationandbisimulationinthe LetS1=(Q1;R1)andS2=(Q2;R2)betwotransitionsystemsandbearelation fromq1toq2(q1q2).dene, S1'S2ifandonlyifS1vS2andS2v?1S1. S1vS2ifandonlyifR1?1R2?1,

suchthats1vs2(respectivelys1's2).weshownowthath;i-simulation and-simulationcoincide. 8IfS1vS2,wesaythatS1-simulatesS2orS2isa-abstractionofS1. Q1Q2,thereexistsaconnection(;)from2Q1to2Q2suchthat Proposition9(Fromvh;itov) S1vS2ifandonlyifS1vh;iS2. LetS1=(Q1;R1)andS2=(Q2;R2)betwotransitionsystems.Foranyrelation S1simulates(respectively,bisimulates)thesystemS2ifthereexistsarelation Proof:Weshowthattheintendedconnectionis(post[];fpre[])(byProposition6, thispairisindeedaconnection).supposethats1vhpost[];fpre[]is2,i.e., Then,aspost[]ismonotonicandIdQ1fpre[]post[],weobtain, post[]pre[r1]fpre[]pre[r2]. Proposition10(Fromvtovh;i) LetS1=(Q1;R1)andS2=(Q2;R2)betwotransitionsystems.Foranyconnection Itcanbeshowninasimilarwaythattheconversealsoholds.Thisproves, post[]pre[r1]fpre[]post[]pre[r2]post[]whichimplies (;)from2q1to2q2thereexistsarelationq1q2suchthat post[]pre[r1]pre[r2]post[]whichisequivalenttor1?1r2?1. S1vh;iS2ifandonlyifS1vS2. S1'hpost[];fpre[]iS2ifandonlyifS1'S2. abstractioninthecasewhereprogrammodelsaretransitionsystemsisthesame. Therefore,wedonotdistinguishinthesequelbetweensimulationsparameterized tationandthatchosenintheframeworkofprocessalgebra.infact,thenotionof byrelationsandthoseparameterizedbyconnections;inanycontextweusethe notionwhichallowstopresenttheresultsinthesimplestway. Proof:DirectfromPropositions7and9. 4.Computingprogramabstractions Thisresultclariestherelationshipbetweentheapproachofabstractinterpre- Intheframeworkofprocessalgebraandofprogramrenement,thenotionofsimulationisingeneralusedinordertodecidefortwogivenprogramsifoneofthem simulatestheother.butouraimis,givenaprogrampandarelationrelating concreteandabstractstates,toconstructanabstractprogrampasuchthatp -simulatespa.obviously,therearemanyprogramswhichare-abstractionsof

i.e.whichisascloseaspossibletotheconcreteprogram. anabstractprogramsatisfying foragiven asmanypropertiesaspossible, theabstractprogrammustalsoberepresentablebysometransitionrelationofthe theabstractsetofstatesisatrivial-abstractionofanyp.weareinterestedin P.InparticulartheprogramChaosdenedbytheuniversaltransitionrelationon Inourframework,wherePisrepresentedbyatransitionsystemS=(Q;R) 9 ofs.insection4.1,wedenerstthecriteriumoffaithfulnesswhichissatised byalltransitionsystemsonqawhicharebisimilartoanysmaller(inthesense ofinclusion)-abstractionsofs.usingtheresultsofsection5,wewillseethat simulation,doesnotnecessarilycorrespondtoasolution,thatmeansafunctionof theformpre[ra]forsometransitionrelationra. formsa=(qa;ra),whereqaisthesetofabstractstates.inthiscasethe faithfulabstractionsarethesetofabstractprogramswhichsatisfyallproperties obviousminimalfunctionpost[]pre[r]fpre[] obtainedfromthedenitionof whicharepossiblysatisedbyany-abstractionofsandwhicharepreservedfrom SAtoS. Itiseasytoseethatingeneral,theremayexistseveral\minimal"-abstractions thecasethatisatotalfunction,pre[]=fpre[]holds,whichtriviallyimpliesthat vwhichwedenoteby.undersomeconditionscoincideswiththenotionof S;thiscasehasbeenwidelystudiedintheliterature(seeforexamplein[25],[10]). Sistheleastabstraction.Then,denesastructurehomomorphismfromSto?1Risafaithfulabstractionifistotalandmoreover=?1holds.In forwardandbackwardsimulationforwhichweobtainstrongerpreservationresults WewillseethattheabstractprogramdenedbyS=(QA;R)withR= andillustratethisonasmallexample. thanforv. abstractionrelationsarerepresentedbypredicatesoversetsofprogramvariables Sisinducedinanobviousmannerbyaslightlystrongernotionofsimulationthan 4.1.Faithfulabstractions Denition6(Faithfulabstractions) GivenS=(Q;R)andQQA,wesaythatSA=(QA;RA)isafaithful InSection4.2,weshowhowScanbecomputediftransitionrelationsaswellas abstractionofsviaifsvsaand8s0=(qa;r0):svs0andr0ra implies90qaqa:sa'0s0. Proposition11LetS=(Q;R)beatransitionsystemandQQA. R=?1R(orequivalently,pre[R]=post[]pre[R]pre[]). GivenS=(Q;R)andQQA,totalonQ,wedeneS=(QA;R)where Notation3(ThesystemS) IfistotalonQ,thenSvS.

Proof:Therstandthethirditemsfollowdirectlyfromthefactthatfpre[]pre[] 10Iffurthermore=?1,thenSisafaithfulabstractionofSvia. ifistotalonq(respectivelyfpre[]=pre[]ifisafunction).forthesecond item,weshowthatforanytransitionsystemsa=(qa;ra)suchthatsvsaand RAR,wehaveSA'?1S,theproofofwhichisgivenintheAppendixA.1. Ifisa(total)function,thenSistheleast-abstractionofS.?1.Thereexistexamplesofinterestingabstractionrelationssuchthatisnot fromthepartitiononqinducedby?1intothepartitionofqainducedby successorby,havethesamesuccessorsby.thismeansthatdenesafunction function.if=?1doesnothold,thensisnotnecessarilyfaithful,andin[12] isgivenawaytocomputefaithfulabstractions. vh;i)whichcoincideswiththenotionofforwardandbackwardsimulationused, e.g.in[21],[22]ifistotal. Sisinducedbyaslightlystrongernotionofsimulationthanv(respectively Noticethat=?1ifandonlyifanytwostatesofQhavingacommon Denition7(andh;i) LetS=(Q;R)andSA=(QA;RA)betransitionsystems,andQQAtotal Lemma1(Characterizationof) onqand(;)atotalconnectionfrom2qto2qa.then, LetS=(Q;R)andSA=(QA;RA)betransitionsystems,andQQAtotal onq;denotes?1=(q;r?1)andanalogouslyforsa.then, SSAifandonlyif?1RRA Sh;iSAifandonlyifpre[R]epre[RA] Now,weconsidertheparticularcasewheretransitionrelationsandabstraction 4.2.Symboliccomputationofprogramabstractions relationsarerepresentedbypredicatesoverprogramvariables.thesetsofstates QaretheCartesianproductofthedomainsofatupleofprogramvariables.For SSAifandonlyifSvSAandS?1vS?1 example,ifx=(x;y),thenwehave,q=dom(x)=dom(x)dom(y). theformr(x;x0)wherex0=(x0;y0)isa\copy"ofx,i.e.,dom(x)=dom(x0). XencodesthesourcestateandX0thetargetstateofanytransitioninR.For example.,ifdom(x)=nanddom(y)=bool,thenr=y^(x0=x+1) Then,binaryrelationsonDom(X)canberepresentedbybinarypredicatesof

representsthetransitionrelationrelatingany(n;true)2nboolwith(n+1;b0) Thisapproachisused,e.g.,in[27],[37].InthesamewayarelationfromDom(X) whereb0maytakeanybooleanvalueasy0isnotconstraintintheexpressionr. todom(xa)isrepresentedbyabinarypredicateoftheform(x;xa). connectives.forexample,thefactthatarelationr1isincludedinr2isexpressed Inthissetting,operationsonsets(respectivelyrelations)areexpressedbylogical 11 Section7. areusedaslabels(names)forsynchronizationpurposesinparallelcompositionin ofbinarypredicatesonthesametupleofvariables,s=fri(x;x0)gi2iwherei2i onthesamesetofvariables. byr1)r2andr1^r2representstheintersectionofr1andr2iftheyaredened ables,theabstractionsofsiscomputedas Then,givenanabstractionrelation(X;Y),whereYisatupleofabstractvari- Weconsiderthataprogramisafamilyoftransitionrelationsrepresentedbysets containingexpressionsinwhich,atleastinthecasewheredom(x)anddom(y) arenite,alloccurrencesofvariablesxandx0canbeeliminated. Example:areader/writerproblem Wedescribeasimplereaders/writerssystembythefollowing\program"RW;in S=f9X9X0:(X;Y)^(X0;Y0)^Ri(X;X0)gi2I factrwdenesafamilyoflabeledtransitionrelationswhereforreadabilityreasons anexplicitlabel((b-read),(e-read),...)ofeachactionisputbetweenparenthesesin frontoftheexpressiondeningthetransitionrelation. RW=f (b-read)(wr>0)^(aw=0)^(wr0=wr?1)^(ww0=ww)^ (e-read)(ar>0) (b-write)(ww>0)^(aw=0)^ (Ar=0) ^(Wr0=Wr+1)^(Ww0=Ww)^ ^(Wr0=Wr)^(Ww0=Ww?1)^ (Ar0=Ar+1)^(Aw0=Aw); (Ar0=Ar?1)^(Aw0=Aw); wherewrandwwarepositiveintegervariablesrepresentingrespectivelythenumbersofwaitingreadersandwaitingwriters,arandawrespectivelythenumbers ofactivereadersandactivewriters.thetransitionrelationassociatedwithrw (e-write)(aw>0) (n-wait) g ^(Wr0=Wr)^(Ww0=Ww+1)^ ((Wr0=Wr+1)_(Ww0=Ww+1))^ (Ar0=Ar)^(Aw0=Aw+1); hasaninnitenumberofstatesaswrandwwcanalwaysbeincreasedbyaction (n-wait). (Ar0=Ar)^(Aw0=Aw?1), (Ar0=Ar)^(Aw0=Aw)

relevantinformationis,whetherthenumberofactivereadersandwritersispositive 12 Wewanttoprovemutualexclusionbetweenreadersandwriters.Then,theonly ornot.therefore,wedeneanabstractionrelationmappingtheprogramvariablesontwobooleanvariablesb1andb2meaningrespectively\thereisnoactive reader"and\thereisnoactivewriter",by thevetransitionrelationsriofrwwehavetocomputetheabstracttransition Asisatotalfunction,RWisafaithfulabstractionofRWvia.Foreachoneof pression: ForthetransitionrelationR1(labeledby(b-read))oneobtainsthefollowingex- (Ri)=9X9X0:(X;Y)^(X0;Y0)^Ri(X;X0) ((Wr;Ww;Ar;Aw);(b1;b2)):=(b1(Ar=0))^(b2(Aw=0)). relation (R1)=9(Ar;Aw;Wr;Ww)9(Ar0;Aw0;Wr0;Ww0): transitionrelations: BydoingasimilarcomputationforallRiweobtainthefollowingfamilyofabstract RW=f(b-read)b2 (b1(ar=0))^(b2(aw=0))^(b01(ar0=0))^(b02(aw0=0))^ (Wr>0)^(Aw=0)^(Wr0=Wr?1)^ (Ww0=Ww)^(Ar0=Ar+1)^(Aw0=Aw) =b2^:b01^b02 (e-read):b1^(b02b2), (e-write):b2^(b01b1), (b-write)b1^b2^b01^:b02, (n-wait) ^:b01^b02, Nowwehavedenedanotionofabstractionandawaytocomputeabstractprograms.Animportantpointistoknowforwhichpropertieswecandeducefrom TheniteglobaltransitionrelationrepresentedbyRWisgivengraphicallyin Figure4.2. 5.Generalresultsonpropertypreservation (b01b1)^(b02b2)g thesatisfactionontheabstractsystemitssatisfactionontheconcretesystem.in allstatesofq2initsimagebysatisfypropertyf.wehavestrongpreservation iftheinverseholdsalso;thismeansintuitivelythatwheneverastateofq1does relatedviasomemonotonicfunction:2q1!2q2,thenthesatisfactionofsome statepropertyfispreservedfroms1tos2viaifforanystateofq1satisfyingf ordertoanswerthisquestion,weconsiderrstthegeneralproblemofproperty notsatisfyf,thenthereexistsastateinitsimagebywhichdoesnotsatisfyf. preservationbetweentwosystems.ifthepropertylatticesofthetwosystemsare

(b1;b2) 13 (b1;b2) e-write e-readb-write (b1;b2) b-reade-write e-read Figure1.Readers/Writersabstraction (b1;b2) b-read e-read expressedbyformulasofalogicallanguagef(p)wherep=fp1;p2;:::gisa setofpropositionalvariables.foragivensystems=(q;r)andaninterpretationfunctioni:p!2q,thesemanticsoff(p)isgivenbymeansofafunction that(;)isaconnection,becauseinsection6weapplythisnotionofpreservation Wegiveusefulcharacterizationsofthesedenitionsifthereexistsafunctionsuch jjs;i:f(p)!2q,associatingwitheachformulaitscharacteristicset,i.e.,theset ofstatessatisfyingit.thisfunctionissuchthat8p2p:jpjs;i=i(p). strongpreservationfrompreservationinbothdirections. tosystemsrelatedviah;i-simulation.wegivealsoatheoremallowingtodeduce omittedwhenevertheirvaluescanbedeterminedbythecontext. Tosimplifynotations,eitheroneorbothofthesubscriptsSandIinjfjS;Iwillbe Letusrstintroducesomenotations.Wesupposethatprogrampropertiesare anyq2, thatpreserves(respectivelystronglypreserves)fforionifandonlyiffor Letf2F(P)beaformula,S1=(Q1;R1)andS2=(Q2;R2)betwotransition If=Q1,weomittomentionthatthepreservationison. Denition8(Preservation) systems,q1,i:p!2q1aninterpretationfunctionand:2q1!2q2.wesay ofs1andpropertiesofs2.preservationmeansthatthefunctioniscompatible withthesatisfactionrelation.inthesequel,wherethefunctionunderconsiderationisalwaysmonotonic,andevensuchthatthereexistsafunction,such Inthisdenition,thefunctionestablishesacorrespondencebetweenproperties q2jfjs1;iimplies(respectivelyifandonlyif)(fqg)jfjs2;i. that(;)isaconnection,weusethefollowingcharacterizationsofthenotionof preservationinordertoestablishpreservationresults.

Letf2F(P)beaformula,S1=(Q1;R1)andS2=(Q2;R2)betwotransition 14 Lemma2(Characterizationofpreservation) systems,i:p!2q1beaninterpretationfunctionand:2q1!2q2. 1.ifismonotonicthen 2.ifthereexistssuchthat(;)isaGaloisconnection,then (A)preservesfforIifandonlyif (jfjs1;i)jfjs2;iimpliespreservesffori andifdistributesover[,theconversealsoholds. Theproofof(2A)isdirectfrom(1)andthelastitemofProposition3.(2B)can (jfjs1;i)=(sq2jfjs1;ifqg)=sq2jfjs1;i(fqg)whichestablishestheresult. Proof:Therstdirectionof(1)isimmediate:fromq2jfjS1;I,weobtainby monotonicityof,(fqg)(jfjs1;i)jfjs2;i.ifdistributesover[,then (B)stronglypreservesfforIifandonlyif jfjs1;i=(jfjs2;i) jfjs1;i(jfjs2;i) thatthereexistsfunctions,0suchthat(;)and(0;0)areconnectionsdoes tos2.noticethatthistheoremusesonlythemonotonicityofand0;thefact bededucedfromthefactthat((fqg)jfjs2;i))q2jfjs1;iisequivalentto Sf(q)jfjS2;IgfqgjfjS1;Iand notallowtoweakentheconditionsrequiredhere.therefore,weuseexactlythis theoreminordertoobtainthestrongpreservationresultsinthefollowingsection. ThefollowingtheoremgivesconditionsunderwhichpreservationbyfromS1to Sf(q)jfjS2;Igfqg=(jfjS2;I)byProposition5. Theorem1(Preservationandstrongpreservation) LetS1=(Q1;R1)andS2=(Q2;R2)betwotransitionsystems.Foranyset S2andpreservationby0fromS2toS1impliesstrongpreservationbyfromS1 that00=0andid0,ifpreservesffori:p!im(0)and0 Q1andforanymonotonicfunctions:2Q1!2Q2and0:2Q2!2Q1such (fqg)jfjs2;i.wehave, preservesfforithenstronglypreservesfforion. Proof:Inordertoshowstrongpreservationbysupposethat,forq2, I=0I0.Thus0I=00I0=0I0=Iwhichimpliesq2jfjS1;I. SinceI:P!Im(0),thereexistsaninterpretationfunctionI0:P!2Q2suchthat 0(fqg)0(jfjS2;I)(monotonicityof0), q20(jfjs2;i)(id0), q2jfjs1;0i(0preservesfforiandlemma2).

augmentedbypasttimemodalities,whichwedenotelp. simulationasdenedinsection3.theuniverseofpropertiesthatweconsideris thesetofpropertiesexpressibleinthepropositionalbranching-time-calculus[24] 6.Preservationofthe-calculus Nowwecantackletheproblemofpreservationbetweensystemsrelatedbyh;i- 15 suchasthebranching-timetemporallogicsctl[9]andctl[14]andalsothe forthefragmentsaugmentedbythecorrespondingpasttimemodalityholdalso. linear-timetemporallogicsasptl[36]andetl[42]. pstandsforlogicscontainingpasttimeoperators).weshowfortwosystemss1and S2that,ifS1vh;iS2,thenpreserves3LfromS1toS2andepreserves2L froms2tos1.ifmoreovers1h;is2holds,thenstrongerpreservationresults Thislogicsubsumesinexpressivenessthecommonlyusedspecicationlogics, i.e.,existenceofsimulationsinbothdirections. tionedabovepreservel(p) Weobtainstrongpreservationofthesefragmentsincaseofsimulationequivalence, Inthecasewherethetwosystemsareh;i-bisimilar,thetwofunctionsmen- Wedenefragmentsofthe-calculuscalledL,2L,2Lp,3L,and3Lp(where tion,wereformulatethevericationmethodsketchedintheintroductionandapply ittothesmallexampleintroducedinsection4.2. 6.1.Thepropositional-calculusanditsfragments andinthesecondsubsectionwegivethepreservationresults.inthethirdsubsec- Intherstsubsection,werecallthedenitionofthe-calculusanditsfragments and,undersomeconditions,theystronglypreserveit. Werecallthesyntaxandthesemanticsofthefutureandpastpropositional-calculus Asusually,thenotionoffreeoccurrencesofvariablesinaformulaisdenedasin formulasoflpisdenedbythefollowinggrammar: Lp.LetPbeasetofatomicpropositionsandXasetofvariables.Thesetofthe therst-orderpredicatecalculusbyconsideringtheoperatorasaquantier.a wherefissyntacticallymonotoniconx,i.e.,anyoccurrenceofxinfis f::=>jp2pjx2xj3fj3pfjf_fj:fjx:f andaninterpretationfunctionfortheatomicpropositionsi:p!2q.aformulaf inwhichthepastoperator3pisnotallowed. formulaisclosediftherearenovariablesoccurringfreeinit.listhefragment ThesemanticsoftheformulasisdenedforagiventransitionsystemS=(Q;R) underanevennumberofnegations. aclosedformulaisinterpretedasasetofstates.theinterpretationfunctionis withnfreevariablesisinterpretedasafunctionjfjs;i:(2q)n!2q.inparticular,

16 inductivelydenedasfollows,foravaluationv=(v1;:::;vn)2(2q)nofthevariablesoccurringfreeinit. j>js;i =Q, jpjs;i =I(P), jxjjs;i(v)=vj, jf1_f2js;i(v)=jf1js;i(v)[jf2js;i(v), j:fjs;i(v)=q?jfjs;i(v), j3fjs;i(v)=pre[r](jfjs;i(v)), j3pfjs;i(v)=post[r](jfjs;i(v)), jx:fjs;i(v)=tfq0q:jfjs;i[q0=x](v)q0g: WeextendLpbyaddingasusuallytheformulas?,f^g,f)g,X:f(X),2fand 2pfwhicharerespectivelyabbreviationsfor:>,:(:f_:g),:f_g,:X::f(:X), :3:fand:3p:f. Aformulaofthisextendedlanguageisinpositivenormalformifandonlyifall thenegationsoccurringinitareappliedtoatomicpropositions.itcanbeshown thatanyformulaoflphasanequivalentformulainpositivenormalform. WedenefragmentsofLpcalled2L,2Lp,3Land3Lp.Theirsetsofformulas aregivenrespectivelybythetwofollowinggrammarswherethepasttimemodalities 2pand3parenotallowedinthefuturefragments2L,respectively3L. g::=>j?jpj:pjxj2gj2pgjg_gjg^gjx:gjx:g h::=>j?jpj:pjxj3hj3phjh_hjh^hjx:hjx:h Noticethatpropertiesexpressedbyformulasof2L(p) involveonlyuniversalquanticationovercomputationsequences(duetotheuseofthe2(or2p)operator) whereasthoseexpressedbyformulasof3l(p) involveonlyexistentialquantication overcomputationsequences. Weconsiderthepositivefragments2L(p)+ and3l(p)+ obtainedfromtheabove languagesbyforbiddingtheuseofthenegationevenonatomicpropositions.we consideralsothefragmentsl(p)+ correspondingtothesubsetofl(p) formulasin positivenormalformwithoutnegations.wecantranslateanyformulaofl(p) which isinpositivenormalformintoanequivalentformulainl(p)+ byreplacingnegated atomicpropositions,i.e.,formulasintheform:p,bynewatomicpropositions. Thus,sinceanyformulaofL(p) hasanequivalentformulainpositivenormalform, wecanexpressinl(p)+ anypropertyexpressibleinl(p),modulothisencodingof theformulas:p.obviously,thesametranslationcanbedonefroml(p) tol(p)+ for2f2;3g. In2Lwecanexpressbranching-timepropertiesasforinstancethesafetypropertieswithrespecttothesimulationpreorder[3].Theclassoftheseproperties correspondstothefragmentof2lwithouttheleastxpointoperator.

pressiblebyanondeterministicbuchiautomaton[6],canbeexpressedin2l[4]. Forexample,thesafetyproperty[26],[29],[34]\alwaysP"canbeexpressedby theformulax:(p^2x).moreover,theguaranteeproperty(accordingto[34]) mulax:(p_2x).propertiesintheotherclassesinthehierarchygivenin[34] \eventuallypinanyinnitecomputationsequence"canbeexpressedbythefor- Furthermore,itcanbeshownthatany!-regularlinear-timeproperty,i.e.,ex- 17 areobtainedbyusingalternationsoftheandtheoperators.thepropertiesof 8CTL*canbeexpressedin2Lifwerestrictourselvestomodelswhosetransition relationistotalas8ctl*allowstoexpressgeneraleventuality.noticethatifthe isexpressedbytheformulax:(p_3true^2x),whichisneitherin2lnorin transitionrelationoftheconsideredmodelsisnotnecessarytotal,\eventuallyp" 3L. :P)X:(:init^2pX). rithmsforinvariantsandeventuallypropertieswhichinsomecasesconvergemuch faster.forexample,theformulainit)x:(p^2x)isequivalentto init.moreover,theymaybeusedinordertodenealternativecomputationalgo- propertieswhichcannotbeexpressedusingonlyfuturemodalities,e.g., X:(init_2pX)holdsexactlyinthesetofstatesreachablefromastatesatisfying Pasttimemodalitiescanbeusedfortwodierentaims:theyallowtoexpress Theformulasof3Larenegationsofformulasof2Landconversely. relatingtwopropertylattices,:2q1!2q2,preservesthemeaningoftheatomic First,wedenethenotionofconsistencywhichexpressesthatachosenfunction 6.2.Preservationresults propositionsdenedbyaninterpretationfunctionion2q1.isconsistentwith :2Q1!2Q2.Then,isconsistentwithIif i.e.,theimagesbyoftheinterpretationofpandof:parenoncontradictory. Lemma3saysthat inthecasethat(;)isaconnection consistencyof atomicpropositions. Denition9(Consistency) LetQ1andQ2betwosetsofstatesandI:P!2Q1aninterpretationandafunction withiexpressesthefactthatestronglypreservestheinterpretationofall IifforallatomicpropositionstheimagesofI(P)andI(P)byaredisjoint, aconnection,thenisconsistentwithiifandonlyif UnderthesameassumptionsasinDenition9,ifthereexistssuchthat(;)is Lemma3(Characterizationofconsistency) 8P2P:(I(P))\(I(P))=; 8P2P:((I(P)))=I(P)

18 Proof:AproofbycontradictioncanbeobtainedusingProposition7. Now,wegiveatheoremaboutthepreservationinthecasethatfortwogiven systemss2ands2arerelatedbys1vh;is2.thetheoremsaysthatpreserves formulasof3lfroms1tos2,epreservesformulasof2l(p) froms2tos1andif evens1'h;is2holds,thenaswellasepreservethewholel.furthermore, ifonereplacesvh;ibyh;i,oneobtainsanalogouspreservationresultsforthe fragmentsaugmentedbythecorrespondingpastmodalities. Theorem2(Preservationof2L(p),3L(p) andl(p) ) LetS1=(Q1;R1)andS2=(Q2;R2)betwotransitionsystemsandI1:P!2Q1, I2:P!2Q2twointerpretationfunctions. 1.IfS1vh;iS2(respectivelyS1h;iS2),then (A)preservestheformulasof3L+(respectively3Lp+ )fori1,andifis consistentwithi1thenpreservestheformulasof3l(respectively3lp) fori1. (B)epreservestheformulasof2L+(respectively2Lp+ )fori2,andifeis consistentwithi2thenepreserves2l(respectively2lp)fori2. 2.IfS1'h;iS2(respectivelyS1h;iS2andS2he;eiS1)thenpreserves theformulasofl+(respectivelylp+ )fori1andifisconsistentwithi1then preservestheformulasofl(respectivelylp)fori1. Proof:TheproofthatpreservesL+ifS1'h;iS2consists,duetoLemma2, inshowingthatforanyformulaf2l+andforanyvaluationv,wehave (jfjs1;i1(v))jfjs2;i1((v)). Theproofisdonebyinductiononthestructureoff,andforalloperators(includingxpointoperators),except3and2weneedonlythemonotonicityofin ordertoestablishthisfact.for3weneedthefactthats1vh;is2andfor2 weneedthefactthats2vhe;eis1.thisproofisgiveninappendixa.2. TheproofofpreservationofLp+ undertheconditionthats1h;is2isobtained bylemma1sayingthatforwardandbackwardsimulationimpliess1vh;is2 ands1?1vh;is2?1(wheresi=(qi;r?1 i))andtheobservationthatpost[r]= pre[r?1]. Finally,ifisconsistentwithI1,itisstraightforwardtodeducethat (j:pjs2;i1)j:pjs1;i1. NoticethatwehavealsopreservationofLp+ byebyexchangingtherolesof andeandofs1ands2andthenusingsymmetricalarguments.now,theproofs of(1a)and(1b)areobviousfromthefactthatforthepreservationof3l(p)+ by weneedonlytheconditionthats1vh;is2(respectivelys1h;is2),and forthepreservationof2l(p)+ byetheconditionthats1vhe;eis2(respectively

S1he;eiS2),whichisequivalenttoS1vh;iS2(respectivelyS1h;iS2). ItisknownthatinordertohavestrongpreservationofLoneneedstheexistenceof abisimulationbetweenthetransitionsystemss1ands2(theorem4givestheexact 19 offragmentsoflundertheweakerconditionthatistheexistenceofamutual conditions).byusingtheorem1,oneobtainsfromtheorem2strongpreservation simulationbetweens1ands2andtheadditionalconditionsrequiredintheorem1: Theorem3(Strongpreservationof2L(p) LetS1=(Q1;R1)andS2=(Q2;R2)betwotransitionsystems.IfS1vh;iS2 ands2vh0;0is1(respectivelys1h;is2ands2h0;0is1)for;0such that00=0,then 1.IfId0forsomeQ1,then Furthermore,ifisconsistentwithI,thenstronglypreserves3L(respectively3Lp)forIon. stronglypreserves3l+(respectively3lp+ and3l(p) 2.IfIde0eforsomeQ2,then Theorem4(StrongpreservationofL(p) Proof:(1)isadirectapplicationofTheorem1usingTheorem2.(2)isobtained inthesamewaybyusingproposition8whichguaranteese0ee0=e0. tively2lp)forion. Furthermore,ifeisconsistentwithI,thenestronglypreserves2L(respec- estronglypreserves2l+(respectively2lp+ ) )onforanyinterpretationi:p!. LetS1=(Q1;R1)andS2=(Q2;R2)betwotransitionsystems.IfS1'h;iS2 (respectivelys1h;is2ands2he;eis1)andee=ethen taineddirectlyfromtheorems1and2byreplacing0byeandusingthefact Proof:Astheprecedingtheorem,theproofofstrongpreservationbyisob- 1.stronglypreservesL(respectivelyLp)onIm(e)foranyinterpretation thatidim(e)e(proposition8)andthefactthatisconsistentwithany 2.estronglypreservesL(respectivelyLp)onIm()foranyinterpretation I1:P!Im(). I1:P!Im(e)and I1:P!Im()byusingthesameargumentsasintheproofofTheorem1.The proofforeissymmetrical.

f22lpandaninterpretationfunctioni:p!2q,onecanproceedasfollowsin 20 6.3.Application Theorem2providesthebasisforourvericationmethodbyusingabstraction. GivenaprogramS=(Q;R),asetPofatomicpropositionsoccurringinformula ordertoverifythatssatisesf,i.e.,jfjs;i=q: (1)GiveanabstractionrelationQQAwhichistotalonQandthecorrespondingabstractionfunction=post[]. (2)ComputetheabstractsystemSandverifywhetherthecharacteristicsetof NoticethatasucientconditionforthisisthatjfjS;I=QAexpressingthat fholdsons.iftheanswerin(2)ispositiveandnoatomicpropositionoccurs negatedinf,thenusingtheorem2.(1b),weobtain (3)SsatisesfwiththeinterpretationfunctioneI,i.e.,jfjS;eI=Q. fons,obtainedusingtheinterpretationfunctioni,iscontainedinthe image(q)ofconcretestates,thatmeanswehavetoverifythat Iffurthermore,I(P)=(eI)(P)foranyP2Pthatoccursinf,then e(jfjs;i)=q. thatthisamountstoevaluateastrongerpropertythanf;therefore,themethod functions(infnegationcanonlybeappliedtoatomicpropositions),wededuce Thismeans(byLemma3)thatinordertoapplythevericationmethodone needstheconsistencyofwithiforallatomicpropositionsoccurringnon negatedinf.forpropositionsp2poccurringonlynegatedinf,computing jfjs;iamountstoevaluatefonswithinterpretatione((i(p)))of:p;as e((i(p)))i(p)isalwaystrueandasalloperatorsinfrepresentmonotonic SsatisesfunderinterpretationI,i.e.,jfjS;I=Q. acounter-example,showingthatoneofthestatesinq0doesnotsatisfyf,orwe havetotrywithamoreprecisesetofabstractstatesandcorrespondingconnection. thatsvh;isa(respectivelysh;isaiffcontainspasttimemodalities). Iftheanswerin(2)isnegative,i.e.,e(jfjS;I)=Q0Q,wecantrytond Obviously,insteadoftheabstractsystemS,wecanuseanysystemSAsuch propositionsoccurringonlynegatedinf. canbeappliedeveniftheconsistencyrequirementidnotfullledforatomic functionsoftheatomicpropositionsiandi.inthatcase,itisshownthatthe fromqtoqasuchthatandearerespectivelyconsistentwiththeinterpretation correspondstoh;i-simulationinducedbyrelationswhicharetotalfunctions Asimilarmethodisappliedin[10].Thenotionofhomomorphismconsideredthere

underthiscondition 2L(p) logic8ctl*ispreservedfroms2tos1undertheconditionthatonlyinnitecomputationsequencesareconsidered.thisresultisgeneralizedbytheorem2since thenotionofexacthomomorphismconsideredtherecorrespondstobisimulations arerespectivelyconsistentwiththeinterpretationfunctionsiandi.ifs1and inducedbyrelationswhicharetotalfunctionsfromqtoqasuchthatande ismoreexpressivethan8ctl*.furthermore, 21 eralnotholdonabstractsystemssuchassortheabstractsystemsproposedin because,ifisatotalfunction,wehave=?1). eventualitypropertiesarepreserved,as,eveniftheyarepreserved,theydoingen- [10]. Itshouldalsobenoticedthatitisnotimportanttochooseaframeworkinwhich ThisresultisgeneralizedbyTheorem2(noticethatthistheoremcanbeapplied S2arerelatedbyanexacthomomorphism,thelogicCTL*isstronglypreserved. -abstractionofsifandonlyifsvsa,i.e.,post[]pre[r]fpre[]pre[ra]). wecanalsodeneothernotionsofsimulation,wherethefunctionrepresentinga Usingthisdenitionweobtainpreservationoftheformulasof2Lfromtheabinition4wereplacethefunctionsprebyfpre,thenS2isah;i-abstractionof transitionsystemischosentobefpre[r],post[r]orgpost[r].takingtherstchoice, weobtainpreservationoftheformulasof3lfromsatos.noticethat,ifindefstractsystemsatotheconcretesystems.asalmostallpropertiesweareinterestedinarein2l,thisisagoodnotionofabstraction.but,asalreadymentioned, Atthispoint,wecanalsodiscussthechoiceofournotionofabstraction(SAisa below;andsimilarasin[12],onemayusesuchapairofapproximationsinorder toevaluateanypropertyofl. ulationfromabove,andpropertiesexpressiblein3lbyapproximatingitfrom propertiesexpressiblein2lcanbeveriedbyapproximatingasystembysim- thesystems1understudyifandonlyifs2he;ei-simulatess1.thatmeansthat it,suchasrr:whereristhesubsetoftransitionrepresenting\stuttering"or \observable"or\non-stuttering"steps. \nonobservable"steps,denotesthetransitiveclosureandr:isthethesetof properties,wemustreplacethetransitionrelationrbysometransitiveclosureof wayassbygroupingsetsofstatesintoasingleabstractstate,andallowing onlythoseabstracttransitionscorrespondingtoatransitionofeverycorresponding concretestate.inordertoobtainabstractsystemsallowingtoverifyreachability fragment3l,doingeneralnotholdonabstractsystems,denedinasimilar However,thereachabilityproperties,whicharetheinterestingpropertiesinthe inanotionofabstractionpreservingonlypastmodalities;however,aswehave seenthisisnotveryinteresting,asbyreplacingvby(whichisnotreallya constraintinpractice)oneobtainspreservationofbothfutureandpastmodalities. ThechoicetorepresentthetransitionrelationsRbythefunctionpost[R],results

hasbeencalculatedinsection4.2. Mutualexclusionbetweenthereadersandthewriterscanbeexpressedbythe Weapplytheabovevericationmethodtoreaders-writersforwhichanabstraction 22 followingformula Reader/writerexamplecontinued impliesthatforanysubsequentstate,mutualexclusionholds. presentationinsuchasmallexample. predicatesonprogramvariablesdirectlyintheformulas,whichsimpliesabitthe tiontranslatingthemintopredicatesonconcreteprogramvariables,weusethese Thisformulastatesthatstartingfromastatewithnoactivereadersandwriters Notice,thatinsteadofpropositionalvariablesandanexplicitinterpretationfunc- f=(ar=0^aw=0))x:((ar=0_aw=0)^2x). Wehavetoshowthatpreservesthefourbasicpredicatesoccurringintheformula, namely(ar=0),:(ar=0),(aw=0)and:(aw=0),i.e., foreachoneofthesepredicates.thiscanbeeasilyveried,e.g.for(ar=0),we havepost[](ar=0)=b1andpre[](b1)=(ar=0). resultingintheformula InordertoverifyfonS,wehavetotranslatetheatomicpredicatesbypost[], fa:(b1^b2))x:((b1_b2)^2x) pre[](post[](p))=p jfajs=true.bytheorem2,wehavethatmutualexclusionholdsontheconcrete todothisevaluationecientlyiftheabstractdomainisnite. program.therecentdevelopmentsofbdds[5]andtoolsmanipulatingthem,allows 7.Compositionalityofsimulationwithrespecttoparallelcomposition Intheprevioussectionswegaveamethodreducingthevericationofaproperty ByusingclassicalsymbolicmodelcheckingforCTL(seee.g.in[38]),weobtain ofthesamepropertyonsomeabstractions=(q;r). Thequestionthenariseswhetheritispossibletocomputeabstractionsofcomplex programsastheparallelcompositionofabstractionsoftheircomponentsinorder ofsomeprogramrepresentedasatransitionsystems=(q;r)totheverication simplerprograms,theapplicationofthismethodrequiresthecomputationofthe correspondingglobaltransitionrelationfromwhichanabstractioncanbecomputed. Whendealingwithcomplexprogramsobtainedastheparallelcompositionof

isguaranteedifthecompositionalityproperty toavoidbuildingthetransitionrelationassociatedwiththecomplexprogram.this (S1v1S10)and(S2v2S20) (S1jjS2)vf(j;1;2)(S10jjS20) 23 holds,wherejjisaparallelcompositionoperatorandf(jj;1;2)anabstraction relationdependingonjj,1and2. parameterizedbyarbitraryrelationsandtherelationusedtoobtaintheabstraction ofthecomposedsystemiscomputedfromtheabstractionrelationsappliedtoits ferentparallelcompositionoperatorsandbytakingf(jj;1;2)=1\2. respecttoparallelcomposition.mostofthemconcernsynchronouscompositionor theparticularcasewherethedomainsofthecomposedprocessesaredisjoint. Inthissectionwepresentcompositionalityresultsfor-simulationforthreedif- components. Thereexistalreadymanycompositionalityresultsforsimulationrelationswith tionofthecomplexprogramandtheabstractionresultingfromtheparallelcom- positionoftheabstractionsofthecomponents. Anotherproblemstudiedinthissectionistherelationshipbetweentheabstrac- Noticethatanimportantdierencewiththeseresultsisthatoursimulationsare AsinSection4.2weconsidertransitionsystemsSdescribedbyfamiliesofoftransitionrelationsrepresentedbysetsofbinarypredicatesonasetofvariablesX, i.e.,s=fri(x;x0)gi2iwheretheelementsofiareconsideredaslabelsusedfor Theseresultsallowtocomparethetwoapproachesconcerningthequalityofthe obtainedabstractions. 7.1.Denitionofparallelcomposition LetSi=fRij(Xi;X0i)jj2Iig;i2f1;2gandAI1I2beasynchronizationset(indicatingwhichrelationsmustsynchronize).Furthermore,takeA1= fij9j:(i;j)2aganda2=fjj9i:(i;j)2ag(aiaretheprojectionsofaoni1 otherscanbeconsideredasparticularcasesofit. Denition10(Parallelcomposition) (jjj)andmixed(j[]j).mixedparallelcompositionisthemostgeneraloneandthe respectivelyi2).wedenetheoperatorsjjj;a;j[a]jasfollows: Weconsiderthreetypesofparallelcomposition,synchronous(),asynchronous labeledtransitionsystemsasitallowsustodeneparallelcompositionofprograms sharingvariables. synchronizationpurposesinparallelcomposition.weusethisrepresentationof mixedcompositionj[a]j:

24S1j[A]jS2=fR1i^R2jj(i;j)2Ag[ synchronouscompositiona: asynchronouscompositionjjj: whereforanysetofvariablesx=fx1;::;xng,stablexisthepredicate S1AS2=fR1i^R2jj(i;j)2Ag (x01=x1)^:::^(x0n=xn). fr1i^stablex2?x1ji62a1g[fr2j^stablex1?x2jj62a2g Comments: Themixedcompositionoperatorforcessynchronizationofpairsoftransition S1jjjS2=fR1i^stableX2?X1ji2I1g[fR2j^stableX1?X2jj2I2g Synchronouscompositionisaspecialcaseofmixedcomposition,whereonlythe j62a2.thisoperatorallowstoexpresstheoperatorsofcsp[19]orlotos lattercorrespondstomovesofeithersomer1ifori62a1orofsomer2jfor [20]bysimulatingmessagecommunicationbycommunicationthroughcommon A,ormovesperformedbyonecomponentwhiletheotherremainsidle.The variables. relationsbelongingtoa.r1j[a]jr2canperformeithermovesresultingfrom executionofsynchronoustransitionsispossible.inthecasewherea=i1i2, thesynchronousexecutionoftransitionsinsomer1iandr2jsuchthat(i;j)2 Asynchronouscompositionisthespecialcaseofthemixedcompositionwhere thisoperatoristhesameas^,andthisistheprogramcompositionoperator operatorsofsccs[33],ofs/rmodels[23]andtheoneusedin[15]. usedintla[27].itcanalsobeusedtodescribetheparallelcomposition nizationsetasbefore.then, Lemma4LetbeSi=fRij(Xi;X0i)jj2Iig;i2f1;2gandAI1I2asynchro- orofsomer2jwherej2i2.thisoperatorisexactlythe\unionoperator"of A=;.ThatmeansthatallmovesaremovesofeithersomeR1iwherei2I1 IfAsuchthatA1=A2=;,thenS1AS2=S1j[A]jS2 S1jjjS2=S1j[;]jS2 Unity[11]. IfA=I1I2,thenR1AR2=R1^R2 S1j[A]jS2=fR1if(i;j)gR2jg(i;j)2AjjjfR1i^stableX2?X1gi62A1jjj S1=jjji2IfR1igwherejjji2Iistheobviousn-aryextensionofjjj fr2j^stablex1?x2gj62a2

theothers.wepreferhowevertoconsiderthethreeoperatorsbecausetheygive eachonerisetospecicresults. variables. Themixedcompositionoperatoristhemostgeneraloneasitallowstoexpress ThelastitemcomesfromthefactthatalltheR1iaredenedonthesamesetof 25 7.2.Compositionalityresults Now,wegiveforalloperatorsofDenition10,conditionsontheabstractionrelationsi,underwhichtherule iswhethertheabstractionsr11jjr22and(r1jjr2)1^2arecomparableinordertoknowwhichwayofcomputingabstractionsgivesbetterapproximations: directcomputationfromthecompoundsystemorcomputationbycompositionof holds.furthermore,weareinterestedinapplyingthisruleintheparticularcase wheres0i=sii(denedasinsection4.1).inthatcaseaninterestingquestion (Comp)(S1v1S10)and(S2v2S20) abstractionsofthecomponents.intuitivelyonewouldthinkthat S1jjS2v1^2S10jjS20 toobtainthismodiedrule,slightlyweakerconditionsthanthoserequiredforthe holdsalways.however,theseconditemsofthefollowingtheoremsshowthatthis isonlytruewithoutrestrictionsforsynchronousparallelcomposition.noticealso that,ifthisimplicationholds,thenalsotherule,obtainedbyreplacingin(comp) thesimulationpreordervbyforwardandbackwardsimulation.infact,inorder seconditemofthefollowingtheoremsarenecessary,butaseventhestrongerones arealmostalwayssatisedinpractice,weproposetheinterestedreadertolookat (R1jjR2)1^2)R11jjR22 implication Assumption1ThroughouttherestofthesectionweconsiderasetofvariablesX forsynchronouscompositionrelativelystrongconditionsarenecessary;whereasfor [28]formoredetails. Thethirditemsofthefollowingtheoremsshowthatinordertoobtaintheinverse asynchronouscompositiontheconditionsarerelativelyeasytofull. oftheformx1[x2wherex1andx2arenotnecessarilydisjoint,twotransition abstractvariables. systemssi=frij(xi;x0i)jj2iig;i2f1;2gandxa=x1a[x2aasetof R11jjR22)(R1jjR2)1^2 variablesandxila=xia?xcathelocalabstractvariablesofsi. localvariablesofsiandanalogouslyxca=x1a\x2a,thesetofcommonabstract WedenotealsoXc=X1\X2,thesetofcommonvariables,Xil=Xi?Xcthe

26Weconsideralsotworelationsrelatingtheconcreteandtheabstractdomains, i(xi;xia),whicharetotalonxiandsuchthat1^2istotalonx.inordertosimplifytheexpressionoftheresultsandbecauseitdoesnotrestrictgenerality,wesupposeinthesequelthattherelationsicanbeputintotheform i=il(xi;xila)^ic(xi;xca),i.e.,theabstractlocalandcommonvariablesdo Theorem5(compositionalitywithrespectto) thetotalityof1l,2land1c^2c. notdependoneachother.thisimpliesthatthetotalityof1^2isequivalentto UnderthehypothesesofAssumption1,onehas 3.Ific:XcA!Xcarefunctionsfori=f1;2g,then 2.(R1AR2)1^2)R11AR22 1.Ific:Xc!XcAarefunctionsfori=f1;2g,then R11AR22=(R1AR2)1^2 S1AS2v1^2S01AS02 SiviS0i;i=1;2 get1.ific=c(xc;xca),then tonicityarguments andisdeferredtoappendixa.3. Theorem6(compositionalitywithrespecttojjj) Proof:Theproofisrathertechnical exceptthatof(2)whichusesonlymono- UnderthehypothesesofAssumption1andiffurthermoreil=il(Xil;XilA),we 3.Ific=c(Xc;XcA)andilareonto(8XilA9Xil:il(Xil;XilA),then 2.Ifil:Xil!XilAarefunctionsfori=f1;2g,then (R1jjjR2)1^2)R11jjjR22 R11jjjR22)(R1jjjR2)1^2 S1jjjS2v1^2S01jjjS02 SiviS0i;i=1;2 Theorem7(compositionalitywithrespecttoj[]j) Proof:ThecompleteproofisgiveninAppendixA.3. UnderthesamehypothesesasinTheorem6,weget

1.Ific:Xc!XcAarefunctionsfori=f1;2g,then 2.Ifil:Xil!XilAarefunctionsfori=f1;2g,then S1j[A]jS2v1^2S01j[A]jS02 SiviS0i;i=1;2 27 Proof:ThefactthatR1j[]jR2canbeexpressedbyusingonlyandjjjasgivenin Lemma4,andthattheconditionsofbothoftheprecedingtheoremsaresatised ineachofthecorrespondingpointsisenoughtoprovethetheorem. 3.Ific:XcA!Xcarefunctionsfori=f1;2gandilareonto,then R11j[A]jR22)(R1j[A]jR2)1^2 (R1j[A]jR2)1^2)R11j[A]jR22 ideaistheuseofh;i-simulationwhichisthesameasthestandardsimulation 8.Conclusion Thepaperstudiespropertypreservingtransformationsforreactivesystems.Akey (parameterizedbyarelation)oftenusedtodeneimplementations.furthermore,h;i-simulationsinduceabstractinterpretationsandthisallowstoapplyan models,labeledtransitionsystemswithsilentactionsandusingthewell-known alencesthataredenedintermsofsimulationsorbisimulationswithsilentactions. Forinstance,onecandeneah;i-observationalequivalencebyconsideringas existingpowerfultheoryforprogramanalysis. factthatobservationalequivalenceisstrongbisimulationequivalenceonamodied transitionrelation. trivialsystems.forthis,akeyproblemisthechoiceofappropriateabstraction Theresultspresentedcanbeadaptedsoastobeappliedtopreordersandequiv- relationsdependingonthepropertiestobeveried.ingeneral,thistaskrequires preservationofthesepredicateshelpndingtheminimalnecessaryabstractdomain.alsotheresultsofsection7arehelpfulfortheuserofthemethodas appropriateabstractionsforcomponentsareeasiertondthanabstractionforthe compoundsystem. Animportantissueistheapplicationoftheresultstothevericationofnon- adeepknowledgeoftheconcreteprogramtobeveriedandcannotbeautomated. However,thepredicatesoccurringintheformulaandtherequirementsforthe abstractionrelationisgiven,therestofthemethodcanbemechanized:computationoftheabstraction,vericationoftheformulaandcheckingpreservationof thepredicates.wehaveimplementedasymbolicvericationtoolsupportingthis Inthecasethatboth,theconcreteandtheabstractdomainsarenite,oncean

methodfornitestateprogramsencodedasbdds[17],[28]:programsareparallelcompositionsofcomponentswhicharepredicates(justastheprogramused obtainedbycomposingandabstractingthecomponentsinanyorderusingabstractionrelationsigivenbypredicatesonabstractandconcretevariables.internally intheexampleinsection4.2)onbooleanvariables.anabstractprogrammaybe (operationonintegers,memoriesandbuers)byacorrespondingabstractoperationonveryreducedabstractdomains.thisexampleshowsthatourresultscanbe computationofniteabstractionsofinnitestatesystemsdeservefurtherstudy. AppendixA A.1.ProofofProposition11 WewanttoshowthatifS=(Q;R)isatransitionsystemandandQQA totalonqsuchthat=?1,thensisafaithfulabstractionofsvia.more First,letusshowthatSv?1SA,thatis, precisely,wewanttoshowthatforanysa=(qa;ra)suchthatsvsaand RAR,SAandSare?1-bisimilar. post[?1]pre[r]fpre[?1]=post[?1]post[]pre[r]pre[]fpre[?1]. Bydenition,wehavepre[R]=post[]pre[R]pre[].Thus,bysubstitution ByProposition2andbythefactthatpre[]=post[?1],weobtain post[?1]pre[r]fpre[?1]=post[?1]pre[r]post[?1]fpre[?1] fpre[]. post[?1]pre[r]fpre[?1]pre[ra]:() inordertobeabletocomputeanappropriateabstractsystem. In[18],weappliedthesamevericationmethodtoaninnitestatesystem,a ample,theabstractprogramcouldnotbeobtainedfullyautomatically.ithasbeen computedfromtheconcreteprogrambyreplacingeveryconcretebasicoperation distributedcachememory[13]whichisknowntobediculttoverify.forthisex- Forthisprotocol,theuseofthecompositionalityresultsofSection7wasessential cationofproperties.usingthistool,wehaveveriedaprotocoldescribedin[13]. allpredicatesarerepresentedbybdds.asymbolicmodelcheckerallowstheveriappliedforthecomputationofabstractionsofinnitestatesystems.however,the Now,since=?1(byhypothesis)and(post[?1];fpre[?1])isaconnection (byproposition6),wehave,post[?1]=post[]and post[?1]fpre[?1]id,thus, post[?1]pre[r]fpre[?1]post[]pre[r]fpre[]. Finally,sincebyhypothesiswehave, SvSA,i.e.,post[]pre[R]fpre[]pre[RA],weobtain, post[?1]pre[r]fpre[?1]pre[ra]whichis(*).

Now,since(?1)?1=(?1),weshowthatSAv?1S,thatistosay, post[?1]pre[ra]fpre[?1]pre[r]: Byhypothesis,wehaveRAR,suchitissucienttoshowthat post[?1]pre[r]fpre[?1]pre[r]() 29 A.2.ProofofTheorem2 InordertocompletetheproofofTheorem2itremainstoshowthatifS1=(Q1;R1) Asintherstpartoftheproofweobtain, post[?1]pre[r]fpre[?1]post[]pre[r]pre[], Lemma2,itissucienttoprovethatforanyformulafinL+andforanyvaluation ands1=(q2;r2)aretransitionsystemsandi:p!2q1isaninterpretation function,suchthats1'h;is2thenpreservestheformulasofl+fori.by whichisequivalentto(**). V,wehave Tosimplifythenotations,weomitthevaluationVwheneveritisnotrelevantina proof. (j?js1;i)j?js2;iand(j>js1;i)j>js2;ias(;)=;and(q2)q1. (jpjs1;i)=jpjs2;ibydenitionoftheinterpretationfunction. (jxjjs1;i(v))=(vj)=jxjjs2;i((v)) (jfjs1;i(v))jfjs2;i((v))orequivalentlyjfjs1;i(v)(jfjs2;i((v))). j2fjs1;i=fpre[r2](jfjs1;i)bydenitionofthesemantics.thedualofthe conditionfors1vhe;eis2isfpre[r1]fpre[r2].bysubstitution,weget, (j3fjs1;i)=pre[r1](jfjs1;i) whichisequivalenttoj2fjs1;i(j2fjs2;i): Byinductionhypothesis (jfjs1;i)jfjs2;i weobtain, j2fjs1;ifpre[r2](jfjs1;i): bydenitionofthesemanticsandmonotonicityof.asidq1,weget (j3fjs1;i)pre[r1](jfjs1;i): j2fjs1;ifpre[r2](jfjs2;i); AsS1vh;iS2,i.e.,pre[R1]pre[R2],weget (j3fjs1;i)pre[r2](jfjs1;i):

(jf2_f1js1;i)=(jf2js1;i[jf1js1;i) 30Byinductionhypothesis (jfjs1;i)jfjs2;i follows bydenitionoftheinterpretationfunction.asdistributesover[,wehave (j3fjs1;i)pre[r2](jfjs2;i)=j3fjs2;i: Ananalogousproofcanbeobtainedforconjunction. jx:fjs2;i((v))=tfp2q2:jfjs2;i[p2=x]((v))p2g, wherevisavaluationonq1ofthefreevariablesoff.as(;)isaconnection Byinductionhypothesis,weobtain (jf2_f1js1;i)jf2js2;i[jf1js2;i=jf2_f1js2;i: (jf2_f1js1;i)=(jf2js1;i)[(jf1js1;i): (IdQ2)andismonotonic, Usingtheinductionhypothesisforfwithvaluation(P2)forXgives implies whichimpliesnallybytransitivity, jfjs1;i[(p2)=x](v)(jfjs2;i[((p2))=x]((v))) (jfjs2;i[((p2))=x]((v)))(p2): jfjs2;i[p2=x]((v))p2() TfP2:()gTfP2:()g,i.e., Thus,everyP2satisfying(*)satisesalso(**).Thisimplies Bydistributivityofoverintersection,weobtain TfP2:jfjS1;I[(P2)=X](V)(P2)g TfP2:jfjS2;I[P2=X]((V))P2g=jX:fjS2;I((V)): \f(p2):jfjs1;i[(p2)=x](v)(p2)g(jx:fjs2;i((v))): jfjs1;i[(p2)=x](v)(p2)(): wededucethat f(p2):jfjs1;i[(p2)=x](v)(p2)gfp1q1:jfjs1;i[p1=x](v)p1g leastxpointjx:fjs1;i(v).fromthefactthat, Itremainstoshowthatf(P2):jfjS1;I[(P2)=X](V)(P2)gcontainsthe Ananalogousproofcanbeobtainedforthegreatestxpoint. whichcompletestheproof. Tf(P2):jfjS1;I[(P2)=X](V)(P2)g TfP1Q1:jfjS1;I[P1=X](V)P1g=jX:fjS1;I(V)

notationforthecompositionofrelations: A.3.ProofsofTheorems5and6 WesupposeallthenotationsandhypothesesintroducedinAssumption1ofSection7.2fortheformulationofthetwotheorems.Furthermore,weusethefollowing IfR1(X;X0)andR2(X0;X00)arepredicatesrepresentingrelations,werepresent 31 9X0:R1(X;X0)^R2(X0;X00). A.3.1.ProofofTheorem5 1.Inordertoshowthestabilityofvwithrespecttosynchronouscomposition byr1r2thecompositionoftherelations,i.e.,r1r2representsthepredicate weusedenition5.moreprecisely,weshowthat, (9(X1l;Xc):1((X1l;Xc);(X1lA;XcA))^9i:R1i((X1l;Xc);(X01l;X0c))) implies (**)(W(i;j)2A(R1i^R2j)?1)(1^2))(1^2)(R01^R02)?1. (*)(Wi2I1R1i)?11)1R0?1 9(X01lA;X0cA):1((X01l;X0c);(X01lA;X0cA))^R01((X1lA;XcA);(X01lA;X0cA))) where(*)canbeexpressedas: 8(X01l;X0c)8(X1lA;XcA): 1and(Wj2I2R2j)?12)2R0?1 2 9(X0cA;X02lA):2((X0c;X02l);(X0cA;X02lA))^R02((XcA;X2lA);(X0cA;X02lA))): and(**)canbeexpressedas: 8(X01l;X0c;X02l)8(X1lA;XcA;X2lA): (9(Xc;X2l):2((Xc;X2l);(XcA;X2lA))^9j:R2j((Xc;X2l);(X0c;X02l))) (9(X1l;Xc;X2l):1((X1l;Xc);(X1lA;XcA))^2((Xc;X2l);(XcA;X2lA))^ ^8(X0c;X02l)8(XcA;X2lA): )9(X01lA;X0cA;X02lA):1((X01l;X0c);(X01lA;X0cA))^2((X0c;X02l);(X0cA;X02lA))^ R01((X1lA;XcA);(X01lA;X0cA))^R02((XcA;X2lA);(X0cA;X02lA))): ItisquiteeasytoseethatifwechoosethesameX0candXcAinpart1 9(i;j)2A:R1i((X1l;Xc);(X01l;X0c))^R2j((Xc;X2l);(X0c;X02l)) R01((X1lA;XcA);(X01lA;X0cA))andR02((XcA;X2lA);(X0cA;X02lA)).Thisimplies X0cAthatcanbechosenthenitisunique,whichinducesby(*)that Thefactthatthaticare(thesame)functionsassuresthatifthereexistsa ondom(x)issucienttobeabletochoosethesamex0casuchthatboth 1((X01l;X0c);(X01lA;X0cA))and2((X0c;X02l);(X0cA;X02lA)). and2of(*),andifwecanchoosethesamexc,thentotalityof1^2

2.Wehavetoshowthat(R1AR2)1^2)R11AR22. 32(**).Noticethattherequiredconditionsoniarealsonecessaryifnomore 3.Wehavetoshowtheinverseimplicationof(2).Weshowthat informationonthetransitionrelationsriandr0iisavailable. (R1AR2)1^2=W(i;j)2A(1^2)?1(R1i^R2j)(1^2). As1^2)i,R1i^R2j)R1iandR1i^R2j)R2j,wehave W(i;j)2A?1 W(i;j)2A(1^2)?1(R1i^R2j)(1^2)) R1i1^R2j2((X1lA;XcA;X2lA);(X01lA;X0cA;X02lA))= 1((X1l;Xc);(X1lA;XcA))^R1i((X1l;Xc);(X01l;X0c))^1((X01l;X0c);(X01lA;X0cA)) 9(X1l;Xc)9(X01l;X0c): 8(i;j)2A:R1i1^R2j2)(R1i^R2j)1^2: 1R1i1^?1 2R2j2whichisequivalenttoR11AR22. isadierentproofofimplication(2)).inordertogettheimplication(3),we Theexpressionfor(R1i^R2j)1^2diersfromthisonebythefactallthe derlined),i.e.,inbothsubexpressionsthesamexcandx0cmustbechosen(this 2((Xc;X2l);(XcA;X2lA))^R2j((Xc;X2l);(X0c;X02l))^2((X0c;X02l);(X0cA;X02lA)) existentialquanticationshavetobeputoutsideofthemainconjunction(un- ^9(Xc;X2l9(X0c;X02l): A.3.2.ProofofTheorem6 onthetransitionrelationsriandr0iisavailable. Xc.Noticethattherequiredconditionisalsonecessaryifnomoreinformation isobviouslyguaranteedbytheconditionthaticarefunctionsfromxcainto mustbesurethatchoosinginbothexistentialquanticationsthesamexcand 1.Inordertoshowthestabilityofvwithrespecttojjj,weuseagainDenition5; thesamex0cwedonotobtainlesstransitionsthanwithoutthisconstraint.this (*)(Wi2I1R1i)?11)1R0?1 soweshowthat analogouslyforr2.weshowtheimplicationforsomer1i. that 8i2I1:(R?1 implies (***)((Wi2I1R1i))jjj(Wj2I2R2j)))?1(1^2))(1^2)(R01jjjR02)?1. Ascompositionofrelationsdistributesoverdisjunction,itissucienttoshow 1i^stableX2l)(1^2))(1^2)(R0?1 1and(Wj2I2R2j)?12)2R0?1 1^stableX2lA)and 2

(9(X1l;Xc;X2l):1((X1l;Xc);(X1lA;XcA))^2((Xc;X2l);(XcA;X2lA))^ 8(X01l;X0c;X02l)8(X1lA;XcA;X2lA): R1i((X1l;Xc);(X01l;X0c))^X2l=X02l (R?1 1i^stableX2l)(1^2))(1^2)(R0?1 1^stableX2lA)canbeexpressedas 33 )9(X01lA;X0cA;X02lA):1((X01l;X0c);(X01lA;X0cA))^2((X0c;X02l);(X0cA;X02lA))^ Thisexpressiondiersfromtherstconjunctof(*)(seeitsexpressioninthe R1((X1lA;XcA);(X01lA;X0cA))^X2lA=X02lA). 2.Weshowthat cienttoshowthat1((x1l;xc);(x1la;xca))and1((x01l;x0c);(x01la;x0ca)) Thisisguaranteedbythethefactthat2ldoesnotdependonXcandthefact 8i2I1:(1^2)?1(R1i^stableX2l)(1^2))(1?1R1i1)^stableX2lA and2((xc;x2l);(xca;x2la))implies2((x0c;x2l);(x0ca;x2la)). that2ccoincideswith1c. previousproofitem(1))byaddingalltheunderlinedparts.thus,itissufandanalogouslyforr2.wehave, whereas R1((X1l;Xc);(X01l;X0c))^1((X01l;X0c);(X01lA;X0cA))^ 9(X1l;Xc;X2l)9(X01l;X0c;X02l):1((X1l;Xc);(X1lA;XcA))^ E1= 2((X0c;X02l);(X0cA;X02lA))^X2l=X02l (1^2)?1(R1i^stableX2l)(1^2)((X1lA;XcA;X2lA);((X01lA;X0cA;X02lA))= E2= (1?1R11^stableX2lA)((X1lA;XcA;X2lA);((X01lA;X0cA;X02lA))= 2((Xc;X2l);(XcA;X2lA))^ 9(X1l;Xc)9(X01l;X0c):1((X1l;Xc);(X1lA;XcA))^R1((X1l;Xc);(X01l;X0c))^ 3.Inordertoobtain(3),i.e.,E2)E1,itissucienttoshowthat thereexistsauniquex2lasuchthat2l(x2l;x2la). ordertoobtaine1)e2itissucienttoshowthate1)(x2la=x02la). wheretheunderliningindicatesthedierencesbetweenthetwoexpressions.in Thisisguaranteedbytheconditionthat2lisafunction,i.e.,thatforanyX2l 1((X01l;X0c);(X01lA;X0cA))^X2lA=X02lA bythefactthat2lisontoandthat2ccoincideswith1c. E2)9X2l:2l(X2l;X2lA)^2c(Xc;XcA)^2c(X0c;X0cA)whichisguaranteed

34 References 4.A.Bouajjani.FromLinear-TimePropositionalTemporalLogicstoaBranching-Time 3.A.Bouajjani,J.-C.Fernandez,S.Graf,J.Sifakis,andC.Rodriguez.Safetyforbranchingsemantics.In18thICALP,Madrid.LNCS510,SpringerVerlag,1991. June1992. tions.inworkshoponcomputer-aidedverication(cav),montreal.lncs630, puterscience,82(2),1991,1988.rstpublishedasreportsrc-29,decresearch 1.M.AbadiandL.Lamport.Theexistenceofrenementmappings.TheoreticalCom- 2.A.Bouajjani,S.Bensalem,C.Loiseaux,andJ.Sifakis.Propertypreservingsimula- 5.R.E.Bryant.Graphbasedalgorithmsforbooleanfunctionmanipulation.IEEE -calculus.rtc15,lgi-imag,grenoble,1989. Centerin1988. 6.J.R.Buchi.Onadecisionmethodinrestrictedsecondorderarithmetic.InInternationalCongressonLogic,MethodandPhilosophicalScience.StanfordUniversity Press,1962. 7.P.CousotandR.Cousot.Systematicdesignofprogramanalysisframework.InProc. Trans.onComputation,35(8),1986. 10.E.M.Clarke,O.Grumberg,andD.E.Long.Modelcheckingandabstraction.In 8.P.CousotandR.Cousot.Comparingthegaloisconnectionandwidening/narrowing 9.E.M.Clarke,E.A.Emerson,andE.Sistla.Automaticvericationofnitestateconcurrentsystemsusingtemporallogicspecication:apracticalapproach.In10thACM SymposiumonPrinciplesofProgrammingLanguages(POPL83),1983.Completever- approachestoabstractinterpretation.plilp'92,lncs631,pp269-295.springer 6thACMSymp.onPrincipleofProgrammingLanguages,1979. 11.K.M.ChandyandJ.Misra.ParallelProgramDesign.Addison-Wesley,Massachusetts, Verlag. 12.D.Dams,O.Grumberg,andR.Gerth.Abstractinterpretationofreactivesystems: sionpublishedinacmtoplas,8(2):244{263,april1986. 13.P.Ernberg,L.Fredlund,andB.Jonsson.Specicationandvalidationofasimple SymposiumonPrinciplesofProgrammingLanguages(POPL92).ACM,January 1992. 1988. Abstractionspreserving8CTL*,9CTL*andCTL*.IFIPconferencePROCOMET'94. 16.S.GrafandC.Loiseaux.Programvericationusingcompositionalabstraction.In 15.O.GrumbergandE.Long.Compositionnalmodelcheckingandmodularverication. 14.E.A.EmersonandJ.Y.Halpern.`Sometimes'and`notnever'revisited:Onbranching TAPSOFT93,jointconferenceCAAP/FASE.LNCS668,SpringerVerlag,April1993. Springer-Verlag,1991. InJ.C.M.BaetenandJ.F.Groote,editors,Concur'91,pages250{265.LNCS527, versuslineartime.in10thacmsymposiumonprinciplesofprogramminglanguages (POPL83),1983.alsopublishedinJournalofACM,33:151-178. overtakingprotocolusinglotos.technicalreportt90006,sics,sweden,1990. 17.S.GrafandC.Loiseaux.Atoolforsymbolicprogramvericationandabstraction. 18.S.Graf.Vericationofadistributedcachememorybyusingabstractions.Conference oncomputeraidedvericationcav'94,stanford.lncs818,springerverlag,1994. InConferenceonComputerAidedVericationCAV'93,HeraklionCrete.LNCS697, SpringerVerlag,1993.

21.H.Jifeng.Varioussimulationsandrenements.InREXWorkshoponStepwiseRe- 20.ISO.ISISO/OSI8807-LOTOS:aformaldescriptiontechniquebasedonthetemporal 19.C.A.R.Hoare.CommunicatingSequentialProcesses.PrenticeHallInternational, orderingofobservationalbehaviour.internationalstandard,iso,1989. nementofdistributedsystems,mook.lncs430,springerverlag,1989. 1984. 35 24.D.Kozen.Resultsonthepropositional-calculus.InTheoreticalComputerScience. 25.R.P.Kurshan.Analysisofdiscreteeventcoordination.InREXWorkshoponStepwise 23.J.KatzenelsonandB.Kurshan.S/R:ALanguageforSpecifyingProtocolsandother 22.B.Jonsson.Ondecomposingandreningspecicationsofdistributedsystems.In 286{292.IEEE,1986. North-Holland,1983. CoordinatingProcesses.In5thAnn.Int'lPhoenixConf.Comput.Commun.,pages REXWorkshoponStepwiseRenementofDistributedSystems,Mook.LNCS430, SpringerVerlag,1989. 28.C.Loiseaux.Vericationsymboliquedeprogrammesreactifsal'aided'abstractions. 29.O.Lichtenstein,A.Pnueli,andL.Zuck.Thegloryofthepast.InConferenceon 26.L.Lamport.Provingthecorrectnessofmultiprocessprograms.IEEETransactions 27.L.Lamport.Thetemporallogicofactions.TechnicalReport79,DEC,Systems ResearchCenter,1991. Thesis,UniversiteJosephFourier,Grenoble,January1994. RenementofDistributedSystems,Mook.LNCS430,SpringerVerlag,1989. 30.N.A.LynchandM.R.Tuttle.AnintroductiontoInput/Outputautomata.Report onsoftwareengineering,se-3(2):125{143,1977. 31.R.Milner.Analgebraicdenitionofsimulationbetweenprograms.InProc.Second 32.R.Milner.Acalculusofcommunicationsystems.InLNCS92.SpringerVerlag,1980. 33.R.Milner.AcalculusforSynchronyandAsynchrony.JournalofTheoreticalComputer LogicsofPrograms,LNCS194.SpringerVerlag,1985. 34.Z.MannaandA.Pnueli.Ahierarchyoftemporalproperties.InProceedingof9th MIT/LCS/TM373,MIT,Cambridge,Massachussetts,November1988. Int.JointConf.onArticialIntelligence,pages481{489.BCS,1971. Science,25,1983. 38.J.P.Queille.Lesystemecesar:Description,specicationetanalysedesapplications 35.O.Ore.Galoisconnexions.Trans.Amer.Math.Soc,55:493{513,February1944. 36.A.Pnueli.TheTemporalLogicofPrograms.In18thSymposiumonFoundationsof ACMSymposiumonPrinciplesofDistributedComputing,1990. 39.LuisE.Sanchis.Datatypesaslattices:retractions,closuresandprojections.In 37.A.Pnueli.Applicationoftemporallogictospecicationandvericationofreactive hout.lncs224,springerverlag,1986. reparties.thesis,universitescientiqueetmedicaledegrenoble,june1982. RAIROTheoricalcomputerscience,vol11,nr4,pages339{344,1977. systems:asurveyofcurrenttrends.incurrenttrendsinconcurrency,nordwijker- ComputerScience,13:45{60,1981. ComputerScience(FOCS77).IEEE,1977.RevisedversionpublishedinTheoretical 40.J.Sifakis.Propertypreservinghomomorphismsandanotionofsimulationoftransition systems.rr332,imag,grenoble,november1982.

3641.J.Sifakis.Propertypreservinghomomorphismsoftransitionsystems.InE.Clarke 42.P.Wolper.Temporallogiccanbemoreexpressive.InformationandControl,56,1983. SpringerVerlag,June1983. andd.kozen,editors,4thworkshoponlogicsofprograms,pittsburgh.lncs164,