A. Baltag, L.S. Moss, S. Solecki
Software Engineering (SEN)
SEN-R9922 November 30, 1999
Report SEN-R9922ISSN 1386-369XCWI
P.O. Box 94079
1090 GB AmsterdamThe Netherlands
CWI is the National Research Institute for Mathematicsand Computer Science. CWI is part of the StichtingMathematisch Centrum (SMC), the Dutch foundationfor promotion of mathematics and computer scienceand their applications.
SMC is sponsored by the Netherlands Organization forScientific Research (NWO). CWI is a member ofERCIM, the European Research Consortium forInformatics and Mathematics.
Copyright © Stichting Mathematisch CentrumP.O. Box 94079, 1090 GB Amsterdam (NL)Kruislaan 413, 1098 SJ Amsterdam (NL)
Telephone +31 20 592 9333Telefax +31 20 592 4199
TheLogicofPublicAnnouncements,CommonKnowledge,andPrivateSuspicions
AlexandruBaltag
CWI
P.O.Box94079,1090GBAmsterdam,TheNetherlands
Email:abaltag@cwi.nl
LawrenceS.MossandSlawomirSolecki
MathematicsDepartment,IndianaUniversity
Bloomington,IN47401,USA
Emails:lsm@indiana.eduandssolecki@indiana.edu
ABSTRACT
Thispaperpresentsalogicalsysteminwhichvariousgroup-levelepistemicactionsareincorporatedintotheobjectlanguage.Thatis,weconsiderthestandardmodelingofknowledgeamongasetofagentsbymulti-modalKripkestructures.Onemightwanttoconsideractionsthattakeplace,suchasannouncementstogroupsprivately,announcementswithsuspiciousoutsiders,etc.Inoursystem,suchactionscorrespondtoadditionalmodalitiesintheobjectlanguage.Thatis,wedonotaddmachineryontopofmodels(asinFaginetal[4]),butwereifyaspectsofthemachineryinthelogicallanguage.
SpecialcasesofourlogichavebeenconsideredinPlaza[13],Gerbrandy[5,6],andGerbrandyandGroene-veld[7].ThelattergroupofpapersintroducealanguageinwhichonecanfaithfullyrepresentallofthereasoninginexamplessuchastheMuddyChildrenscenario.Inthatpaperwefindoperatorsforupdatingworldsviaannouncementstogroupsofagentswhoareisolatedfromallothers.Weadvancethisbyconsideringmanymoreactions,andbyusingamoregeneralsemantics.
Ourlogiccontainstheinfinitaryoperatorsusedinthestandardmodelingofcommonknowledge.Wepresentasoundandcompletelogicalsystemforthelogic,andwestudyitsexpressivepower.
1991MathematicsSubjectClassification:03B45,68T27,68T301991ACMComputingClassificationSystem:F.4.1,I.2.4,I.2.11
KeywordsandPhrases:epistemicactions,epistemiclogic,knowledgeupdate,publicannouncements
Note:WethankJelleGerbrandyandRohitParikhforusefulconversationsonthiswork.Anearlierversionofthispaperwaspresentedatthe1998ConferenceonTheoreticalAspectsofRationalityandKnowledge.
2
TableofContents
1
Introduction:ExampleScenariosandTheirRepresentations.....................1.1Models.........................................................1.2EpistemicActions.................................................1.3TheIssues.......................................................1.4FurtherContentsofThisPaper......................................2ALogicalLanguagewithEpistemicActions..................................
2.1Syntax.........................................................2.2Semantics.......................................................2.3MoreonActions..................................................3ALogicforL([α])......................................................4ALogicforL([α],2∗)...................................................5CompletenessforL([α],2∗)...............................................
5.1SomeSyntacticResults............................................5.2Completeness....................................................5.3TwoExtensions..................................................6ResultsonExpressivePower..............................................
6.1AnnouncementsaddExpressivePowertoModalLogicwith2∗.............6.2PrivateAnnouncementsAddExpressivePower..........................7ConclusionsandHistoricalRemarks........................................8Appendix:thelexicographicpathorder.....................................References.................................................................35810111111111315171919212424262728293132
References
1.Introduction:ExampleScenariosandTheirRepresentations
3
1.Introduction:ExampleScenariosandTheirRepresentations
Weintroducetheissuesinthispaperbypresentingafewepistemicscenarios.TheseareallbasedontheMuddyChildrenscenario,well-knownfromtheliteratureonknowledge.Theintentionistoexposetheproblemsthatwewishtoaddress.Theseproblemsarefirstofalltogetmodelswhicharefaithfultoourintuitions,andthentobuildandstudylogicalsystemswhichcapturesomeofwhatisgoingoninthescenarios.
Thecastofcharactersconsistsofthreechildren:A,B,andC.Sothatwecanusepronounsfortheminthesequel,weassumethatAismale,andBandCarefemale.Furthermore,AandBaredirty,andCisclean.Eachofthechildrencanseeallandonlytheothers.Itisknowntoall(say,asaresultofashoutfromoneoftheparents)thatatleastonechildisdirty.Furthermore,eachchildmusttrytofigureouthisorherstateonlybystating“IknowwhetherI’mdirtyornot”or“Idon’tknowwhetherI’mdirtyornot.”Theymusttellthetruth,andtheyareperfectreasonersinthesensethattheyknowallofthesemanticconsequencesoftheirknowledge.Theopeningsituationandtheserulesareallassumedtobecommonknowledge.
Scenario1.Afterreflection,AandBannouncetoeveryonethatatthatpointtheydonotknowwhethertheyaredirtyornot.(ThereasonwearehavingAandBmakethisannouncementratherthanallthreechildrenisthatitfitsinbetterwithourscenariostofollow.)Letαdenotethisannouncement.
AsintheclassicalMuddyChildren,thereareintuitionsaboutknowledgebeforeandafterα.Herearesomeofthoseintuitions.Beforeα,nobodyshouldknowthatheorsheisdirty.However,AshouldthinkthatitispossiblethatBknows.(ForifAwereclean,Bwouldinferthatshemustthedirtyone.)Afterα,AandBshouldeachknowthattheyaredirty,andhencetheyknowwhethertheyaredirtyornot.Ontheotherhand,Cshouldnotknowwhethersheisdirtyornot.
Scenario1.5.Thisscenariobeginsafterα.Atthispoint,AandBannouncetoallthreethattheydoknowwhetherornottheyaredirty.We’llcallthiseventα′.Ourintuitionisthatafterα′,Cshouldknowthatsheisnotdirty.Moreover,AandBshouldknowthatCknowsthis.Actually,thedirty-or-notstatesofallthechildrenshouldbecommonknowledgetoallthree.
Scenario2.Asanalternativetothefirstscenario,let’sassumethatCfallsasleepforaminute.Duringthistime,AandBgottogetherandtoldeachotherthattheydidn’tknowwhethertheyweredirtyornot.Letβdenotethisevent.Afterβ,Cwakesup.PartofwhatwemeanbyβisthatCdoesnotevenconsideritpossiblethatβoccurred,andthatit’scommonknowledgetoAandBthatthisisthecase.Thenourintuitionsarethatafterβ,Cshould“know”(actually:believe)thatAdoesnotknowwhetherheisdirty(andsimilarlyforB);andthisfactaboutCiscommonknowledgeforallthreechildren.Ofcourse,itshouldalsobecommonknowledgetoAandBthattheyaredirty.Scenario2.5.FollowingScenario2,weagainhaveα′:AandBannouncethattheydoknowwhethertheyaredirtyornot.Ourintuitionsarenotentirelyclearatthispoint.SurelyCshouldsuspectsomekindofcheatingormiscalculationonthepartoftheothers.However,wewillnothavemuchtosayabouttheworkingsofthiskindofreal-worldsensibility.Ourgoalwillbemoreinthedirectionofmodelingdifferentalternatives.
Scenario3.NowwevaryScenario2.CmerelyfeignedsleepandthoughtsheheardbothAandBwhispering.Ccannotbesureofthis,however,andalsoentertainsthepossibilitythatnothingwascommunicated.(Inreality,AandBdidcommunicate.)AandBfortheirpart,stillbelievethatCwassleeping.Wecallthiseventγ.
4
OnemightatfirstglancethinkthatAandB’s“knowledge”ofC’sepistemicstateisunchangedbyγ.Afterall,thecommunicationwasnotaboutC.However,weworkwithasemanticnotionofknowledge,andafterγ,AandBknowthattheyaredirty,hencethenknowthatCknowsthattheyaredirty.AandBdidnotknowthisattheoutset.
Soweneedtorevisetheinitialintuition.WhatiscorrectisthatifCknowssomefactϕbeforeγ,thenafterγ,AandBknow(orrather,believe)thatCknowsϕ.Thisisbecauseafterγ,AandBnotonlyknowtheclean-or-dirtystateofeveryone,they(therefore)alsoknowexactlywhichpossibilitieseveryoneisawareof,whichtheydiscardasimpossible,etc.SoeachofthemcanreconstructC’sentireepistemicstate.Theybelievethattheirreconstructioniscurrent,butofcourse,whattheyreconstructisC’soriginalone,beforeγ.
Conversely,ifafterγ,AandB“know”thatCknowsϕ,thenbeforeγ,Creallydidknowϕ.Thatis,thereconstructionisaccurate.Forexample,afterγ,AbelievesthatCshouldnotconsideritpossiblethatAknowsthatheisdirty.However,CthinksitispossiblethatAknowsheisdirty.Thereisastrongerstatementthatistrue:Cknowsϕbeforeγiffafterγ,itiscommonknowledgetoAandBthateachofthemknowsthatCknowsϕ.Intuitively,thisholdbecauseeachofAandBknowsthatbothofthemareabletocarryoutthereconstructionofC’sstate.
Ourfinalintuitionisthatafterγ,CshouldknowthatifAweretosubsequentlyannouncethatheknowsthatheisdirty,thenCwouldknowthatBknowsthatsheisdirty.
Scenario3.5.Again,continueScenario3byα′.Atthispoint,Cshouldknowthathersuspicionswereconfirmed,andhencethatsheisnotdirty.Fortheirpart,AandBshouldthinkthatCisconfusedbyα′:theyshouldthinkthatCisasshewasfollowingScenario2.5.
Scenario4.AandBareononesideofthetableandCisontheother,dozing.CwakesupatwhatlookstoherlikethemiddleofajointconfessionbyAandB.Thetwosidesstareeachotherdown.Infact,AandBhavealreadycommunicated.Wecallthisactionδ.SoCsuspectsthatδiswhathappened,butcan’ttellifitwasδornothing.Fortheirpart,AandBseethatCsuspectsbutdoesnotknowthatδhappened.
Thebasicintuitionisthatafterδ,itshouldbecommonknowledgetoallthreethatCsuspectsthatthecommunicationhappened.EvenifCthinksthatAandBdidnotcommunicate,Cshouldnotthinkthatsheissureofthis.
Onerelatedintuitionisthatafterδ,itshouldbecommonknowledgethatCsuspectsthatAknowsthatheisdirty.Asithappens,thisintuitioniswrong.Hereisadetailedanalysis:Cthinksitpossiblethateveryoneisdirtyattheoutset,andifthiswerethecasethentheannouncementofB’signorancewouldnothelpAtolearnthatheisdirty;fromA’spointofview,hestillcouldbecleanandBwouldnotknowthatsheisdirty.C’sviewonthisdoesnotchangeasaresultofδ,soafterwards,CstillthinksthatitcouldbethecasethatAsays,“It’spossiblethatBandCarethedirtyoneandIamclean,HenceCwouldseemycleanfaceandnotsuspectthatIknowthatIamdirty.”SoitcertainlyshouldnotbecommonknowledgethatCsuspectsthatAknowsheisdirty.
NoticealsothatCwouldsayafterδ:“Ithinkitispossiblethatnoannouncementoccurred,andyetAthinksitpossiblethatBistheonlydirtyone.Inthatcase,whatAwouldthinkthatIsuspectthatAtoldBthatheknowsthatheisnotdirty.Ofcourse,thisisnotwhatIactuallysuspect.”ThepointisthatC’sreasoningaboutAandB’sreasoningaboutherinvolvessuspicionofadifferentannouncementthanweatfirstconsidered.
Scenario4.5.Onceagain,wecontinuewithα′.OurintuitionisthatthisistantamounttoanadmissionofprivatecommunicationbyAandB.Ifwedisregardthisandonlylookathigherorderknowledgeconcerningwhoisandisnotdirty,weexpectthattheepistemicstateafterα′isthesameforallthreechildrenasitisattheendofScenario1.5.
1.Introduction:ExampleScenariosandTheirRepresentations
5
1.1Models
Nowthatwehavedetailedafewscenariosandourintuitionsaboutthem,itistimetoconstructsomeKripkemodelsasrepresentationsforthem.
TheModelsUandV.Webeginwitharepresentationofthesituationbeforeα.WetaketheKripkemodelUwhoseworldsareu1,...,u7andwhosestructureisgiveninthetableontheleftbelow:
World
A
••••
C
u1,u5u2,u6u3,u7u4
u1,u5u2,u6u3,u7
B→
u1
u2,u3u2,u3u4,u5u4,u5u6,u7u6,u7
WorldA
•••
C
v1,v5v3,v7v1,v5v6v3,v7
B→
v1v3v5v6,v7v6,v7
Forexample,inworldu3,Aisclean,butBandCaredirty.Also,theworldswhichAthinksarepossibleareu3andu7.Thatis,AseesthatBandCaredirty,soAinfers√thattheworldiseitheru3oru7.Therestofthestructureisexplainedsimilarly,exceptforthemarknexttou6.Thisspecifiesu6astheactualworldinthemodel,theonewhichcorrespondstoourdescriptionofthemodelbeforeα.NotethatUincorporatessomeoftheconventionsstatedinScenario1.Forexample,ineachworld,eachchildhasacompleteandcorrectassessmentofwhichworldsarepossibleforallthreereasoners.
Eachofourintuitionsaboutknowledgebeforeαturnsintoastatementinthemodallogicofknowledge.ThislogichasatomicsentencesDA,DB,andDCstandingfor“Aisdirty”,etc.;ithasknowledgeoperators2A,2B,and2Calongwiththeusualbooleanconnectives.WearegoingtousethestandardKripkesemanticsformulti-modallogicthroughoutthispaper.Sogivenamodel-worldpair,sayA,a,andsomeagent,sayD,we’llwrite
K,k|=2Dϕiff
D
linK,wehaveK,l|=ϕ.wheneverk→
Thebooleanconnectiveswillbeinterpretedclassically.Wecanthencheckthefollowing:
U,u6|=¬2ADA∧¬2A¬DA∧¬2BDB∧¬2B¬DB∧¬2CDC∧¬2C¬DC
U,u6|=3A2BDB
ThemodelafterαistheKripkemodelV,shownontherightabove.ThewaywegotVfromUwastodiscardtheworldsu2andu4ofU,sinceinUateachofthoseworlds,eitherAorBwouldknowiftheyweredirty.Wealsochangedtheu’stov’stoavoidconfusion,andtostressthefactthatwegetanewmodel.Turningbacktoourintuitions,wecanseethatthefollowingholds:
V,v6|=2ADA∧2BDB∧¬(2CDC∨2¬DC)
TheModelW.Scenario1.5elaboratesScenario1bytheeventα′.SowediscardtheworldswherethisisfalseinV,andweobtainaone-worldmodelW:
World
A
•
C
w6
B→
w6
(Wehaverenamedv6tow6.)Thismodelreflectsourintuitionthatatthispoint,Cshouldknowthatsheisnotdirty.
6
TheModelX.ThiscorrespondstoScenario2.WestartwithUandseetheeffectoftheprivateannouncementβ.TheresultingmodelXis:
World
A
C
x1,x5x3,x7x1,x5x6
x3,x7
B→
•••
u1
u2,u3u4,u5u6,u7u6,u7
Noticethattheworldsu1,...,u7arealsoworldsinX.WedidnotputanyinformationinthechartaboveforthoseworldssinceitshouldbeexactlythesameasinUabove.Thereasonforhavingthese“oldworlds”inXisthatsinceCwasasleep,theworldsthatCconsiderspossibleafterβshouldbejusttheonesthatwerepossiblebeforeβ.Wecancheckthat
X,x6|=¬2C(2ADA∨2A¬DA).
Letϕbethesentenceabove.Thenalso,X,x6|=2∗{A,B,C}ϕ.Thisisourformalstatementthatitiscommonknowledgeinthegroupofthreechildrenthatϕholds.ThesemanticsofthisisthatforallsequencesD1,...,Dm∈{A,B,C}∗,X,x6|=2D1···2Dmϕ.NotethatwehavenowayofsayinginthemodallanguagethatCsuspectsthatanannouncementhappened;thebestwecandois(roughly)tosaythatCthinksthatsomesentenceψispossibleinthesensethatψholdsinsomepossibleworld.Ofcourse,wehavenowaytosaythatAandBknowthatCwasasleep,either.
C
x6.Inotherwords,therealworldwouldnotbeNoteaswellthatinX,wedonothavex6→
possibleforC.Thisissomeindicationthatsomethingstrangeisgoingoninthismodel.Further,weconsiderthemodelofwhathappensafterAandB’sannouncement.Theninthismodel,noworldswouldbeaccessibleforCfromtheactualworld.Theseanomaliesshouldjustifyourinterestinthemorecomplicatedscenariosandmodelsinvolvingsuspicionsofannouncements.
TheModelobtainedbyannouncingα′inX.Thiswouldbetheone-worldmodelbelow:
World
A
•
C
x∗6
B
→
WehavenotonlydeletedtheworldswhereeitherAorBdoesnotknowthattheyaredirtyinX,butwealsodiscardedallworldsnotreachablefromthenewversionx∗6ofx6.TheanomalyhereisthatCthinksnoworldsarepossible.
TheModelY.WeconsiderγfromScenario3,inwhichCthoughtshemighthaveheardAandB,whileAandBthinkthatCisunawareofγ.WegetthemodelYdisplayedinFigure1below.Yhas24worlds,andsowewon’tjustifyallofthemindividually.WewillgiveamoreprincipledconstructionofYfromWandγ,oncewehavesettleonamathematicalmodelofγ.Fornow,theideasarethattheyworldsarethosewheretheannouncementhappened,andthey′worldsarethoseinwhichitdidnot.Notethatsomeoftheyworldsaremissing,sincethetruthfulannouncementbyAandBpresupposesthattheydon’tknowwhethertheyaredirtyinUatthecorrespondingworld.Thex’sandu’sarefromabove,andtheyinherittheaccessibilityrelationswhichwehaveseen.NowourmainintuitionhereisthatU,u6|=2CϕiffY,y6|=2+{A,B}2Cϕ.(Thesentence+
2{A,B}χmeansthatAknowsϕ,AknowsBknowsχ,etc.Itdiffersfrom2∗{A,B}χinthatitdoesnot
C
u6,u7andnootherworlds.Andtheonlyworldsentailthatχistrue.)Toseethis,notethatu6→
1.Introduction:ExampleScenariosandTheirRepresentations
7
B→
WorldAC
•••••••
x1,x5x3,x7x1,x5x6
x3,x7u1,u5u2,u6u3,u7u4
u1,u5u2,u6u3,u7
′y1,y1
′′y3,y2,y3
′′y5,y4,y5
′′
y6,y7,y6,y7
′′
y6,y7,y6,y7
′y1,y1
′′y3,y2,y3
′′y3,y2,y3
′
y5,y4,y5
′′y5,y4,y5
′′
y6,y7,y6,y7
′′
y6,y7,y6,y7
Figure1:ThemodelY
ABC
reachablefromy6usingoneormore→or→transitionsfollowedbya→transitionareagainu6andu7.
AnotherintuitionisthatinY,y6,CshouldthinkthatitispossiblethatAknowsthatheisdirty.
C
Thisisjustifiedsincey6→x6,andY,x6∼=X,x6(thatis,thesubmodelsofXandYgeneratedbyx6areisomorphic),andX,x6|=2ADA.
OurfinalintuitionisthatinY,y6,CshouldknowthatifAweretosubsequentlyannouncethatheknowsthatheisdirty,thenCwouldknowthatBknowsthatsheisdirty.Tocheckthis,weneed
′
tomodifyYbydeletingtheworldswhereAdoesnotknowthatheisdirty.Theseincludey7,y6and′
y7.Intheupdatedmodel,theonlyworldaccessibleforCfrom(thenewversionof)y6isy6itself,andaty6inthenewstructure,Bcorrectlyknowssheisdirty.
TheModelobtainedbyannouncingα′inY.Aswhenα′isannouncedinX,weonlykeepthe
′′
worldsofYworldswherebothAorBdoknowtheyaredirty.Sowedropy7,y6,andy7.
World
A
•
•
C
x#6x#6
B
→
∗y6
Wealsoonlykeeptheworldsaccessiblefromy6(thischangeisharmless).Cknowssheisnotdirty.Technically,AandB“know”this,butthisisforthenonsensicalreasonthatthey“know”thatCknowseverything.
8
TheModelZ.CorrespondingtoScenario4,wegetthemodelZshownbelow.
w
A
••••
C
z1,z5z2z3,z7z4z1,z5z6z3,z7
B→
w
′z1,z1
′′
z2,z3,z2,z3
′′
z2,z3,z2,z3
′′
z4,z5,z4,z5
′′
z4,z5,z4,z5
′′
z6,z7,z6,z7
′′
z6,z7,z6,z7
A
•
•••
C
′′z1,z5′′z2,z6′′z3,z7′′z4,z6′′z1,z5′′z2,z6′′z3,z7
B
→
′z1,z1
′′
z2,z3,z2,z3
′′
z2,z3,z2,z3
′′
z4,z5,z4,z5
′′
z4,z5,z4,z5
′′
z6,z7,z6,z7
′′
z6,z7,z6,z7
RecallourlastpointinScenario4,thatweneedtoconsiderafewpossibleannouncementsforCto
suspect.Thisisreflectedinthefactthatthezworldsareofthreetypes.Inz2,Bannouncedthatsheknowswhethersheisdirty,andAannouncedthathedoesn’t.Similarremarksapplytoz4.Inallotherzworlds,bothannouncedthattheydonotknow.Theworldsaccessiblefromeachoftheseisbasedontherelevantannouncement.Forexample,inz2,neitherAnorBthinksanyotherworld
A
z6.Butinz6,Bcouldnotannouncethatsheknowssheisispossible.(Onemightthinkthatz2→
dirty.Soiftheworldwerez2andtherelevantannouncementmade,thenAwouldnotthinkz6ispossible.)Thez′worldsarethoseinwhichnoannouncementactuallyhappened.
OurkeyintuitionwasthatitiscommonknowledgethatCsuspectsthatδhappened.ThiswillnotcorrespondtoanythingintheformallanguageL([α],2∗)introducedlaterinthispaper.(However,itwillberepresentableinanauxiliarylanguageaboutactions;seeExample2.3.)Informally,the
′C
intuitionisvalidforZbecauseforeveryzi(orzi)thereissomezj(unprimed)suchthatzi→zj
′C
(orzi→zj).Inaddition,inthisparticularmodelthereisasentenceinourformallanguagewhichhappenstoholdonlyattheworldswhereanannouncementoccurred.Hereisone:
χ
≡
2ADA∨2A¬DA∨3∗{A,B}3C2ADA
SoZ,z6|=2∗{A,B,C}3Cχ.
AC
z3,and3A2ADAfailsinz7→TheexplanationofthemistakenintuitioninScenario4isthatz6→
′′
z2,z3,z2,andz3..Overall,Z,z6|=¬2C2A3A2ADA.
ThepointthatC’ssuspicionvariescorrespondstothefactthat3C3A3C2A¬DAholdsatZ,z6.
′A′CC
Indeedz6→z6→z2→z2,andZ,z2|=2A¬DA.
AfewmoreinvolvedstatementsaretrueinZ.Forexample,2{A,B,C}∗(2ADA→3C2ADA).ItiscommonknowledgetoallthreethatifAknowsheisdirty,thenCthinksitpossiblethatAknowsthis.
TheModelobtainedbyannouncingα′inZ.ThismodelisWfromabove.(Actually,itisbisimilartoW;seeSection2.2.)ThiscorrespondstotheintuitionthatScenarios2.5and4.5leadtothesamemodel.
1.2EpistemicActions
WewillformalizealanguageinSection2alongwiththenotionsof(epistemic)actionstructureandactions.Beforewedothat,itmakessensetopresenttheideainformallybasedontheexampleswhichwehavealreadydealtwith.
αandα′:announcementstoeveryone.WefirstconsiderαofScenario1.Letψbegivenby
ψ
:=
¬(2ADA∨2A¬DA)∧¬(2BDB∨2B¬DB)
(1.1)
1.Introduction:ExampleScenariosandTheirRepresentations
9
SoψsaysthatneitherAnorBknowwhetherornottheyaredirty.Thisisthepreconditionoftheannouncement,butitisnotthestructure.Thestructureofthisannouncementisquitesimple(somuchsothatthereaderwillneedtoreadfurthertogetanideaforwhatwemeanbystructure).Itis
D
kforallD∈{A,B,C}.thefollowingKripkestructureK:wetakeonepoint,callitk,andwesetk→
WecallK,kanactionstructure.AlongwithK,wealsohaveaprecondition;thiswillbeψfrom(1.1).Todealwithactionstructureswithmorethanonepoint,thepreconditionwillbeafunctionprefromworldstosentences.Inthiscase,thefunctionpreisjust{k,ψ}.ThetupleK,k,prewillbeanexampleofwhatwecallanaction.Thisparticularactionisourmodeloftheannouncementα.Henceforthweusethesymbolαtoreferambiguouslytothepretheoreticnotionoftheannouncementeventandtoourmathematicalmodelofit.
Anotherexampleofanannouncementtoeveryoneisα′.Herewejustchangeψfrom(1.1)tothesentenceψ′whichsaysthatbothAandBknowwhetherornottheyaredirty.Yetanotherexampleisthenullannouncement.Thismodelstheannouncementofatautologytruetoeveryone.We’llwritethisasτ.
β:asecureannouncementtoasetofagents.Next,supposewehaveanannouncementmadetosomepossiblypropersubsetB⊆AinthemannerofScenario2.Sothereissomedisputeastowhathappened:theagentsinBthinkthattherewasanannouncement,whilethoseoutofBaresurethatnothinghappened.WeformalizethiswithaKripkestructureoftwopoints,landt.
DDD
tforallD.ThepointisthatlhereisthetforD∈/B,andt→lforallD∈B,l→Wesetl→
actualannouncement,andtheagentsinBknowthatthisistheannouncement.TheagentsnotinBthinkthattisforsuretheonlypossibleaction,andtinthismodelwillbehavejustlikethenullannouncement.Thepreconditionfunctionwillbecalledpreinallofourexamples.Herepreisgivenbypre(l)=ψandpre(t)=true,whereψisfrom(1.1).TheactionoverallisL,l,pre,whereL={l,t}.Wecallthisactionβ.
γ:anannouncementwithasuspiciousoutsider.structurehasfourpoints,asfollows:
World
A→
ThisisbasedonScenario3.Theassociated
C→
ltltψtrueψtrue
Theideaisthatmisthe(private)announcementthatCsuspects,andnisotherannouncementthatCthinksispossible(wherenothingwascommunicatedbyAandB).Thenifmhappened,AandBweresurethatwhathappenedwasl;similarly,ifnhappened,AandBwouldthinkthattwaswhathappened.Wecallthisactionγ;technicallyitis{m,n,l,t},m,pre.Wegetadifferentaction,sayγ′ifweusethesamemodelasabovebutchangethedesignated(“real”)worldfrommton.δ:anannouncementwithcommonknowledgeofsuspicion.CorrespondingtoScenario4,wehavethefollowingmodel.InitψAdenotesthesentencesayingthatAknowswhetherheisdirtybutBdoesnot,ψBthesentencesayingthatBknowswhethersheisdirtybutAdoesnot,andψ∅thesentencestatingthatneitherknows.
World
A→
opqrs
C→
ψψAψBψ∅true
10
Wecallthisactionδ.Therearefivepossibleactionshere,dependingonwhetheritwasψ,ψA,ψB,ψ∅ornothingwhichwasannounced.Ineachcase,AandBaresureofwhathappened.Evenifnothingactuallyhappened(s),Cwouldsuspectoneoftheotherfourpossibilities.Inthose,Cstillconsidersitpossiblethatnothinghappened.
Stilltocome.Thereaderisperhapswonderingwhattheactualconnectionisbetweenthe(formal)actionsjustintroducedandtheconcretemodelsoftheprevioussection.Theconnectionisthatthereisawayoftakingamodelandanactionandproducinganothermodel.WhenappliedtothespecificmodelUandtheactionsofthissection,wegetthemodelsV,...,Z.WedelaythisconnectionuntilSection2.2below,sinceitishightimethatweintroduceourlanguageofepistemicactionsanditssemantics.Thepointisthatthereisaprincipledreasonbehindthemodels.
Thequestionalsoarisesastowhetherthereareanyprinciplesbehindtheparticularactionswhichweputdowninthissection.Asithappens,thereismorewhichcanbesaidonthismatter.WepostponethatdiscussionuntilSection2.3,afterwehaveformallydefinedthesyntaxandsemanticsofourlogicallanguages.
1.3TheIssues
Themainissueweaddressinthispaperistoformallyrepresentepistemicupdates,i.e.,changesintheinformationstatesofagentsinadistributedsystem.Wethinkofthesechangesasbeinginducedbyspecificinformation-updatingactions,whichcanbeofvarioustypes:(1)information-gatheringandprocessing(e.g.,realizingthepossibilityofotheragents’hiddenactions,andmoregenerally,learningofanykindofnewpossibilityviaexperiment,computation,orintrospection);(2)information-exchangeandcommunication(learningbysending/receivingmessages,publicannouncements,secretinterceptionofmessages,etc.);(3)information-hiding(lyingorotherformsofdeceivingactions,suchascommunicationoversecretchannels,sendingencryptedmessages,holdingsecretsuspicions);(4)information-lossandmisinformation(beingliedto,startingtohavegratuitoussuspicions,non-introspectivelearning,wrongcomputationsorfaultyobservations,paranoia);(5)andmoregenerallysequentialorsynchronouscombinationsofalloftheabove.
Specialcasesofourlogic,dealingonlywithpublicorsemi-publicannouncementstomutuallyisolatedgroups,havebeenconsideredinPlaza[13],Gerbrandy[5,6],andGerbrandyandGroeneveld[7].ThesedealwithactionssuchasαandβinourIntroduction.Ourexamplesγandδgobeyondwhatispossibleinthesettingofthesepapers.Butouroverallsettingismuchmoreliberalsetting,sinceitallowsforalltheabove-mentionedtypesofactions.Wefeelitwouldbeinterestingtostudyfurtherexampleswithaneyetowardsapplications,butweleavethistootherpapers.
Inourformalsystem,wecaptureonlytheepistemicaspectoftheserealactions,disregardingother(intentional)aspects.Inparticular,forsimplicityreasons,weonlydealwith“‘purelyepistemic”actions;i.e.,theonesthatdonotchangethefactsoftheworld,butaffectonlytheagents’beliefsabouttheworld.However,thisisnotanessentiallimitation,asourformalsettingcanbeeasilyadaptedtoexpressfact-changingactions(seetheendofSection2.3andalsoSection5.3).
Onthesemanticalside,themainoriginaltechnicalcontributionofourpaperliesinourdecisiontorepresentnotonlytheepistemicstates,butalsotheepistemicactions,byKripkestructures.Whileforstates,thesestructuresrepresentintheusualwaytheuncertaintyofeachagentconcerningthecurrentstateofthesystem,wesimilarlyuseaction-structurestorepresenttheuncertaintyofeachagentconcerningthecurrentactiontakingplace.Theintuitionisthatwearedealingwithpotentially”half-opaque/half-transparent”actions,aboutwhichtheagentsmaybeincompletelyinformed,orevencompletelymisinformed.Besidesthestructure,actionshavepreconditions,definingtheirdomainofapplicability:noteveryactionispossibleineverystate.Wemodeltheupdateofastatebyanactionasapartialupdateoperation,givenbyarestrictedproductofthetwostructures:theuncertaintiespresentinthegivenstateandthegivenactionaremultiplied,whilethe“impossible”combinationsofstatesandactionsareeliminated(bytestingtheactions’preconditionsonthestate).Theunderlyingintuitionisthattheagent’suncertaintiesconcerningthestateandtheonesconcerningtheactionare
2.ALogicalLanguagewithEpistemicActions
11
mutuallyindependent,exceptfortheconsistencyoftheactionwiththestate.
Onthesyntacticalside,weuseamixtureofdynamicandepistemiclogic,withdynamicmodalitiesassociatedtoeachaction-structure,andwithcommon-knowledgemodalitiesforvariousgroupsofagents(inadditiontotheusualindividual-knowledgeoperators).Wegiveacompleteanddecidableaxiomatizationforthislogic,andweprovevariousexpressivityresults.Fromaproof-theoreticalpointofview,themainoriginalityofoursystemisthepresenceofourActionRule,aninferencerulecapturingwhatmightbecalledanotionof“epistemic(co)recursion”.WeunderstandthisruleandourKnowledge-ActionAxiom(ageneralizationofRamsey’saxiomtohalf-opaqueactions)asexpressingfundamentalformalfeaturesoftheinteractionbetweenactionandknowledgeinmulti-agentsystems,featuresthatwethinkhavenotbeenformallyexpressedbefore.
1.4FurtherContentsofThisPaper
Section2givesourbasiclogicL([α])ofepistemicactionsandknowledge.Theideaistodefinethelogictogetherwiththeactionstructureswhichwehavejustlookedatinformally.SoinL([α])wefinallywillpresentthepromisedformalversionsoftheannouncementsofSection1.2.InSection3wepresentasoundandcompleteaxiomatizationofL([α]).WeaddthecommonknowledgeoperatorstogetL([α],2∗)inSection4.CompletenessforthislogicisprovedinSection5.TworesultsontheexpressivepowerarepresentedinSection6.AnAppendixcontainssometechnicalresultswhich,whileneededforourwork,seemtointerrupttheflowofthepaper.
2.ALogicalLanguagewithEpistemicActions2.1Syntax
WebeginwithasetAtSenofatomicsentences,andwedefinetwosetssimultaneously:thelanguageL([α]),andasetofactions(overL([α])).
L([α])isthesmallestcollectionwhichincludesAtSenandwhichisclosedunder¬,∧,2AforA∈A,and[α]ϕ,whereαisanactionoverL([α]),andϕ∈L([α]).
Anactionstructure(overL([α]))isapairK,pre,whereKisafiniteKripkeframeoverthesetAofagents,andpreisamappre:K→L.WewillusuallywriteKfortheactionstructureK,pre.Anaction(overL([α]))isatupleα=K,k,pre,whereK,preisanactionstructure
D
overL([α]),andk∈K.Eachactionαthusisafinitesetwithrelations→forD∈A,togetherwithapreconditionfunctionandaspecifiedactualworld.
TheactionsthemselvesconstituteaKripkeframeActionsinthenaturalway,bysetting
Whenα=K,k,pre,wesetpre(α)=pre(k).Thatis,pre(α)isthepreconditionassociatedtothedistinguishedworldoftheaction.Forthisreason,weoftenwritepre(α)insteadofpre(k).Examples2.1AllofthesentencesmentionedinSection1.1aresentencesofL([α]),exceptfortheonesthatuse2∗{A,B,C}.Thisconstructgivesusamoreexpressivelanguage,asweshallsee.Thestructuresα,τ,β,γ,γ′,δ,andδ′describedinformallyinSection1.2arebonafideactions.As
D
αexamplesoftheaccessibilityrelationontheclassofactions,wehavethefollowingfacts:α→
DBCCABCABAB
andτ→τforallD∈{A,B,C};β→β;β→β;β→τ;γ−→β;γ,γ′→γ,γ′;γ′−→τ;δ−→δ,′AB′′C′δ−→δ,andδ,δ→δ,δ.
Manyothertypesofexamplesarepossible.Wecanrepresentmisleadingepistemicactions,e.g.lying,ormoregenerallyactingsuchthatsomepeopledonotsuspectthatyouractionispossible.Wecanalsorepresentgratuitoussuspicion(“paranoia”):maybeno“real”actionhastakenplace,exceptthatsomepeoplestartsuspectingsomeaction(e.g.,someprivatecommunication)hastakenplace.2.2Semantics
Aswiththesyntax,wedefinetwothingssimultaneously:thesemanticrelationW,w|=ϕ,andapartialoperation(W,w,α)→W,wα.Beforethis,weneedanotherdefinition.GivenamodelWandanactionstructureK,wedefinethemodelWKasfollows:
D
L,l,pre′iffK,k,pre→
D
linK.K=L,pre=pre′,andk→
(2.1)
12
1.TheworldsofWKarethepairs(w,k)∈W×KsuchthatW,w|=pre(k).2.Forsuchpairs,
A
(w′,k′)(w,k)→
AA
k′.w′andk→iffw→
(2.2)
3.WeinterprettheatomicsentencesbysettingvWK((w,k))=vW(w).Thatis,pistrueat(w,k)
inWKiffpistrueatwinW.Givenanactionα=K,kandamodel-worldpairW,w,wesaythatW,wαisdefinediffW,w|=pre(k),andinthatcasewesetW,wα=W,wK,k=WK,(w,k).Onecannowcheckthatthefollowingholdsforthesedefinitions.
A
W,wα→W,xβ
iff
AA
W,wαandW,xβaredefined,w→xinW,andα→β.
Thesemanticsisgivenbyextendingtheusualclausesformodallogicbyoneforactions:
W,w|=[α]ϕ
iffW,wαisdefinedimpliesW,wα|=ϕ.
W,wαisdefinedandW,wα|=ϕ.
Asiscustomary,weabbreviate¬[α]¬ϕbyαϕ.Thenwehave
W,w|=αϕ
iff
Wealsoabbreviatethebooleanconnectivesclassically,andwelettruedenotesometautologysuchas
p∨¬p.
ThelargerlanguageL([α],2∗)WealsoconsideralargerlanguageL([α],2∗).Thisisdefinedbyaddingoperators2∗BforallsubsetsB⊆A.(Whenwedothis,ofcoursewegetmoreactionsaswell.)Thesemanticsworksbytaking2∗Bϕtoabbreviatetheinfinitaryconjunction
2A1···2Anϕ.
A1,...,An∈B∗
HereB∗isthesetofallsequencesfromB.Thisincludestheemptysequence,so2∗Bϕlogicallyimpliesϕ.
BisimulationGiventwomodels,sayKandL,overthesamesetofAofagents,abisimulationbetweenKandLisarelationR⊆K×LsuchthatifkRlandA∈A,then:1.Forallatomicp,K,k|=piffL,l|=p.
AA
l′suchthatk′Rl′.k′thereissomel→2.Forallk→
AA
3.Foralll→l′thereissomek→k′suchthatk′Rl′.
Giventwomodel-worldpairsK,kandL,l,wewriteK,k≡L,liffthereissomebisimulationRsuchthatkRl.ItisastandardfactthatifK,k≡L,l,thenthetwopairsagreeonallsentencesofstandardmodallogic.Inoursetting,wealsocanspeakaboutactionsbeingbisimilar:wechangecondition(1)abovetorefertosaythatpre(k)=pre(l).Itiseasynowtochecktwothingssimultaneously:(1)bisimilarpairsagreeonallsentencesofL([α]);and(2)ifK,k≡L,landα≡β,thenK,kα≡L,lβ.Furthermore,theseresultsextendtoL([α],2∗).
Examples2.2WelookbackatSection1.1forsomeexamples.Weuse∼=todenotetherelationofisomorphismonmodel-worldpairs.Itisnothardtocheckthefollowing:U,u6α∼=V,v6,
δ∼γ∼β∼U,u6=X,x6,U,u6=Y,y6,andU,u6=Z,z6.Forexample,theisomorphismwhich
2.ALogicalLanguagewithEpistemicActions
13
′
showsthatU,u6δ∼=Z,z6is(ui,o)→zifori=2,4,(u2,q)→z2,(u4,p)→z4,and(ui,r)→ziforalli.
Letα′betheactionofannouncingtoallagentsthatbothAandBdoknowwhethertheyare
′′
dirty.ThenV,v6α∼=X,x6.Moreover,Z,v6α≡X,x6.Notethatinthiscaseweonlyhavebisimilarity.However,weknowthatourlanguageswillnotdiscriminatebetweenbisimilarpairs,sowecanregardthemasthesame.ThismodelsourintuitionthattheepistemicstatesattheendofScenarios1.5and4.5shouldbethesame.
Finally,allofthesemanticfactsaboutthevariousmodelsinSection1.1nowturnintoprecisestate-ments.Forexample,U,u6|=[α]3A2BDB.Also,U,u6|=[α][α′]2∗A,B,C2CDC.ThisformalizesourintuitionthatifwestartwithU,u6,firstannouncethateachofAandBdonotknowtheirstate,thensecondannouncethattheyeachdoknowit,thenatthatpointitwillbecommonknowledgetoallthreethatCknowssheisdirty.
2.3MoreonActions
Inthissection,wehaveafewremarksonactions.ThepointhereistoclarifytherelationbetweenthescenariosofSection1andtheintuitionsconcerningthem,andthecorrespondingactionsofSection1.2.Firstandforemost,herearethetheconceptualpointsinvolvedinourformalization.Theideaisthatepistemicactionspresentalotofuncertainty.Indeed,whatmightbethoughtofasasingleaction(orevent)isnaturallyinterpretedbyagentsindifferentways.Thevariousagentsmightbeunclearonwhatexactlyhappened,andagaintheymightwellhavedifferentinterpretationsonwhatishappening.OurformalizationreflectsthisbymakingepistemicactionsintoKripkemodels.Soouruseofpossible-worldsmodelingofactionsisonaparwithotherusesofthesemodels,anditinheritsallofthefeaturesandbugsofthoseapproaches.
Next,wewanttospelloutinwordswhatourproposalamountsto.ThebasicproblemistodecidehowtorepresentwhathappenstoaKripkemodelWafteranannouncementα.(Ofcourse,wearemodelingαbyanactioninourformalsense.)OursolutionbeginsbyconsideringcopiesofW,oneforeachactiontokenkofαinwhichpre(α)holds.WecanthinkoftaggingtheworldsofWwiththeworldsofα,andthenwemustgiveanaccountoftheaccessibilityrelationbetweenthem.Theintuitionisthattheagents’relationstoalternativeworldsshouldbeindependentfromtheirrelationstootherpossibilitiesforα.SotheaccessibilityrelationsofKandWshouldbecombinedindependently.Thisisexpressedformallyin(2.2).
ˆhasasatomicsentencesallsentencesϕofL([α],2∗).IthasallbooleanTheauxiliarylanguageL
connectives,standardmodaloperators2AforA∈A,andalsogroupknowledgeoperators2∗BforB⊆A.
ˆonactionsusingthestandardclausesfortheconnectivesandmodaloperators,andWeinterpretL
byinterpretingtheatomicsentencesasfollowsK,k|=piffpre(k)=p.
Examples2.3Theideahereisthattheauxiliarylanguageformalizestalkaboutwhatthedifferentagentsthinkishappeninginourannouncements.WereferbacktotheactionsofSection1.2.Forexample,α|=2∗{A,B,C}ψ.Intuitively,inα,itiscommonknowledgethatψwasannounced.Anotherexample:that
δ|=2∗{A,B,C}3C(ψ∨ψA∨ψB).
Thatis,inδ,itiscommonknowledgethatCthinksitispossiblethatsomenon-trivialannouncement
happened.Recallthatthiswasoneofourbasicintuitionsaboutδ,onewhichisnotingeneralstatableinourmainlanguageL([α],2∗).
ˆ.ThenχcharacterizesDefinitionLetK,kbeamodel-worldpair,andletϕbeasentenceofL
K,kiffforallL,l,L,l|=χiffL,l≡K,k.
14
ˆProposition2.4LetK,kbeamodel-worldpairwithKfinite.ThenthereisasentenceχofL
whichcharacterizesK,k.
ProofByreplacingK,kbyitsquotientunderthelargestauto-bisimulation,wemayassumethatifl=m,thenK,l≡K,m.Itiswell-knownthattherelationofelementaryequivalenceinmodallogicisabisimulationonmodelsinwhicheachworldhasfinitelymanyarrowscominginandout.ItfollowsfromthisandtheoverallfinitenessofKthatwecanfindsentencesϕlforl∈Kwiththepropertythatforalllandm,K,m|=ϕliffm=l.Letψbethefollowingsentence
ψ≡3Aϕl′ϕl′∧ϕl→2A
AAl∈K,A∈Al→l′l→l′GoingbacktoouroriginalK,k,letχbeϕk∧2∗Aψ.ItiseasytocheckthateachK,lsatisfiesψ;
∗
henceeachsatisfies2Aψ.ThereforeK,k|=χ.WeclaimthatχcharacterizesK,k.Toseethis,supposethatJ,j|=χ.ConsidertherelationR⊆K×Jgivenby
k′Rj′
iffJ,j′|=ϕk′∧2∗Aψ.
ItissufficienttoseethatRisabisimulation.We’llverifyhalfofthis:supposethatk′Rj′and
AAj′→j′′.Byusingψ,weseethatthereissomek′′suchthatk′→k′′andj′′|=ϕk′′.Andalso,since
∗′′
|=2∗=2∗⊣Aψ→2A2Aψ,weseethatJ,j|Aψ.Thiscompletestheproof.
Theconnectionofthisresultandourdiscussionofactionsisthatitisoftendifficulttogofromaninformaldescriptionofananepistemicactiontoaformalonealongourlines.(Forexample,ourformulationofδwasthelastofseveralversions.)Presumably,onewaytogetaformalactioninoursenseistothinkcarefullyaboutwhichpropertiestheactionshouldhave,expressthemintheauxiliarylanguage,andthenwriteacharacterizingsentencesuchasψintheproofofProposition2.4.Thenonecanconstructthefinitemodelbystandardmethods.Althoughthiswouldbeatediousprocess,itseemsworthwhiletoknowthatitisavailable.
Ourformalizationofactionsreflectssomechoiceswhichonemightwishtomodify.Oneofthesechoicesistotaketherangeofthefunctionpretobesomelanguage.Anotheroptionwouldbetohavetherangetobethepowersetofthatlanguage.ThiswouldmakeactionsintoKripkemodelsoverthewholesetofsentences.(Andsowhatwehavedoneislikeconsideringmodallogicwiththerestrictionthatatanyworldsatisfiesexactlyoneatomicsentence.)Takingthisotheroptionthusbringsactionsandmodelscloser.ThisideaispursuedinBaltag[1],acontinuationofthisworkwhichdevelopsa“calculusofepistemicactions.”Thisreplacesthe“semantic”actionsofthispaperwithactionexpressions.Theseexpressionshavenicerpropertiesthantheauxiliarylanguageofthispaper,butitwouldtakeustoofarafieldtodiscussthisfurther.
Onadifferentmatter,itmakessensetorestrictattentionfromthefullcollectionofactionsaswe
A
havedefinedittothesmallercollectionofS5actions,whereeachaccessibility→isanequivalencerelation.Thiscorrespondstothestandardmoveofrestrictingattentiontomodelswiththisproperty,andthereasonsfordoingthisaresimilar.Intuitively,anS5actionisoneinwhicheveryagentisintrospective(withrespecttotheirownsuspicionsaboutactions).Moreover,theintrospectionisaccurate,andthisfactiscommonknowledge.
Afinalmodificationwhichisquitenaturalistoallowactionswhichchangetheworld.Onewoulddothisbyaddingtoournotionofactionasententialupdateu.ThiswouldbeafunctiondefinedonAtSenandwrittenintermsofupdateequationssuchasu(p):=p∧q;u(q)=false,etc.Weareconfidentthatourlogicalsystemscanbemodifiedtoreflectthischange,andwediscussthisatcertainpointsbelow.Wedecidednottomakethischangemostlyinordertokeepthebasicnotionsassimpleaspossible.
Withrespecttobothofthechangesmentionedinthelasttwoparagraphs,itisnothardtomodifyourlogicalworktogetcompletenessresultsforthenewsystems.WediscussallofthisinSection5.3.
3.ALogicforL([α])
15
BasicAxioms
Allsententialvalidities([α]-normality)(2A-normality)∗(2∗C-normality)ActionAxioms
(AtomicPermanence)(PartialFunctionality)(Action-Knowledge)∗MixAxiom
∗CompositionAxiomModalRules(ModusPonens)([α]-necessitation)(2A-necessitation)∗(2∗C-necessitation)∗ActionRule
⊢[α](ϕ→ψ)→([α]ϕ→[α]ψ)
⊢2A(ϕ→ψ)→(2Aϕ→2Aψ)
∗∗
⊢2∗C(ϕ→ψ)→(2Cϕ→2Cψ)
⊢[α]p↔(pre(α)→p)⊢[α]¬χ↔(pre(α)→¬[α]χ)
A
β})⊢[α]2Aϕ↔(pre(α)→{2A[β]ϕ:α→
{2A2∗Cϕ:A∈C}
⊢2∗Cϕ→ϕ∧
⊢[α][β]ϕ↔[α◦β]ϕFrom
FromFromFrom
⊢ϕand⊢ϕ→ψ,infer⊢ψ⊢ψ,infer⊢[α]ψ⊢ϕ,infer⊢2Aϕ⊢ϕ,infer⊢2∗Cϕ
Letψbeasentence,andletCbeasetofagents.Lettherebesentencesχβforallβsuchthat
α→∗Cβ(includingαitself),andsuchthat1.⊢χβ→[β]ψ.
A
γ,then⊢(χβ∧pre(β))→2Aχγ.2.IfA∈Candβ→
Fromtheseassumptions,infer⊢χα→[α]2∗Cψ.
Figure2:ThelogicalsystemforL([α],2∗).ForL([α]),wedropthe∗axiomsandrules.3.ALogicforL([α])
InFigure2belowwepresentalogicforL([α],2∗)whichweshallstudylater.Inthissection,weshallrestrictthelogictothesimplerlanguageL([α]).Wedosopartlytobreakupthestudyofasystemwithmanyaxiomsandrules,andpartlytoemphasizethesignificanceofaddingtheinfinitaryoperators2∗BtoL([α]).Tocarryouttherestriction,weforgettheaxiomsandrulesofinferenceinFigure2whicharemarkedbya∗.Inparticularα◦βwillbedefinedlater(Section4).
Therulesofthesystemareallquitestandardfrommodallogic.TheActionAxiomsaretheinterestingnewones.IntheAtomicPermanenceaxiom,pisanatomicsentence.Theaxiomthensaysthatannouncementsdonotchangethebrutefactofwhetherornotpholds.Thisaxiomreflectsthefactthatouractionsdonotchangeanykindoflocalstate.(WediscussanextensionofoursysteminSection5.3wherethisaxiomisnotsound.)ThePartialFunctionalityAxiomcorrespondstothefactthattheoperationW,w→W,wαisapartialfunction.ThekeyaxiomofthesystemistheAction-KnowledgeAxiom,givingacriterionforknowledgeafteranannouncement.Wewillchecksoundnessofthisaxiomleavingcheckingsoundnessofotherunstarredaxiomsandrulestothereader.
16
Proposition3.1TheAction-KnowledgeAxiom
[α]2Aϕ↔(pre(α)→
issound.
ProofWeremindthereaderthattherelevantdefinitionsandnotationarefoundinSection2.2.LetαbetheactionK,k.FixapairW,w.IfW,w|=¬pre(α),thenbothsidesofourbiconditionalhold.WethereforeassumethatW,w|=pre(α)intherestofthisproof.AssumethatW,wα|=2Aϕ.
AA
k′.Letβ.ThisβisoftheformK,k′forsomek′suchthatk→Takesomeβsuchthatα→
Aw→w′.Wehavetwocases:W,w′|=pre(k′),andW,w′|=¬pre(k′).Inthelattercase,W,w′|=[β]ϕtrivially.We’llshowthisintheformercase,soassumeW,w′|=pre(k′).Then
A
(w′,k′).NowourassumptionthatW,wα|=2Aϕ(w′,k′)isaworldofWK,andindeed(w,k)→
impliesthatWK,(w′,k′)|=ϕ.ThismeansthatW,w′β|=ϕ.HenceW,w′|=[β]ϕ.Sinceβand
w′werearbitrary,W,w|=β2A[β]ϕ.Theotherdirectionissimilar.⊣TherestofthissectionisdevotedtothecompletenessresultforL([α]).Thereadernotinterestedinthismayomittherestofthissection,butatsomepointslaterwewillreferbacktothetermrewritingsystemRwhichweshalldescribeshortly.OurcompletenessproofisbasedonatranslationofL([α])toordinarymodallogicL.AndthistranslationisbasedonatermrewritingsystemtobecalledR.TherewritingrulesofRare:
[α]p[α]¬ψ[α](ψ∧χ)[α]2Aψ
;;;;
pre(α)→ppre(α)→¬[α]ψ[α]ψ∧[α]χ
A
β}pre(α)→{2A[β]ψ:α→
A
β}){2A[β]ϕ:α→
Asinallrewritesystems,weapplytherulesofRatarbitrarysubsentencesofagivensentence.(For
example,considerwhathappenswithsomethinglike[α][β]ϕ.Wemightrewrite[β]ϕusingsomerule,saytoψ.Thenwemightrewrite[α]ψtosomethinglike[γ]ψ,etc.)Lemma3.2Thereisarelation 3.Asentenceϕ∈L([α])isanormalformiffitisamodalsentence(thatis,ϕcannotberewritteniffnoactionsoccurinϕ).Thistakessomework,andbecausethedetailsarelessimportantthanthefactsthemselves,wehaveplacedtheentirematterinanAppendixtothispaper.(TheAppendixalsodiscussesanextensionoftherewritesystemRtoasystemR∗forthelargerlanguageL([α],2∗),soifyoureaditatthispointyouwillneedtokeepthisinmind.) Inthenextresult,weletLbeordinarymodallogicoverAtSen(whereofcoursetherearenoactions). Proposition3.3Thereisatranslationt:L([α])→Lsuchthatforallϕ∈L([α]),ϕissemanticallyequivalenttoϕt.Proof EverysentenceϕofL([α])mayberewrittentoanormalform.ByLemma3.2,thenormal 4.ALogicforL([α],2∗)17 formsofϕisasentenceinL.Wethereforesetϕttobeanynormalformofϕ,saytheoneobtainedbycarryingoutleftmostreductions.Thesemanticequivalencefollowsfromthefactthattherewriterulesthemselvesaresound,andfromthefactthatsemanticequivalenceispreservedbysubstitutions. ⊣Lemma3.4(Substitution)Letϕbeanysentence,andlet⊢χ↔χ′.Supposethatϕ[p/χ]comesfromϕbyreplacingpbyχatsomepoint,andϕ[p/χ′]comessimilarly.Then⊢ϕ[p/χ]↔ϕ[p/χ′].Proof Byinductiononϕ.Thekeypointisthatwehavenecessitationrulesforeach[α]. ⊣ Theorem3.5ThislogicalsystemforL([α])isstronglycomplete:Σ⊢ϕiffΣ|=ϕ. ProofThesoundnesshalfbeingeasy,weonlyneedtoshowthatifΣ|=ϕ,thenΣ⊢ϕ.First,Σt|=ϕt.Sinceoursystemextendsthestandardcompleteproofsystemofmodallogic,Σt⊢ϕt.NowforeachχofL([α]),⊢χ↔χt.(Thisisaneasyinductionon 4.ALogicforL([α],2∗) Atthispoint,weturntothecompletenessresultforL([α],2∗).Itiseasytocheckthatthereisnohopeofgettingastrongcompletenessresult(whereonehasarbitrarysetsofhypotheses).Thebestonecanhopeforisweakcompleteness:⊢ϕifandonlyif|=ϕ.Also,incontrasttoourtranslationsresultsforL([α]),thelargerlanguageL([α],2∗)cannotbetranslatedintoLoreventoL(2∗)(modallogicwithextramodalities2∗B).WeprovethisinTheorem6.2below.Socompletenessresultsfor ∗ L([α],2)cannotsimplybebasedontranslation. OurlogicalsystemislistedinFigure2above.Wediscussedthefragmentofthesystemwhichdoes ∗ nothavethe∗axiomsandrulesinSection3.The2∗C-normalityAxiomand2C-necessitationRulearestandard,asistheMixAxiom.Weleavecheckingtheirsoundnesstothereader.ThekeyfeaturesofthesystemarethustheCompositionAxiomandtheActionRule.WebeginwiththeActionRule,restatedbelow: TheActionRuleLetψbeasentence,andletCbeasetofagents.Lettherebesentencesχβforallβsuchthatα→∗Cβ(includingαitself),andsuchthat1.⊢χβ→[β]ψ. A 2.IfA∈Candβ→γ,then⊢(χβ∧pre(β))→2Aχγ. Fromtheseassumptions,infer⊢χα→[α]2∗Cψ. RemarkWeuse→∗Casanabbreviationforthereflexiveandtransitiveclosureoftherelation∗A A∈C→.Recallthatthereareonlyfinitelymanyβsuchthatα→Cβ,sinceeachisdeterminedbyaworldofthesamefiniteKripkeframethatdeterminesα.SoeventhoughtheActionRulemightlooklikeittakesinfinitelymanypremises,itreallyonlytakesfinitelymany. Anotherpoint:ifonesodesires,theActionRulecouldbereplacedbya(morecomplicated)axiomschemewhichwewillnotstatehere. Lemma4.1W,w|=α3∗CϕiffthereisasequenceofworldsfromW w=w0 →A1 w1 →A2 ··· →Ak−1 wk−1 →Ak wk 18 wherek≥0,andalsoasequenceofactionsofthesamelengthk, α=α0 suchthatAi∈CandW,wi|=pre(αi)forall0≤i≤k,andW,wk|=αkϕ.RemarkThecasek=0justsaysthatW,w|=α3∗=αϕ.CϕisimpliedbyW,w| ProofW,w|=α3∗=pre(α)andWα,(w,α)|=3∗=pre(α)andCϕiffW,w|Cϕ;iffW,w| α thereisasequenceinW, (w,α)=v0 wherek≥0suchthatAi∈CandWα,vk|=ϕ.NowsupposesuchsequencesexistinWα.ThenwegetasequenceofworldswiinWandactionsαisuchthatvi=(wi,αi)andW,wi|=pre(αi).TheconditionthatWα,vk|=ϕtranslatestoW,wk|=αkϕ.Conversely,ifwehaveasequenceinWwiththeseproperties,wegetoneinWαbytakingvi=(wi,αi).⊣Proposition4.2TheActionRuleissound. ProofAssumethatW,w|=χαbutalsoW,w|=α3∗C¬ψ.AccordingtoLemma4.1,thereisalabeledsequenceofworldsfromW w=w0α=α0 wherek≥0andeachAi∈C,andalsoasequenceofactionsoflengthk,withthesamelabels,suchthatW,wi|=pre(αi)forall0Nowwearguethecasek>0.Weshowbyinductionon1≤i≤kthatW,wi|=χαi∧[αi]ψ.Inparticular,W,wk|=[αk]ψ.Thisisacontradiction.⊣WeclosewithadiscussionoftheCompositionRule,beginningwithageneraldefinition. DefinitionLetα=K,kandβ=L,lbeactions.Thentheactioncompositionα◦βistheactiondefinedasfollows.ConsidertheproductsetK×L.WeturnthisintoaKripkeframeusingtherestrictionoftheproductarrows.Wegetanactionstructurebysetting pre((k′,l′)) Finally,wesetα◦β=K×L,(k,l). = pre(k′)∧[K,k′]pre(l′). →A1 α1 →A2 ··· →Ak−1 αk−1 →Ak αk →A1 w1 →A2 ··· →Ak−1 wk−1 →Ak wk →A1 v1 →A2 ··· →Ak−1 vk−1 →Ak vk →A1 α1 →A2 ··· →Ak−1 αk−1 →Ak αk Proposition4.3Concerningthecompositionoperation:1.(Wα)β∼=Wα◦βviatherestrictionof((w,k′),l′)→(w,(k′,l′))to(Wα)β.2.TheCompositionAxiomissound:[α][β]ϕ↔[α◦β]ϕ.3.α◦(β◦γ)∼=(α◦β)◦γ. 4.α◦τ∼=α∼=τ◦α,wherethenullactionτisfromSection1.2. ProofLetα=K,kandβ=L,l.For(1),notethattheworldsof(Wα)βareoftheform((w,k′),l′),where(w,k′)∈WαandWα,(w,k′)|=pre(l′).Forsuch((w,k′),l′),W,w|=pre(k′)andW,w|=[K,k′]pre(l′).Thatis,(w,(k′,l′))∈Wα◦β.Theconverseissimilar,andtherestoftheisomorphismpropertiesareeasy. Part(2)followsfrom(1).Weusetheobviousisomorphism((k,l),m)→(k,(l,m))inpart(3).WeusetheCompositionand[α]-necessitationaxiomstoshowthatthisisomorphismpreservestheprefunctionuptologicalequivalence.Part(4)iseasy,usingthefactthat|=[τ]ϕ↔ϕ.⊣ 5.CompletenessforL([α],2∗)19 ExtendingtherewritingsystemRtoL([α],2∗).WeconsiderL([α],2∗).TherewritingsystemRextendsnaturallytothislargerlanguage,takingnewsymbolsfortheoperators2∗B.WealsoaddarulecorrespondingtotheCompositionAxiom:[α][β]ϕ;[α◦β]ϕ.WecallthisrewritingsystemR∗.Lemma4.4Thereisarelation 3.Ifψisapropersubsentenceofϕ,thenψ<ϕ. 4.Asentenceϕ∈L([α],2∗)isanormalformiffitisbuiltfromatomicsentencesusing¬,∧,2A, ∗ and2∗B,orifitisoftheform[α]2Bψ,whereαisanactioninnormalform,andψtooisinnormalform. ∗ β,pre(β)isanormalformsentence.5.Anactionαisanormalformifwheneverα→∗ 6.Ifα→β,then[α]2∗Cψ>[β]ψ. 7.nf(ϕ)≤ϕ. Onceagain,thedetailsareintheAppendix. InSection3,wesawatranslationtfromL([α])toLcanbeextendedtoatranslationfromL([α],2∗)totheinfinitarylanguageL∞,wherewehavecountableconjunctionsanddisjunctions.ThisextensionisdefinedusingPart(4)ofLemma4.4.Theadditionalclausesinthedefinitionoftare tt (2∗ϕ)=BA1,...,An∈B∗(2A1···2Anϕ)tt([α]2∗=Bψ)A1,...,An∈B∗([α]2A1···2Anψ)Inthisway,weseethatL([α],2∗)mayberegardedasafragmentofinfinitarymodallogic. RemarkItispossibletodroptheCompositionAxiominfavorofamoreinvolvedversionofthe ActionRule.ThepointistheCompositionAxiomsimplifiesthenormalformsoftheL([α],2∗):WithouttheCompositionAxiom,thenormalformsofsentencesofL([α],2∗)wouldbeoftheform[α1][α2]···[αr]ψ,whereeachαiisanormalformactionandψisanormalformsentence.TheCompositionAxiominsuresthatthenormalformsareoftheform[α]ψ.SoifweweretodroptheCompositionAxiom,wewouldneedaformulationoftheActionRulewhichinvolvedsequencesofactions.Itisnotterriblydifficulttoformulatesucharule,andcompletenesscanbeobtainedbyanelaborationoftheworkwhichweshalldo.Wedidnotpresentthiswork,mostlybecauseaddingtheCompositionAxiomleadstoshorterproofs. ThiscompletesthediscussionoftheaxiomsandrulesofourlogicalsystemforL([α],2∗).5.CompletenessforL([α],2∗) Inthissection,weprovethecompletenessofthelogicalsystemforL([α],2∗).Section5.1hassometechnicalresultswhichculminateintheSubstitutionLemma5.3.ThisisusedinsomeofourworkonnormalformsintheAppendix,andthatworkfiguresinthecompletenesstheoremofSection5.2.5.1SomeSyntacticResults Lemma5.1ForallA∈Candallβsuchthatα→Aβ,1.⊢[α]2∗Cψ→[α]ψ. ∗2.⊢[α]2∗Cψ∧pre(α)→2A[β]2Cψ. ProofPart(1)followseasilyfromtheMixAxiomandmodalreasoning.Forpart(2),westartwitha 20 ∗∗∗ consequenceoftheMixAxiom:⊢2∗Cψ→2A2Cψ.Thenbymodalreasoning,⊢[α]2Cψ→[α]2A2Cψ. ∗ BytheAction-KnowledgeAxiom,,wehave⊢[α]2∗⊣Cψ∧pre(α)→2A[β]2Cψ. DefinitionLetαandα′beactions.Wewrite⊢α↔α′ifαandα′arebasedonthesameKripkeframeWandthesameworldw,andifforallv∈W,⊢pre(v)↔pre′(v),wherepreistheannouncementfunctionforα,andpre′forα′.Wenotethefollowingbisimulation-likeproperties:1.If⊢α↔α′,thenalso⊢pre(α)↔pre(α′). ′′∗ 2.Wheneverβ′issuchthatα′→∗Cβ,thenthereissomeβsuchthat⊢β↔βandα→Cβ. Thesefolloweasilyfromthewaywedefinedpreonactionsintermsoffunctionsonframes.Lemma5.2If⊢α↔α′,thenforallψ,⊢[α]ψ↔[α′]ψ. ProofByinductiononψ.Forψatomic,ourresultiseasy.Theinductionstepsfor¬and∧aretrivial.Thestepfor2Aisnothard,andsoweomitit.Assumingtheresultforψgivestheresultfor[χ]ψ,usingtheCompositionAxiomandtheinductionhypothesis.Thisleavesthestepforsentencesof ∗′∗ theform2∗Bψ,assumingtheresultforψ.WeusetheActionRuletoshowthat⊢[α]2Cψ→[α]2Cψ. ′ Foreachβ′,weletχβ′be[β]2∗Cψ,whereβissuchthat⊢β↔β.Weneedtoshowthatforallrelevantβ′andγ′, ′ a.⊢[β]2∗Cψ→[β]ψ;and ′∗Ab.Ifβ′→γ′,then⊢[β]2∗Cψ∧pre(β)→2A[γ]2Cψ. For(a),weknowfromLemma5.1,part(1)that⊢[β]2∗Cψ→[β]ψ.Byinductionhypothesisonψ,⊢[β]ψ↔[β′]ψ.Andthisimplies(a).For(b),Lemma5.1,part(2)tellsusthatundertheassumptions, ∗ ⊢[β]2∗Cψ∧pre(β)→2A[γ]2Cψ.Asweknow,⊢pre(β)↔pre(β′).Thisimplies(b). Thiscompletestheinductiononψ. ⊣ Lemma5.3(Substitution)LettbeasentenceoractionofL([α],2∗),andlet⊢χ↔χ′.Supposethatt[p/χ]comesfromtbyreplacingpbyχatsomepoint,andt[p/χ′]comessimilarly.Then⊢t[p/χ]↔t[p/χ′].Proof Byinductionont,usingLemma5.2. ⊣ Lemma5.4Foreverysentenceϕ∈L([α],2∗)thereissomenormalformnf(ϕ)≤ϕsuchthat⊢ϕ↔nf(ϕ). ProofGivenϕ,thereisafinitesequenceϕ0;···;ϕn=ϕ′suchthatϕ0=ϕ,andϕnisinnormalform.Thisisaconsequenceofthefactthat 5.2Completeness TheproofofcompletenessanddecidabilityisbasedonthefiltrationargumentforcompletenessofPDLduetoKozenandParikh[10].Weshowthateveryconsistentϕhasafinitemodel,andthatthesizeofthemodelisrecursiveinϕ.WeshallneedtousesomeresultsconcerningtherewritingsystemR∗fromSection4. DefinitionLets(ϕ)bethesetofsubsentencesofϕ,includingϕitself.Thisincludesallsentencesoccurringinactionswhichoccurinϕandtheirsubsentences.Forfutureuse,wenotethat ∗∗ s([α]2∗ϕ)={[α]2ϕ,2ϕ}∪s(ϕ)∪{s(pre(β)):α→∗(5.1)CCCCβ} Wedefineafunctionf:L([α],2∗)→P(L([α],2∗))byrecursiononthewellfoundedrelation = ===== ∪∪∪∪ {p} f(ϕ)∪{¬ϕ} f(ϕ)∪f(ψ)∪{ϕ∧ψ}f(ϕ)∪{2Aϕ} ∗ f(ϕ)∪{2∗Bϕ}∪{2A2Bϕ:A∈B} ∗ {2A[β]2∗Cϕ:α→Cβ&A∈C} ∗ {[β]2∗Cϕ:α→Cβ&A∈C} {f(χ):(∃β)α→∗Cβ&χ∈s(pre(β))}∗f(2Cϕ) {f([β]ϕ):α→∗Cβ} Forϕnotinnormalform,letf(ϕ)=f(nf(ϕ)).(Notethatweneedtodefinefonsentenceswhichare notnormalforms,becausef([β]ψ)figuresinf([α]2∗Cϕ).Also,thedefinitionmakessensebecausethecallstofontheright-handsidesareall 3.Ifψ∈f(ϕ),thenf(ψ)⊆f(ϕ).4.Ifψ∈f(ϕ),thens(ψ)⊆f(ϕ). ∗∗∗ 5.If[γ]2∗Cχ∈f(ϕ),γ→Cδ,andA∈C,thenf(ϕ)alsocontains2A[δ]2Cχ,[δ]2Cχ,pre(δ),andnf([δ]χ). ProofAllofthepartsarebyinductiononϕinthewell-order<.Forpart(1),notethatif[α]2∗Cψis ∗ anormalform,theneachsentence2A[β]2Cψandallsubsentencesofthissentencearenormalforms.Forpart(2),notethatwhenϕisanormalform,ϕ∈f(ϕ). Inpart(3),weonlyneedtoconsiderϕinnormalform.Theresultisimmediatewhenϕisan ∗ atomicsentencep.Theinductionstepsfor¬,∧,and2Aareeasy.For2∗Bϕ,notethatsinceϕ<2Bϕ,ourinductionhypothesisimpliestheresultforϕ;weverifyitfor2∗Bϕ.Theonlyinterestingcaseis ∗ whenψis2A2BϕforsomeA∈B.Andinthiscase f(ψ) = ∗ f(2∗Bϕ)∪{2A2Bϕ} ⊆ f(2∗Bϕ). 22 ∗ Tocompletepart(3),weconsider[α]2∗Cϕ.Ifthereissomeχ<[α]2Cϕsuchthatψ∈f(χ)andf(χ)⊆f([α]2∗Cϕ),thenweareeasilydonebytheinductionhypothesis.Thiscoversallofthecases ∗∗ exceptforψ=[β]2∗Cϕandψ=2A[β]2Cϕ.Forthefirstofthese,weusethetransitivityof→Cto ∗ checkthatf([β]2∗Cϕ)⊆f([α]2Cϕ).Andnowthesecondcasefollows: f(2A[β]2∗Cϕ) = ∗ f([β]2∗Cϕ)∪{2A[β]2Cϕ} ⊆ f([α]2∗Cϕ). Part(4)issimilartopart(3),usingequation(5.1)atthebeginningofthissubsection. ∗ Forpart(5),assumethat[γ]2∗Cχ∈f(ϕ).Bypart(1),[γ]2Cχisanormalform.Weshowthat ∗∗ 2A[δ]2∗Cχ,[δ]2Cχ,pre(δ),andnf([δ]χ)allbelongtof([γ]2Cχ),andthenusepart(3).Thefirsttwoofthesesentencesareimmediatebythedefinitionoff;thethirdonefollowsfrompart(4);andthelastcomesfrompart(2)sincenf([δ]χ)∈f([δ]χ)⊆f([γ]2∗⊣Cχ.Theset∆=∆(ϕ)Fixasentenceϕ.Weset∆=f(ϕ)(i.e.,wedropϕfromthenotation).Thisset∆istheversionforourlogicoftheFischer-Ladnerclosureofϕ.Let∆={ψ1,...,ψn}.GivenamaximalconsistentsetUofL([α],2∗),let [[U]] = + ψn, wherethesignsaretakeninaccordancewithmembershipinU.Thatis,ifψi∈U,thenψisaconjunctof[[U]];butifψi∈/U,then¬ψiisaconjunct. Two(standard)observationsareinorder.Noticethatif[[U]]=[[V]],then[[U]]∧[[V]]isinconsistent.Also,forallψ∈∆, ⊢ψ↔{[[W]]:Wismaximalconsistentandψ∈W}.(5.2) ⊢¬ψ↔ {[[W]]:Wismaximalconsistentand¬ψ∈W}. (5.3) and (Thereasonisthatψisequivalenttothedisjunctionofallcompleteconjunctionswhichcontainit.However,someofthosecompleteconjunctionsareinconsistentandthesecanbedroppedfromthebigdisjunction.Theothersareconsistentandhencecanbeextendedtomaximalconsistentsets.)DefinitionThefiltrationFisthemodelwhoseworldsaretheequivalenceclasses[U],whereUisamaximalconsistentsetinthelogicforL([α],2∗),andtheequivalencerelationisU≡Viff[[U]]=[[V]](iffU∩∆=V∩∆).WesetF,[U]|=piffp∈U∩∆.Furthermore, A [V]inF[U]→ iffwhenever2Aψ∈U∩∆,thenalsoψ∈V.(5.4) Thisconditionisindependentofthechoiceofrepresentatives:weusepart(4)ofLemma5.5tosee thatif2Aχ∈∆,thenalsoχ∈∆.Agoodpathfrom[V0]forα3∗CψisapathinF [V0] →A1 [V1] α1→A2 →A2 suchthatk≥0,eachAi∈C,andsuchthatthereexistactions α=α0 suchthatpre(αi)∈Viforall0≤i≤k,andαkψ∈Vk. →A1 ··· →Ak−1 ··· →Ak−1 [Vk−1]αk−1 →Ak →Ak [Vk]αk TheideabehindagoodpathcomesfromconsideringLemma4.1inF.Ofcourse,thespecialcaseofthatresultwouldrequirethatF,[Vi]|=pre(αi)ratherthanpre(αi)∈Vi,andsimilarlyforαkψandVk.TheexactformulationabovewasmadeinorderthattheTruthLemmawillgothroughforsentencesoftheformα3∗Cψ(seethefinalparagraphsoftheproofofLemma5.8).Proof ∗∗Lemma5.6Let[α]2∗Cψ∈∆.Ifthereisagoodpathfrom[V0]forα3C¬ψ,thenα3C¬ψ∈V0. Byinductiononthelengthkofthepath.Ifk=0,thenα¬ψ∈V0.Ifα3∗/V0,thenC¬ψ∈ 5.CompletenessforL([α],2∗)23 [α]2∗Cψ∈V0.ByLemma5.1,part(1),wehave[α]ψ∈V0.Thisisacontradiction.Assumetheresultfork,andsupposethatthereisagoodpathfrom[V0]forα3∗C¬ψoflengthk+1. ∗∗ Thenthereisagoodpathoflengthkfrom[V1]forα13C¬ψ.Also,[α1]2Cψ∈∆,byLemma5.5,part(5).Byinductionhypothesis,α13∗C¬ψ∈V1. ∗∗ Ifα3C¬ψ∈/V0,then[α]2Cψ∈V0.ByLemma5.1,part(2),V0contains[α]2∗Cψ∧pre(α)→ ∗∗ 2A[α1]2Cψ.SoV0contains2A[α1]2Cψ.Again,thissentencebelongsto∆byLemma5.5,part(5). A inF,weseethat[α1]2∗Nowbydefinitionof→⊣Cψ∈V1.Thisisacontradiction. ∗ Lemma5.7If[[V0]]∧α3∗Cψisconsistent,thenthereisagoodpathfrom[V0]forα3Cψ. Wewillneedthefollowingstandardclaim,anargumentforwhichcanbefoundinKozenandParikh[10].WewillalsousethisclaimintheproofofLemma5.8.ClaimIf[[U]]∧3A[[V]]isconsistent,then[U]→A[V]. ProofofClaimAssume2Aψ∈U∩∆.Ifψ∈V,then¬ψ∈V,sosinceψ∈∆,⊢[[V]]→¬ψ.Thus,⊢3A[[V]]→3A¬ψ,andso⊢[[U]]∧3A[[V]]→2Aψ∧3A¬ψ,whence[[U]]∧3A[[V]]isinconsistent.Thiscontradictionestablishestheclaim. ProofForeachβsuchthatα→∗Cβ,letSβbethe(finite)setofall[W]∈Fsuchthatthereisno ∗ goodpathfrom[W]forβ3Cψ.Weneedtoseethat[V0]∈/Sα;supposetowardacontradictionthat[V0]∈Sα.Let χβ={[[W]]:W∈Sβ}. Notethat¬χβislogicallyequivalentto{[[W′]]:[W′]∈FandW′∈/Sβ}.SinceweassumedV0∈Sα,wehave⊢[[V0]]→χα. Wefirstclaimthatχβ∧βψisinconsistent.Otherwise,therewouldbe[W]∈Sβsuchthatχβ∧βψ∈W.NotethatbythePartialFunctionalityAxiom,⊢βψ→pre(β).Butthentheone-pointpath[W]isagoodpathfrom[W]forβ3∗/Sβ,andthisisacontradiction.Cψ.Thus[W]∈Soindeed,χβ∧βψisinconsistent.Therefore,⊢χβ→[β]¬ψ. WenextshowthatforallA∈Candallβsuchthatβ→Aγ,χβ∧pre(β)∧3A¬χγisinconsistent.Otherwise,therewouldbe[W]∈Sβwithχβ,pre(β),and3A¬χγinit.Then{3A[[W′]]:W′∈Sγ},beingequivalentto3A¬χβ,wouldbelongtoW.Itfollowsthat3A[[W′]]∈WforsomeW′∈Sγ.Bytheclaim,[W]→A[W′].Since[W′]∈/Sγ,thereisagoodpathfrom[W′]forγ3∗Cψ.But ∗ sinceβ→AγandWcontainspre(β),wealsohaveagoodpathfrom[W]forβ3Cψ.Thisagaincontradicts[W]∈Sβ.Asaresult,forallrelevantA,β,andγ,⊢χβ∧pre(β)→2Aχγ.BytheActionRule,⊢χα→[α]2∗[V0]]→χα.So⊢[[V0]]→[α]2∗C¬ψ.Now⊢[C¬ψ.Thiscontradictstheassumptionwithwhichwebeganthisproof.⊣Lemma5.8(TruthLemma)Considerasentenceϕ,andalsotheset∆=f(ϕ).Forallχ∈∆and[U]∈F:χ∈UiffF,[U]|=χ. ProofWearguebyinductiononthewellfounded [U]→[V].Thenbydefinitionof→,ψ∈V.Theinductionhypothesisappliestoψ,sinceψ<2Aψ,andsinceψ∈∆byLemma5.5,part(4).Sobyinductionhypothesis,F,[V]|=ψ.Thisgiveshalfofourequivalence.Conversely,supposethatF,[U]|=2Aψ.Supposetowardsacontradictionthat3A¬ψ∈U.So[[U]]∧3A¬ψisconsistent.Weuseequation(5.3)andthefactthat3Adistributesoverdisjunctionstoseethat[[U]]∧3A¬ψislogicallyequivalentto([[U]]∧3A[[V]]),wherethedisjunctionis 24 takenoverallVwhichcontain¬ψ.Since[[U]]∧3A¬ψisconsistent,oneofthedisjuncts[[U]]∧3A[[V]]mustbeconsistent.Theinductionhypothesisagainapplies,andweuseittoseethatF,[V]|=¬ψ. A BytheclaimintheproofofLemma5.7,[U]→[V].WeconcludethatF,[U]|=3A¬ψ,andthisisacontradiction. Forχoftheform2∗Cψ,weusethestandardargumentforPDL(seeKozenandParikh[10]).ThisisbasedonlemmasthatparallelLemmas5.6and5.7.Theworkissomewhateasierthanwhatwedobelowforsentencesoftheform[α]2∗Cψ,andsoweomitthesedetails. Weconcludewiththecasewhenχisanormalformsentenceoftheform[α]2∗Cψ∈∆.Assumethat∗∗ [α]2Cψ∈∆.First,supposethat[α]2Cψ∈/U.ThenbyLemma5.7,thereisagoodpathfrom[U]for ∗ α3C¬ψ.WewanttoapplyLemma4.1inFtoassertthatF,[U]|=α3∗C¬ψ.Letkbethelengthofthegoodpath.Fori≤k,pre(αi)∈Ui.Noweachpre(αi)belongsto∆byLemma5.5,part(5),andis<[α]2∗=pre(αi).WealsoneedtocheckthatCψ.Sobyinductionhypothesis,F,[Ui]| F,[Uk]|=αk¬ψ.Forthis,recallfromLemma5.5,part(5)that∆containsnf([αk]ψ)≤[αk]ψ.ByLemma4.4,nf([αk]ψ)≤[αk]ψ<[α]2∗Cψ.Sincethepathisgood,Ukcontainsαk¬ψandhence¬[αk]ψ.Italsomustcontainthenormalformofthis,byLemma5.4.Sobyinductionhypothesis,F,[Uk]|=nf(¬[αk]ψ).Bysoundness,F,[Uk]|=αk¬ψ.NowitdoesfollowfromLemma4.1thatF,[U]|=α3∗C¬ψ. Goingtheotherway,supposethatF,[U]|=α3∗C¬ψ.ByLemma4.1,wegetapathinFwitnessingthis.Theargumentofthepreviousparagraphshowsthatthispathisagoodpathfrom ∗ [U]forα3∗⊣C¬ψ.ByLemma5.6,Ucontainsα3C¬ψ.Thiscompletestheproof.Theorem5.9(Completeness)Forallϕ,⊢ϕiff|=ϕ.Moreover,thisrelationisdecidable.ProofByLemma5.4,⊢ϕ↔nf(ϕ).Letϕbeconsistent.BytheTruthLemma,nf(ϕ)holdsatsomeworldinthefiltrationF.Sonf(ϕ)hasamodel;thusϕhasone,too.Thisestablishescompleteness.Fordecidability,notethatthesizeofthefiltrationiscomputableinthesizeoftheoriginalϕ.⊣5.3TwoExtensions WebrieflymentiontwoextensionsoftheCompletenessTheorem5.9.TheseextensionscomefromourdiscussionattheendofSection2.3. First,considerthecaseofS5actions.WechangeourlogicalsystembyrestrictingtotheseS5actions,andweaddtheS5axiomstoourlogicalsystem.WeinterpretthisnewsystemonS5models.ItiseasytocheckthatapplyinganS5actiontoanS5modelgivesanotherS5model.Further,theS5actionsareclosedundercomposition.Finally,ifαisanS5actionandα→Aβ,thenβalsoisanS5action.Theseeasilyimplythesoundnessofthenewaxioms.Forcompleteness,weneedonlycheckthatifweassumetheS5axioms,thenthefiltrationFfromtheprevioussectionhastheproperty A thateach→isanequivalencerelation.Thisisastandardexerciseinmodallogic(see,e.g.,Faginetal[4],Theorem3.3.1). Oursecondextensionconcernsthemovefromactionsaswehavebeenworkingthemtoactionswhichchangethetruthvaluesofatomicsentences.Ifwemakethismove,thentheaxiomofAtomicPermanenceisnolongersound.However,itiseasytoformulatetherelevantaxioms.Forexample,ifwehaveanactionαwhicheffectsthechangep:=p∧¬q,thenwewouldtakeanaxiom[α]p↔(pre(α)→p∧¬q).Havingmadethesechanges,alloftherestoftheworkwehavedonegoesthrough.Inthisway,wegetacompletenesstheoremforthislogic. 6.ResultsonExpressivePower Inthissection,wepresenttworesultswhichshowthataddingannouncementstomodallogicwith3∗addsexpressivepowerasdoesaddingprivateannouncementstomodallogicwith3∗andpublicannouncements.ToshowtheseresultsitwillbesufficienttotakethesetAofagentstobe{A,B}andconsideronlylanguagescontainedinalanguagebuilt-upfromtheatomicsentencespandq,using 6.ResultsonExpressivePower 25 ∗∗ 3A,3B,3∗A,3B,and3AB,andtheactions[ϕ]A,[ϕ]BofannouncingϕtoAorBprivately,and[ϕ]ABtheactionofannouncingϕtoAandBpublicly.LetLallstandforthislanguage.Weuseherethecustomarynotation([ϕ]A,[ϕ]B,[ϕ]AB)forannouncements,but[ϕ]Aissimplytheactionwiththe A fromktokandpre(k)=ϕ.Wethinkof[ϕ]Bsimilarly.[ϕ]ABisKripkestructureK={k}with→ AB theactionwiththeKripkestructureK={k}with→and→goingfromktokandpre(k)=ϕ.Weneedtodefinearank|ϕ|onsentencesfromLall.Let|p|=0forpatomic,|¬ϕ|=|ϕ|,|ϕ∧ψ|=max(|ϕ|,|ψ|),|¬ϕ|=|ϕ|,|3Xϕ|=1+|ϕ|,forX=AorX=B,|3∗Xϕ|=1+|ϕ|forX=A,X=B,orX=AB,and|[ϕ]Xψ|=max(|ϕ|,|ψ|)forX=A,X=B,orX=AB. Firstwepresentalemmawhichallowsus,incertaincircumstances,todothefollowing:fromtheexistenceofasentenceinalanguageL1whichisnotequivalenttoanysentenceinalanguageL0inferthatthereexistsasentenceinL1notequivalenttoanytheoryinL0. Lemma6.1LetL0bealanguageincludedinLall,andletψbeasentenceinLall.AssumethatforeachnwehavemodelsFnandGnwithsomeworldsfn∈Fnandgn∈GnsuchthatFn,fnsatisfies¬ψ,Gn,gnsatisfiesψ,andFn,fnandGn,gnagreeonallsentencesinL0ofrank≤n.Then3AψisnotequivalentwithanytheoryinL0. ProofForasequenceofmodel-worldpairsHn,hn,n∈D⊆ω,weletn∈D(Hn,hn)beamodel-worldpairdefinedasfollows.Lethbeanewworld.TakedisjointcopiesoftheHn’sandaddanA-arrowfromhtoeachhn.AllotherarrowsarewithintheHn’sandstaythesameasinHn.Noatomicsentencesaretrueath.AtomicsentencestrueintheworldsbelongingtothecopyofHnin n)arepreciselythosetrueatthecorrespondingworldsofHn.n∈D(Hn,h m n∈ω(Fn,fn)withthenewworlddenotedbyf.DefinealsoF,form∈ω,tobeLetFbem whereHm=Gm,hm=gmandforalln=m,Hn=Fnandn∈ω(Hn,hn)withthenewworldf hn=fn Nowassumetowardsacontradictionthat3AψisequivalentwithatheoryΦinL0.Clearly3AψfailsinF,f.Thussomesentenceϕ∈ΦfailsinF,f.Ontheotherhand,eachFm,fmsatisfies3Aψ,whenceFm,fmsatisfiesϕ.Letm0=|ϕ|.ThefollowingclaimshowsthatbothF,fandFm0,fm0makeϕtrueorbothofthemmakeitfalse,whichleadstoacontradiction. ClaimLetϕbeasentenceinΦofrank≤m.LetHn,Kn,n∈D,withhn∈Hnandkn∈KnbemodelssuchthatHn,hnandKn,knagreeonsentencesinΦofrank≤m.Thenn(Hn,hn),handn(Kn,kn),kagreeonϕ. Thisclaimisprovedbyinductiononcomplexityofϕ.Itisclearforatomicsentences.Theinductionstepsforbooleanconnectivesaretrivial.Amomentofthoughtgivestheinductionstepfor3and3∗withvarioussubscripts.Itremainstoconsiderthecasewhenϕ=[ϕ1]Aϕ2.(Thecaseswhenϕ=[ϕ1]Bϕ2andϕ=[ϕ1]ABϕ2aresimilar.)FixHn,Kn,hn∈Hn,kn∈Kn,withn∈D,suchthatHn,hnandKn,knagreeonsentencesinΦofrank≤m.Notethat,foreachn∈D,Hn,hn|=ϕ1 ′ ifandonlyifKn,kn|=ϕ1.LetD1bethesetofalln∈DforwhichHn,hn|=ϕ1.LetHnand′ KnbemodelsobtainedbyupdatingHnandKnby[ϕ1]A.Bythedefinitionofrankandthefactthat ′′ |ϕ1|≤m,wehavethatHn,hnandKn,knagreeonsentencesfromΦofrank≤m.Therefore,byourinductivehypothesis ′′ Hn,h|=ϕ2iffKn,k|=ϕ2. n∈D1 n∈D1 However, ′ Hn,h|=ϕiffHn,h|=ϕ2 n n∈D1 and Kn,k|=ϕiff n n∈D1 ′ Kn,k|=ϕ2, andwearedone.⊣ 26 6.1AnnouncementsaddExpressivePowertoModalLogicwith2∗ Intheresultbelow,therewillbeonlyoneagentA,andsoweomittheletterAfromthenotation. ∗ WeletL([],3∗)bemodallogicwithannouncements(tothisA)and3∗=3∗A.WealsoletL(3)betheobvioussublanguage. Theorem6.2ThereisasentenceofL([],3∗)whichcannotbeexpressedbyanysetofsentencesofL(3∗). ProofWeshowfirstthat[p]3+q=[p]33∗qcannotbeexpressedbyanysinglesentenceofL(2∗).(Incidentally,thesameholdsfor[p]3∗q.)Fixanaturalnumbern.WedefinestructuresA=AnandB=Bnasfollows.FirstBhas2n+3pointsarrangedcyclicallyas 0→1→···→n→n+1→−n→···→−1→0. Fortheatomicsentences,wesetptrueatallpointsexceptn+1,andqtrueonlyat0. narrangedasThestructureAisacopyofBwithnmorepoints 0→ n→0. TheshapeofAisafigure-8.Inbothstructures,everypointisreachablefromeverypointbythe transitiveclosureofthe→relation.Atthepoints 6.ResultsonExpressivePower 27 WereturntothemodelsAandBdescribedinthebeginningofthisproof.For0≤i≤n,weletSi⊆A×Bbethefollowingset Si = ∪∪ {(0,0),...,(n,n),(n+1,n+1),(−n,−n),...,(−1,−1)} n−1,−2)...,(1,−n)}{( {(n−i,n−i)} Inthecaseofi=n,thenthelastdisjunctisempty.NotethatS0⊃S1⊃···⊃Sn.Also,for 0≤i≤n,everypointofonestructureisrelatedbySitosomepointoftheother.Claim2If0≤i≤nand(a,b)∈Si,thenA,a∼iB,b. Theproofisbyinductiononi.Ifi=0,thisisduetothefactthatpairsinS0agreeontheatomicformulas.Assumethestatementfori,andthati+1≤n.Let(a,b)∈Si+1.WeonlyneedtoshowthatIIcanrespondtoanyplayandhavetheresultingpairbelongtoSi.SupposefirstthatIplaysa3-move.Supposealsothata=b,sothat(a,a)comesfromthefirstsubsetofSi+1.Inthiscase,weonlyneedtonoticethat(a+1,a+1)∈Siif|a|≤n,(−n,−n)∈Siifa=n+1,and( a,a)belongsto thethirdsubsetofSi+1,thena≤n−(i+1)=n−i−1.Soa+1≤n−i,and( 28 Theproofisbyinductiononϕ.Forϕ=p,theresultisclear,asaretheinductionstepsfor¬and∧.For3Aϕ,supposethataj|=3Aϕ.Eithera∞|=ϕ,inwhichcasea∞|=3Aϕ,orelseaj−1|=ϕ.Inthelattercase,byinductionhypothesis,a∞|=ϕ;whencea∞|=3Aϕ.Theconverseissimilar.Thecaseof3Bϕistrivial:aj|=¬3Bϕanda∞|=¬3Bϕ. ∗ For3∗Aϕ,notethatsincewehaveacycle(6.1)containingallpoints,thetruthvalueof3Aϕdoes ∗ notdependonthepoint.Thecasesof3∗Bϕand3ABϕaresimilar. For[ϕ]ABψ,assumetheresultforϕandψ,andlet|[ϕ]ABψ| AB havesome→-successorinH(e.g.,a∞),andtherewouldbeno→edges.Soweassumeb∈H.Thusai∈/Hforsomei WegetTheorem6.3directlyfromtheclaim,theobservationthatGn,an|=χandGn,a∞|=¬χ,andLemma6.1.⊣Wefeelthatourtworesultsonexpressivepowerarejustasampleofwhatcouldbedoneinthisarea.Wedidnotinvestigatethenextnaturalquestions:Doannouncementswithsuspiciousoutsidersextendtheexpressivepowerofmodallogicwithallsecureprivateannouncementsandcommonknowledgeoperators?Andthendoannouncementswithcommonknowledgeofsuspicionaddfurtherexpressivepower? 7.ConclusionsandHistoricalRemarks Theworkofthispaperbuildsonthelongtraditionofepistemiclogicaswellastechnicalresultsinotherareas.Inrecenttimes,oneveryactivearenaforworkonknowledgeisdistributedsystems,andthemainsourceofworkinrecenttimesonknowledgeindistributedsystemsisthebookReasoningAboutKnowledge[4]byFagin,Halpern,Moses,andVardi.Wedepartfrom[4]byintroducingthenewoperatorsforepistemicactions,andbydoingwithouttemporallogicoperators.Ineffect,ourKripkemodelsaresimpler,sincetheydonotincorporatealloftherunsofasystem;thenewoperatorscanbeviewedasacompensationforthat.Wehavenotmadeadetailedcomparisonofourworkwiththelargebodyofworkonknowledgeondistributedsystems,andsuchacomparisonwouldrequirebothtechnicalandconceptualresults.Onthetechnicalside,wesuspectthatneitherframeworkistranslatableintotheother.Onewaytoshowthiswouldbebyexpressivityresults.Perhapsanotherwaywouldusecomplexityresults.Inthisdirection,wenotethatHalpernandVardi[8]examinesninety-sixlogicsofknowledgeandtime.Thirty-twoofthesecontaincommonknowledgeoperators,andofthese,allbuttwelveoftheseareundecidable.Butoverall,ourlogicsarebasedondifferingconceptualpointsandintendedapplications,andsoweareconfidentthattheydiffer. Asfarasweknow,thefirstpapertostudytheinteractionofcommunicationandknowledgeinaformalsettingisPlaza’spaper“LogicsofPublicCommunications”[13].Asthetitlesuggests,theepistemicactionsstudiedareannouncementstothewholegroup,asinourαandα′.Perhapsthemainresultofthepaperisacompletenesstheoremforthelogicofpublicannouncementsandknowledge.ThisresultiscloselyrelatedtoaspecialcaseofourTheorem3.5.ThedifferenceisthatPlazarestrictsattentiontothecasewhenalloftheaccessibilityrelationsareequivalencerelations. 8.Appendix:thelexicographicpathorder 29 Incidentally,Plaza’sproofinvolvesatranslationtomulti-modallogic,justasoursdoes.Inadditiontothis,[13]containsanumberofresultsspecialtothelogicofannouncementswhichwehavenotgeneralized,anditalsostudiesanextensionofthelogicwithnon-rigidconstants. OtherpredecessorstothispaperarethepapersofGerbrandy[5,6]andGerbrandyandGroene-veld[7].Thesestudyepistemicactionssimilartoourβ,whereanannouncementismadetosetofagentsinaprivatewaywithnosuspicions.Theypresentedalogicalsystemwhichincludedthecommonknowledgeoperators.AnimportantresultisthatallofthereasoningintheoriginalMuddyChildrenscenariocanbecarriedoutintheirsystem.Thisshowsthatinordertogetaformaltreat-mentoftheproblem,oneneednotpositmodelswhichmaintainhistories.Theydidnotobtainthecompleteness/decidabilityresultfortheirsystem,butitwouldbetheversionofTheorem5.9restrictedtoactionswhicharecompositionsofprivateannouncements.SoitfollowsfromourworkthatallofthereasoningintheMuddyChildrencanbecarriedoutinadecidablesystem. Weshouldmentionthatthesystemsstudiedin[5,6,7]differfromoursinthattheyarevariantsofdynamiclogicratherthanpropositionallogic.Thatis,announcementsareparticulartypesofprogramsasopposedtomodalities.Thisisanaturalmove,andalthoughwehavenotfolloweditinthispaper,wehavecarriedoutastudyofexpressivepowerissuesofvariousfragmentsofadynamiclogicwithannouncementoperators.Wehaveshown,forexample,thatthedynamiclogicformulationsaremoreexpressivethanthepurelypropositionalones.Detailsonthiswillappearinaforthcomingpaper. Incidentally,thesemanticsin[5,6,7]usenon-wellfoundedsets.Inotherwords,theyworkwithmodelsmodulobisimulation.TheadvantagesofmovingfromthesetoarbitraryKripkemodelsarethatthelogiccanbeusedbythosewhodonotknowaboutnon-wellfoundedsets,andalsothatcompletenessresultsareslightlystrongerwithamoregeneralsemantics.Therelevantequivalenceofthetwosemanticsisthesubjectoftheshortnote[11].Thefollowingarethenewcontributionsofthispaper: 1.Weformulatedalogicalsystemwithmodalitiescorrespondingtointuitivegroup-levelepistemicactions.Theseactionsincludenaturalformalizationsofannouncementssuchasγandδ,whichallowvarioustypesofsuspicionbyoutsiders.Ourapparatusalsopermitsustostudyepistemicactionswhichapparentlyhavenotyetbeenconsideredinthislineofwork,suchasactionsinwhichnothingactuallyhappensbutoneagentsuspectsthatasecretcommunicationtookplace.2.Weformulatedalogicalsystemwiththesemodalitiesandwithcommonknowledgeoperatorsforallgroups.BuildingonthecompletenessofPDLandusingabitoftermrewritingtheory,weaxiomatizedthevaliditiesinoursystem.3.Weobtainedsomeresultsonexpressivepower:inthepresenceofcommonknowledgeopera-tors,itisnotpossibletotranslateawaypublicannouncements,andinourframework,privateannouncementsaddexpressivepowertopublicones.8.Appendix:thelexicographicpathorder Inthisappendix,wegivethedetailsonthelexicographicpathordering(LPO),bothingeneralandinconnectionwithL([α])andL([α],2∗). Fixsomemany-sortedsignatureΣofterms.InordertodefinetheLPO Hereishowthisisappliedinthispaper.Weshalltaketwosorts:sentencesandactions.Oursignaturecontainstheusualsentence-formingoperatorsp(forp∈AtSen)¬,∧,and2AforallA∈A.Hereeachpis0-ary,¬and2Aareunary,and∧isbinary.Wealsohaveanoperatorapptakingactionsandsentencestosentences.Wethinkofapp(ψ,α)asmerelyavariationon[α]ψ.(Theorderofargumentstoappissignificant.)Wefurtherhaveabinaryoperator◦onactions.(Thisisadeparturefromthetreatmentofthispaper,sinceweused◦asametalinguisticabbreviationinsteadofasaformalsymbol.ItwillbeconvenienttomakethischangebecausethisleadstoasmoothertreatmentoftheCompositionAxiom.)Finally,foreachfiniteKripkeframeKoverL([α])andeach1≤i≤|K|,wehaveasymbolFiKtaking|K|sentencesandreturninganaction.Eachsentenceϕhasaformalversion α.Thesearedefinedbytherecursionwhichisobviousexceptfortheclauses ϕ, α = FiK( pre(kn)) Hereα=K,ki,prewithK={k1,...,kn}insomespecifiedorder.However,outsideoftheproof ofProposition8.2weshallnotexplicitlymentiontheformalversionsatall,sincetheyarehardertoreadthanthestandardnotation. Wemustalsofirstfixawellfoundedrelation 31 ui Proposition8.2ConsidertheLPO β,thenpre(β)<α.1.Ifα→ ∗ 2.Ifα→β,then[β]ψ<[α]2∗Cψ. 3.pre(α)→p<[α]p.4.pre(α)→¬[α]ψ<[α]¬ψ. 5.[α]ψ∧[α]χ<[α](ψ∧χ). A β}<[α]2Aψ.6.pre(α)→{2A[β]ψ:α→7.[α◦β]ϕ<[α][β]ψ. Inparticular,forallrulesϕ;ψoftherewritingsystemR∗,ψ<ϕ. ProofPart(1)holdsbecauseweregardαasatermα=Fiγn),forsomeframeKandi.K( ∗ Sowheneverα→β,eachpre(β)isapropersubtermofα. α).Nowlexicograph-Hereistheargumentforpart(2):Weneedtoseethatapp(ψ,ically,(ψ,α).Soweonlyneedtoknowthatα).Letα=Fiγn).K( γ1,...,Nowaccordingtoequation(2.1)inSection2.1, γi β References 1.AlexandruBaltag,Alogicofepistemicactions,ms.CWI,Amsterdam,1999. 2.WilfriedBuchholz,Proof-theoreticanalysisofterminationproofs,AnnalsofPureandApplied Logic75(1995)57–65.3.NachumDershowitz,Orderingsforterm-rewritingsystems,TheoreticalComputerScience17 (1982)279–301.4.RonaldFagin,JosephY.Halpern,YoramMoses,andMosheY.Vardi,ReasoningAboutKnowl-edge,MITPress,1995.5.JelleGerbrandy,Dynamicepistemiclogic,inLawrenceS.Moss,etal(eds.)Logic,Language, andInformation,vol.2,CSLIPublications,StanfordUniversity1999.6.JelleGerbrandy,BisimulationsonPlanetKripke,Ph.D.dissertation,UniversityofAmsterdam, 1999.7.JelleGerbrandyandWillemGroeneveld,Reasoningaboutinformationchange,JournalofLogic, Language,andInformation6(1997)147–169.8.JosephY.HalpernandMosheY.Vardi,Thecomplexityofreasoningaboutknowledgeand time.I.Lowerbounds.JournalofComputerandSystemSciences38(19),no.1,195–237.9.S.KaminandJ.J.Levy,Twogeneralizationsoftherecursivepathorderings,Unpublishednote, DepartmentofComputerScience,UniversityofIllinois,UrbanaIL(1980).10.D.KozenandR.Parikh,AnelementaryproofofthecompletenessofPDL,TheoreticalComputer Science(1981)113–118.11.LawrenceS.Moss,FromhypersetstoKripkemodelsinlogicsofannouncements,inJ.Gerbrandy etal(eds.)JFAK.EssaysDedicatedtoJohanvanBenthemontheOccasionofhis50thBirthday,Vossiuspers,AmsterdamUniversityPress,1999.12.DavidA.Plaisted,Terminationorderings,inD.Gabbay(etal)eds.,HandbookofLogicin ArtificialIntelligenceandLogicProgramming,vol.I,273–3.13.JanPlaza,Logicsofpubliccommunications,Proceedings,4thInternationalSymposiumonMethod-ologiesforIntelligentSystems,19.14.FrankVeltman,Defaultsinupdatesemantics,JournalofPhilosophicalLogic25(1996)221–261.
因篇幅问题不能全部显示,请点此查看更多更全内容
Copyright © 2019- 517ttc.cn 版权所有 赣ICP备2024042791号-8
违法及侵权请联系:TEL:199 18 7713 E-MAIL:2724546146@qq.com
本站由北京市万商天勤律师事务所王兴未律师提供法律服务