You are on page 1of 20

First-Order-Defi

nableModal
Systems

I Motivation
Our dcvclopmcnt of modaLllogic has bccn ahistorical or cvcn anti-historical: modal deduction systems,eachcharacterizinga deductiveconscqucnccrelation, wcrc introduccd and studied long bclbrc possiblc
worlds semanticswas discovered.But the semanticalviewpoint providcs a supcdor pcrspcctivcfrom vr'hichto cxplain why thcrc arc many
modal systcmsof intcrcst bcsidcs5M (S5).In thjs scction,wc introducc
an cxtra componcnt into intcrprctations, whosc paramctcrs can bc sct
in dilltrent ways to generatedillerent (scmanticallydelined) systems.
In the SM semantics,if sometring is possible at one wor1d,it is so al
ovcryworld. But this excludesan idea about possibility which has somc
intuitive force, the idea that what is possible(at a world) is detcrmined
in part by how things in fact arc (at that world). Pcrhapsccrtain statcs
of al'lairs arc impossible,given the way things actually are,but if things
had been ditTerent,those statesof aflairs would havebeen possible.In
othcr words, what is possiblc may vary from world to worldThcrc arc no uncontrovcrsialcxamplcsof this phcnomcnon,but hcrc
is a conlrovcrsial onc. lt is hcld by sornc philosophcrsthat if an organism in lact dcvclops fiom a cc ain cnljty or cntitics, then tftat organ
ism could not have developedliom dif'lerent entities-[jor instance,if l{
is a human being who actually originatesfiom a sperm .s,and an egge,,
then accordingto this view H could not have originated from different
entities. Thc idea is that anlthing devcloping from diffcrcnt cntitics
would notbc H, but someoneelse,evenif very similar to H. But thc vicw
as statcd is imprecise:lhe claim may be that H cor.rldnot havc odginated from a difl'erent egg dnd a difi'erent sperm, or tbat H could not havc
originatcd from a different egg or a different sperm. Supposcwc takc

Chapter3 : First-Order-Definable
lrodal Systems

thc first rcading. Thcn it is allowcd that H could havc originatcd liom
a dillerent egg er, so long as it is sr which fertilizes er, and that H could
have originated tiom a ditTerent sperrn s2, so long as it is e1 which s,
f?rtjlizes. So given that H originates fiom s, and er, that is, that in v/o
'propagules',
these irre I{'s
a world u in which I{ originates from.sr and
, is possiblc, and a world u' in which H originatcs from.st and e, is possiblc, but a world v in which H originates from .ri and s, is impossible.
But the impossibility is from the perspectivc of the world u.,* wherc H
originates from s, and r. Assuming that the doctrine about origin is
not itsclf sensitive to how things in fact arc, thcn if wc considcr mattcrs
from the point of vicw of ll, whcrc H originatcs from s, and u,, we find
that thc world vis possiblc, sincc onc of thc entitics from which H originates (in a) is retained in v. Thus yis possiblc rclativc to u, u is possiblc
relative to )a,",but v is impossible relative to ]r": relative possibility is
not transitivc.

2 Systems
'[o

captu-retheseideas,we introduce the relation of relative possibiJity


inlo th notion of interpretation, and we w te Rav to mean that v is
possiblerelative to u, or as it is often read, v is acces'sible
from ui tlf,formally, modal logicians oftcn rcad as Rav as u can see v. As cxplaincd
in Chapter 1.3, a binary rclation on a domain of djscourscD is a sct of
ordcrcd pairs of clementsof D. fhus our ncw accountof intcrprctation
is as fbllows:
A,rrinterpretation I with relative possibility of an LSMLsequent,
a general interp retation, is a set t4lof worlds and a sct li of
ordered pairs of elementsof I44each]v I4lis associatedwith
an assifJnmentof truth-values to the sentence-letterswhich
occur in the sequent,every]4/assitining1 to ',\'; and one )v I4l
is designatedth actual world, conventionallywritten 'lv"'"
Wc usc 'stzmdard inte.rpretation' for the kind of interpretation discusscdin Chaptcr 2.2,in which thcrc is no il" As wc leamed in Chaptcr
1"3,a binary relation can have a variety of stluct]ral properties, such
as reflexivity, slrnnetry or transitivity. We can generate different
modal systcms scmantically by dcf.ining diffcrcnt semantic conse-

52: Systems 9 l

qucncc rclations tss,ts,,,ctc., whcrc thc diffcrcncc lics in thc structu-ral


requircmcnts wc imposc on R. For instance, thcrc will bc a systcm fbr
thc cascwhcrc.l{ is rcflcxivc, a systcmfor thc cascwhcrc it is tTansitivc,
one for the casewhere it is both transitive and reUexive,and so on, as
wcll as a basic system where thcrc arc no constraints on R (howcvcr,
thc fact that onc collcction of constTaintson R diffcrs from anothcr
docs not guarantccthat thc systcmsthcy dctcrminc will diffcr). Wc will
use appropriate subscriptson the semanticconsequenceslmbol io distinguish the various systemsliom eachother^A system whose semantic consequencerelation is defined by stipulations on R expressiblein
first-order language(LFOL)is said to be a first-order definablem<>dal
systcm. Ifl the rest of tllis section we will introducc a few of the bestknown first-ordcr definable systcms (wcll ovcr a hundrcd havc bccn
studied).
But before going any further, we have to tie in the relative possibjlity
rclation to thc cvaluation clauscsfor the modal opcrators. Intuitively,
a statcmcnt 'op' should hold at a world w iff therc is some world u
wlic}Jis possible relat:ive to w w}J1crfp holds. Notice that we spcak now
of worlds, not possible worlds; the 'possible' has become a characterization of a relationship betlveen worlds. Irr place of the evaluation
clausesfor 5M, we usc the following evaluationclauscs,\'ith thc oncs
for the modal operatorsbeing changcd:
(EA): Vu' W, wlrrl = r ilf w.'rr + r;
(Eno): Vwe W,wlnpl = T iff (Va e W)(Rwu- ulpl: a);
(E.J: Vra/ W, wl.pl: T itf (la e W)(Rwu& ulpl = r);
(tr): W(p)= T iff )a/tlpl= r
For the rest of this section,we use (E!*) and (E.,.) in place of (Er) and
(Ea).(ED,r)saysthat p is necessaryat a world }v iff p is true at all worlds
lrycan see,and (EaR)saysthat p is possible at a world lryiff p is tTuc at
somc world ]a,can seeOur initial first-order definable system is called 'Gen' since it ariscs
by placing no resffictions on R. We will symbolize the semantic conse
quencerelauon of Gen by'er' to indicate the absenceof .restrictions.
This relation is deiined bv
p t,...,p, F a 4 iff there is no generalinterprctation on which all
of p1,...,p,are true while qisfalse.
Gen is the collection of semanticallycorrect sequntsp,,...,p, e q, and.
"

92

Chapter3r First-Order-Definable
ModalSystems

wc will immediatcly scc that *a nA - A. Conscqucntly,Cen and SM


(S5)are ditlerent systems,since of course Fe\atrA - AExample3.2.1:Sbowts6 !A * AInterpretation: W = {y*}, R : @,w": A - L. Fxplanation:The condition
'(Vu e W\(Rw"'u ' ulAl = T)'holds since the
or y instance is'(Ru/*w*
- )v{Al - T)', which is true because'R)a.,*yr'"'
is false. I-Iencew.[!Al T. In words, sincc )ry*can sce no worlds, 'A' holds at cvcry world which
\y* can sce.
It is crucial to the counterexampleto 'uA - A' that the world )ry*not
b able to see itself- If instead of R - O we had set fi : (w*,yi *), tlen
rcgardlessof what assignment)y* makesto 'A' wc would havc 'trA - A'
truc at )v*; in particular, if r4,*;A* r, thcn by (Elr), w.[nAl = -Las wcll,
since there is a world lry* can see,itself, where 'A' is not truc; and i{
w"llAl = r, then rry*ltrA- A] : T. But it may seem unreasonablc to
allow interpretations in which there are worlds which cannot see themselves.How can what happensat a world fail to be pos$ible relative to
that very world? ln other words, it sccms that thc relation of relativc
possibility should bc ref'lexive:(Vw)Rww.rSo our second first-o.rdcr
del'inablesystem is the systcm Rf, whose semantic consequencerclation is delined by:
pt,...,p, ts* 4 iff there is no generalinterprctation ? in which .ll
is rcflcxivc and on which all of p,,...,p,,ate true whilc 4 is falsc.
'l'hc

systcm Rf is the collectionof scmanticallycorrcct scquentspr,.,.,p,


FRr4, a collection which contains the collection of semanticallycorrect
sequentspv..,p, ea q- For if pb...,p" Fi, q, then no generalinterpretation makespr,...,p,all truc and 4 false,so in particula.rno generalinterprctation with rcflcxivc R makcs p.,...,p, aJ\ truc and 4 falsc; hcnce
pr...,pn t=nr4. Thus any sequentwhjch belongs to Gen, belongs to Rf.
On the other hand, there are sequents which belong to Rf but not to
Gen: tso DA - A but FRrnA - A. On tlfs accountwe say that Gen is a
proper subsystemof Rf and that Rf properly containsGen. More generally, S is a propcr subsystcmof S', and S' properly contains S, if cvc.ry
S-sequentis an s'-sequcntwhilc somc Slscqucnt is not an S-scqucnt.
r 'lbis does not mean that there is no intqest in systcms of moda.l logic in which ti is
not required to be reflexive. It merely means that it would be implausible to intcrpret the
'n' of such systcmsas exprcssingbrcadly logicalneccssity-Systemswithou t rcncxivc R
arc usualJy motilated by very differcnt readings of 'D'.

52: Systems 93

'DA
Thc percepuvereadermay havc noted that it sufficcs to validatc
- A' that we restrict ourselvesto just those generalinterprctations in
which R is reflexiveat w",that is, in which Rw"w*, evenif R is not 'globally' reflexive, that is, even if there is some other world n such tJrat
'Rau" We could introduce a semantic consequencerelation for this
wider class of general interpretations, and wc would obtain a somcwhat
'(Vw)Rww'
dift'ercnt system from Rf. However,global conditions like
yield systems with certain convenicnt formal featurcs-if p is a valid
formula in such a system,so is ap so in this chapterwe consider onJy
conscqucnccrelations defincd globally: wc say that a conditjon is global iff aI its terms are bound variablcs.
An altcrnativc global constraint on rclativc possibility which wc
might impose is that of slmmctry: (Vw)(VuXRwu- Ruw).We have
already proved otrA FssA, and henceby the soundnessof F* for Frr,
otrA trsMA, which is also easyto seein its own right: if 'otrA' holds at
].r.,"in a sta.ndard interpretation I'l then for some world u in ty, u[trA] :
r, and so'A'holds at every worldin W, including ]r". However,if n is
not rcquired to be s)mmeuic, we can easily give a gcneral intcrprctation which relutcs this sequent-Whetheror not n is refletve makes no
difference,but we n'ill give an examplewhere it is.
Example..:].2.2:Show oaA *RrA.
LctW= {y/",u},R: {(}i/*,a),(y/",}y*),(u,u)},ta/*;A* r, u.'A * T. Wc can
give a dia$am of this interpretation in which the directed arrows indicate lines of sight, and absenceof an arrow rurrring from onc world to
anothcr indicatcs that the fi-rst cannot seethe second-So this interprctation is as Dicturedbelow:
\ ,
u
Wc havc w*[A] = r. To seethat ]v*[o nA] = T, note that by (En,,),we have
[!Al = T, since'A'is true at every world u can see(Ir cannot se w*)^
Then since w* can see a, u,"[onAl = T by (EoJ. Thus .trA FRrA, and
ipso facto otrA Fo A.
In Example:1.2.2the failurc of R tobc symmc ic is crucial. Our third
tirst-order definable sysremis the system Sym,whose semanticconsequencerelation tss,.is defined by:

chapter3: First-Order-Definable
ModalSystems

p t,...,p, tssv.4 iff there is no gcncral intcrprctation in which R is


synmetric and on which a]l of pr,...,p, arc true whilc 4 is falsr.
The system Sym is the collection of semantically correct sequents
p r. ..,pr r=rnn
q, a collectionwhich containsthe collectionof scmantically
q,Ihenno generalintert=@
corfectsequentsp b...,p"tsaq.Forif p1,...,p"
prctation makcs pr,-..,p,all truc and 4 falsc, so in pafiicular no gcncral
interpretation with slmmetric R makes p,,...,p, ?l'l true and q fatse;
hence p,,...,p, Fsy-4" Thus any sequcnt which belongsto Cen, belongs
to Sym- On the other hand, olA Fa A while ocA Fsy.A. So Gen is a
proper subsystemof Sym. But cven though otrA trRrA, it does not follow that Rf is a subsystemof Sym, since there may be sequentswhich
bclong to Rf but not to Sym. And there arc; for instance, FRrBA - A,
but it is easyto chcck that *5y. trA * A (this is an exercise).So neither
Sym nor Rf is a subsystemof the other"
The remaining main structural property of binary relations is that of
& Ruv)* Rwv).The example about bi}
transitivity: (VwXVUXVvX(Rwu
bgical origin we gavein S1 in xplanationof thc notion of rclative possibilily motivates the vicw that rclativc possibility is not tTansitive.For
if )v is a world in which H odginates liom s, and e, and u a world in
which FI originatesfrom s, and er, then a is possiblerelative to lv (Rwu);
and if v is a world in whjch H originatesfrom s, and e,, then v is possiblc rclative to a (Ruv).But v is not possible relativc to Ir (-Rwv),and so
Fansitivity of R fails in this particular example. However, wc will ccrtainly want to investigate a system with transitive R, so our fourth firstorder definablesystemis the systemTrn, whosesemanticconsequence
rclation Fr," is dcfincd by:
p,,...,p^*r* q lff therc is no generalinterpretation in which R is
transitive and on which all of p,,...,p, arc truc whilc 4 is falsc"
'lhc

system Trn is the collection of semantically co.ffect sequents


pt,...,p, p*,4, a collection which, as can be shown by a now familiar
argumcnt, contains the collection of semantically coffcct sequents
pt,...,p,ca 4. In addition,wc have ooA F@oA, while ooA Frr" .A. So
Gen is a proper subsystemofTrn. In i'act,we also have aoA tssy.oA
oA.
and ooA tsRr

52: Systems

oA and ooA *RroAExample3.2.3:Show oaA trsvm


Interpretation: ln the following interpretation, R is rcflcxivc and symmetric. w:
lw*,u,v,\, R : {(w*,a),(a,v),(u,w*),(v,u),(w",w"l,(u,u),
(v,v)),ly".'A * -1,a.'A * f , y: A * T, or as pictuled:

t ,
A* r

l )

u
A* r

Hcrc wc have w*[A] = r, and also wo[on] = 1, since u is the only other
world w" can see and u[A] = r- However,a[aA] : T sincc u can scc v
and v[A] - T. But if uloAl : T, thcn ]t"[ooA] = T since v/" can scc u. In
te.rmsof oul example,supposethat 'A' means'H originatesfrom s, and
er'; then this interpretation illustrates that if H acflrally or.iginatesfrom
s, and e], it is not possible that H odginate from .ri and e2,but it could
havc bccn possible;it is possibly possible,so to speak.Sincenon'transitivity is crucial to the interprctation, it also shows that Trn contains
a sequentthat Rf and Sym both lack. However,each of thcse systems
also contain sequentsTrn lacks, so none of the three systemswith constraints on I{ is a subsystemof any other"
lhc sequentswe haveused to illustrate djffercnccs betwecn the systems havenot been chosenat random-Thcy are in a certain sensecharactenslrcof thcir associatedscmanlic consequencerelations.
. FRroA - A, but if C is any global constraint on R which does
not entail that R is reflexive,then tscDA - A.4
. o trA Fsv.A, but if C is any global constraint on R which does
not entail that R is slanrnetric,then atrA *c A.
. oaA Fr- oA, but if C is any gklbal constraint on R which
does not cntail that n is ffansitivc, then o oA trc oA.
Wc could continuc in thc sanc vcin, using futher structural propcrties to define new semanticconsequencerelations.But the best-known
systems of modal logic are obtained by combining some of tfle proper'
ties of R we have consideredso far. There are th,reesuch systems,corresponding to the global constraints that R be (i) reflexive and
4 Thc icstriction to global constnints is essential,as alreadyindjcatcd.If Cis simply thc
constraint that Rr4,*}v,,the sct ol valid formulae is not closedundcr Necessiratron-

ModalSystems
Chapter3 : First-Order-Definable

sFnmetric, (ii) rcflcivc and transitivc, and (iii) rcflcxivc, symmctric


and transitive, that is, an equivalenccrclation (secpage 59). The associated scmantic consequcnccrclations arc labcled ts*, tsR,and truo
respectively.They are defined by:
pv..,p, t-asq iff therc is no gcneralintcrprctation in $'hich R is
refledve and symmctric and on which all of pr,...,p, are truc
while a is falsc.
pv..,p, t=w4 iff there is no generalinterpretation in which R is
rcflexive and transitive and on which all of p,,.-.,p, are true
while q is falsc.
p,...,p" t=q 4 ifT there is no generalintrpretation in which R is
an cquivalence rclation and on which all of pr,..-,p. arc truc
while q is false.
The systemsRS,RT and Eq arc thc conesponding collectionsof semantically correct sequents.Cen is a propcr subsystemof all threc, neithcr
of RSnor RT is a subsystemof the other, and every sequcnt of Rs or
RT is in Eq-But Eq is distinct from both RSand RT-It is an exerciseto
show that oA tsRsDoA and oA FRrtroA, and it is easy to see that oA
tr& !oA. For if 'oA' holds at wn in W then )r* can seesome u in I/ such
that 'A' holds at l,|.Let v bc any world in I/ which ]v* can scc. Sinccwo
can seev, then by symmetry v can sce rry*.Also, sincc )v" can see ,, by
'oA'holds at v. And sincc
lTansilivity v can see a. But'A'holds at a; so
v was chosen arbitraflly from the worlds w" can see,this means that
'oA'holds at
'noA'holds at ]1,*-FlenceRs
evervworld wo can scc.Thus
and RT arc botr subsystcmso[ Eq.
We have now introduccd sevcnfirst-ordff dcfinablc systcms of scntcntial modal logic, Gen, Rf, Sym, Trn, RS, RT, and Eq, and havc
explainedthe notion of one system'sbeing a subsystemof another. To
keep track of the relationshipsamong thesesystems,the following diagram is useful, in which the system at thc tail of each arrow is a subsystem of the system at its tip. Arows illc transitive, ]ikc thc
subsystemrelationship; that is, any two arrows which connect tip to
tail can be considereda single anow Thus one systemis a proper subsystem of anotherif there is a path through the diagram t'rom the first
to thc second"Whereneither of two systemsis a subsystemof the oth-

S2:Systems

Eq
_t Y.
. /
RS

,/\r/\
u/^

r----.-

RT

R"ft

t
----t,/

r)n

,,-'-,

Gen

er, no path of a-rlowsconnectsthcm.


As the diagram suggests,thcrc is plcnty of'logical space' for the
inse.rtionof other systems-For example,therc is a group of frequently
studied systems which lie on the path from RT to Eq, each of $'hich
contains RT and is contained in Eq. One of thesc is the system RTc,
which has thc semanticconsequencerelation trRrcdefined by:
pb...,p, F=R.rc
q itf rhereis no generalinterpretation in which n is
reflexive, transitive and connectedand on which p,,...,p" are all
tTuc while q is false.
Thc rcquircmcnts of rcflcxivity and transitivity ensurc that every
sequcnt of RT is a sequent of RTC,since a sequent not in RTC has a
reflcxive transitive counterexample and so is not in RT-For a characteristic sequentfor RTC,consider the followingExample.3.2.4:Show E(oA v oB) FRrtroA v troB.
Interpretation:W = {w*,u,v},R = 1@",ul,lw",v),(w",w*),(u,n),(v,v)},
w * . ' A * r , B * T , u - - A* T , B * I , v . ' A * l , B * T .
x ,
;
A- T
B * T

a
A*'T
B - T

i
A--L
B * T

'oB'holds at so 'oA
In this interpretation, 'aA'holds at u and v/* and
%
'!(oA v oB)'
v oB'holds at every world which ),|,'"can see,and hence
holds at )v*.But'troA'is falsc al r.v"since'oA'docs not hold at cvcry

Chapler J: first-Order-Definablei,loddl Systems

'trcB'is
'oB'
falsc at w* sincc
world which r4.,"can scc (considcr v), and
docs not hold at cvcry world which w" can scc (considcr a). Howcvcr,
|hc lack ol' any conncction belween u and v is cruciitl Lo I his countcrcx
ample to thc scqucnt; any rcflexive fansitive counte.rexample to il will
havc at least two unconnected worlds visiblc from w".'ltrus u(oA v oB)
r=RrctroA v troB. Sincc thc.rc is no countcrcxamplc in which li is an
cquivalcncc rclation, and sincc oA FRrctr aA, wc also havc RTC a propcr
subsystem of Eq.
Wc end this discussion of first-ordcr definablc systcms with thrcc
points of intercst. First, it is natu-ral to ask what relationship thcrc is, if
any, betlvcen these systems iuld thc systcm SM (S5) of standard modal
logic which wc introduccd in Chaplcr 2" Wc arlluc that SM is Eq. Rccall
th.rl an cquivalcncc relation on a domain partitions thc domain into
mutually exclusive and jointly exhaustive equivalcnce classcs. So if 7 is
a gcncral intcrpretation in which R is an equivalenre relation, there is
a uniquc cquivalcncc class 4.I.,to which )v'- bclongs (in this spccial casc,
'Ilrv'can
bc rcad as'a and v can scc cxacllv thc samc worlds'). Morcovcr, thc truth-valuc of any scntencc p on'1, that is, lhe lruth-valuc ()1'
p at )v", dcpcnds only on thc lruth-valucs of scntcnccs at lhc othcr
wo.rlds in a-I,,.,,
for if wc bcgin tbc cvaluation of p at w*, thcn no mattcr
how many modal opcrators p contains, we arc ncvcr lcd outside L/-.,
sincc no world in {/., can scc any world outsidc a/.," constqucntly, thc
'discard' all
truth-valucs of all scntcnccs a[ w* rcmain thc samc if wc
worlds in W whjch arc not in [.r., and all pairs ol worlds in ii in which
some member is not in L/... lhis leaves us with a new interprctation 1;
which makes cxactly the same scntenccs truc as docs 1" llut in'1l, n is
univcrsal-(Vw)(Vu)Rwu sincc ?+'sdomain t// is just U*, and cvcry
wo.rld in a/,,,can scc cvcry othcr (as wcll as itscu). Howcvcr, whcn,R js
univcrsal, thc cvaluation clauscs (ElR) and (l1oR)ol this chaptcr arc
equivaleft to the clauses (Ur) and (Eo) of SM" lhus, for any world t,t in
ty#, thcro is a world v in llzowhich I,,can scc and at \a'hich a scntcncc O
'which
is truc iff therc is a world v in l4l+at which tf is truc
u can scc'
is rcd[ndant. Conscqucntly, if wc transform I into a standard intc{prctalion .',fby removing mcntion of R, and cvaluatc scntcnccs in .7 by thc
standard rather than thc general cvaluation clauscs, cxactly lhc same
s(:ntcnces will come out flue in J as are true in 1#,and hcncc as in ?.
lhis mcans that SM and Eq are exactly thc samc systcms. Iror whcncvcr
wc havc a gcncral intcrprctation 7 which cstablishcs pr,.,.,p,, iirq 4, thcn
by lbllowing thc proccdurc just described wc gct a stand.lrd intcrprctation J which establishcspt,...,p"*su 4. And convcrscly, if a standard

52:Systems

J cstablishcs pt,.. ,p, *su 4, thcn by adding a univcrsal i?, wc gct a gcncral ? which cstablishcs pr,...,p, *rq 4.
n sccond point ofnolc is that il would bc wronlt to concludc lrom thc
constnrction of thc cxtcnsions of Gen that any first-order del'inable
constraint on R dctcrmincs its own systcm, Each such constraint determincs a systcm, of coursc, but it may bc thc samc as onc dctcrmincd
by a di{Tcrcnt constrajrt. Ihc discussion of Eq in thc prcvious par '
graph illustralcs this. l}]c condjtion o{' univcrsality, (Vw)(Vu)Rwu, is a
strongcr constraint on /l tllan the condition that R be an equivalence
rclation (a univcrsal rclation is an cquivalcncc rclation, bul not cvcry
cquivalcncc rclation is a univcrsal rclation)- Yet as wc havc jus{ st:cn,if
wc wcrc to usc uflivcrsality to dciinc a scmantic conscqucncc rclation,
thc rcsulting systcm would bc Eq ovcr again. Anothcr cxamplc is irrcflcxivily, (Vw)-Rww: the semantic consequence relation Fr,, in fact coin
cjdcs with Fz (this is an excrcis(r,
'fhird
'first-order d,ctinablc
and lastly, usc of thc nomcnclatura
systcm' sullgcsts that thcrc arc such things as second'order dcl'irablc systcms (somc lamiliarity with sccond ordcr logic is prcsumcd in thc ncxt
l]!vo paragraphs). Sincc first-ordcr logic is contained in sccond ordcr
lo,jic, first{rder delinable systems are ipso facto sccond order dclin
ablc, so by a sccond-t.rrdcr dcfinablc systcm wc mcan onc thal is eisendally sccond-ordcr: at lcast onc of tbc constraints on R dcfining thc
scmantic conscqucncc rolation is cxprcsscd by a formula contalining
second-ordcr quantilicrs, and no flrst-ordcr conditions capture thc
samo scmantic conscqucncc relation. Simply using sccond-order languagc to impose some constraints on li is insufficicnt to gcncratc such
a systcm, sincc tho sccond ordcr Iormulac may bc logically cquivalcnt
to lirst-ordcr oncs. It is not cvcn suificicnt to usc somc conccpt which
we know is nol. first-order cxpressible. For cxamplc, [hcrc is no way ol'
'thcrc
cxpressing
arc at most finitely many x such that' in lirsl ordcr
languagc, but it can bc cxprcsscd w'ith sccond ordc-r quantificrs. Yet il
docs not follow tiom this that thc systcm F whosc scmantic consc
qucncc rclation FF is givcn by 'pr,...,p" ,-t q rtt t'hcrcis no gcncral inLcrprcl.alion with linite t'f on which all ol pt,...,p, arc truc whilc q is l'alsc'
is essentially second-order. ln lact, FF again coincidcs with pa"
'lo
construct an (essentially) second order definable system, dcfine
an n-chain in an intcrpretation to be a sequence of worlds wo, wt, tv2...
such that I{}ro}vr, nv,,r}/r, ctc. (Wc do not -rcquirc that thc worlds in an /{chain be distinct, so any world which can scc itscll' immcdiatcly givcs
dsc to an intinitcly long R-chain.) Thc scmantic consequence rclation

IOO Chapter3: First-Order-Definable


l,iodalSystems
trc dclincd as
pt,.".,p, F=cq iff there is no generalinterpretation in which R is
transitive, and every l?-chain is finite, and on which all of
pt,...,p,aretrue while 4is false

\"7

yiclds a systcm lnown as G which is essentidlll second-ordcr.In othcr


wo.rds,therc is no collection C of ftst-ordcl conditions on n klobal or
othcrfi/ise)such tJratFr...,p, t=cq ltf p,,...,p" Fc 4. G contajns Trn and
hence Gen, but differs from all the systemsof tJ]is section in view of
thc scqucnt Fc D(trA -' A) - aA. Thc formula'tr(trA -- A) * tlA'is
known as Ltib's Formula, and can be understood as expressing an
important corollary of Godcl's SccondIncomplctcncssThcorcm (G is
so-labclled after Gitdel). At this point, modal logic links up with the
main results of twentieth-century mathematical logic"

u Exercises
I

Show the following, explaining your solutions:


(1)
(3)
(5)
(7)
*(9)
(11)
(I3)

II

(2) oA +z aA;
DA Fo oA;
(4) A -^ tso0-A;
trA&trBFoA-B;
*(6) trA - A Fo rlA - .A;
trA - oA tso DA * A;
(8) aaA F6 oEA;
trA tso trtrA;
(10) trA * A Fo tr(DA ' AX
oEA tso troA;
(12) A * B Fu DA - B;
D(!(A * !A) * A) tsoA;
n A - o ( B- C ) F o o ( B- ( D A - o C ) .

Show the following, explaining you,r solutions:


(l)
(3)
(5)
(7)
(9)
(11)
(12)
(13)

A FortroA;
D(nA ' A) *n, rA;
oA FRsnoA;
D(A v oB) #RsuA v oB;
EoA tsrqoEA;
tr(tr(A -- trA) - A) FRrA
FRstr(nA - B) v D(nB* A)
tr(o.A - .A)*RsflA - DDA

(2)
(4)
(6)
(8)
(10)

trA FRrtrEA;
Fsv.aA * A;
.A FRrtr.A;
Fsv.o-,r;
tr(trA ' A) FeqnA;

III For each of the following, suggesta global first-order condition or


set of condiuons C which makes it correct (perhapsdifferent C in each
case)and explain your suggestion,C may not be any of the condjtions

in semantically
definedsystems l0l
53:Deducibility
or scts of conditions uscd to dcfinc systcmsin this scction,and should
be as weak a condition or sct of conditions as possiblc:
(1) Fc D(oA * A); (2) Fc DA - oA; (3) Fc cA.
IV Showthat RTC is a propcr subsystemof Eq (this has two parts, (a)
every RTCsequentis an Eq scquent,and (b) some Eq scquentis not an
RTC scqucnt).
V Explain why trr,,coincideswith Fo. lHint for the non-trivial direction, supposethat somegeneralintcrprctation ? cstablishcspr,...,p, tso
4. lndicate how an irleflexive interpretation J could be constructed
from 'l which wor.rldestablish pu...,p, *n 4. Considcr replacing each
world in I which can seeitself with two worlds of a certain sort.l

3 Deducibilityin semantically
definedsystems
We turn now to the question of providing rules of proof for some of the
systemsintroduced in the previous section"Ideally,what we sekis the
following: for eachfirst'order definablesystem.t with smanticconscqucncc rclation +, wc would likc to constTucta dcduction systcm .9'
with deductive consequencerelation r9 such that ry and *r coincidc.
In other words, the following two conditions should bc fulfilled:
(Sndj".J:if pr,...,p,t-s' qthenpb...,p,Esqi
(Comp.t,.r):
if p,,...,pnF=
s q r}]rcnp b...,p,,t-. q.
According to (Sndy"),thc deductivc systcm S' is sound with respectto
(or lbr) the semanticallydefined systemS,which meansthat eachprovable scquent in S' is scmantically corrcct on thc scmantics for S" And
accordingto (CompsrJ,the deductivesystemS' is completewith respect
to (or for) the semantically dcfined systcm S, which mcans that for
every sequentthat is semanticallycorrect on the semanticsof S, therc
is a proof of it using thc rulcs of S'. Wc also spcakof thc dcductivc consequencerelation ry being sound and/or completetbr,/with respect to
the semanticconsequencerelation trs.
Thcrc is no guarantcethat for eachof the systemsin 52 some sould
and completc dcductivc conscquenccrclation cxists; wc might bc pa.rticularly doubtful of finding such a relation for an csscnrially sccond-

l02

Chapter3: First-Order-Definable
ModalSystems

ordcr systcm likc G. Wc alrcady know, &ough, that thc dcductiv(: systcm 55 is sound and complclc I'or Eq, sincc Eq is lhc samc system as
SM, and wc statcd carlicr (wilhout prool) that 55 is sound and (:omplctc
for SM- ltemarkably, every system in $2 can be provided wilh sound
ilnd (:omplctc rulcs of proof. But thcre is a catch. Onc's lirst thoughl is
that Lhc rulcs lor thc othc.r systcms will bc variations on thc 55 rulcs
thc othcr scmantic conscqucncc rclations might bc capturablc simply
by adjusling thc constraints on El and ol- Ilowever, while it is possiblc
to obtain thc systcm RT in this way, tbcrc is no known way of obtaininit
any of thc othe.r systems in 52 similarly. A gcncrirl approach to dcduc
tion which works for all systems is rather diffcrcnt in naturo from nat
ural dcduction, and it is rathcr unwicldy. For that rcason, rvc wlll
d i s (u s s l h c m c l h o d I a i r l yh r i c l l y .
We definc a deductive conscqucncc rclation r* which is sound and
complete for Gen in the fbl[owing n'ay.
(i) lhe deductive system K has as non-modal basis all lhc.rulcs
of NK for thc scntcntial corincctivcs.
(ii) K has a nrlc callcd Neces'sitation,which is a rcstdctcd vcrsion
of trI; whcn a formula p at linc j in a proof has been inferred
Irom prcvious lincs in such a way [h:rt iL dcpcnds on no
assumptions or prcmiscs at j, or if it occurs at .i by fl, thcn
at a latcr linc k wc may add 'np', also dcpcnding on no prcmiscs or assumptions. k is labcld l, Ncc'.
(iii) K has a rulc I)fo, likc thc NK-rulc Df lbr'*', n'hich allows us
lo add a new line k to a proof by rcplacing any occurrcncc ol'
'-r-'
in a fbrmula at line i with'o', or in the ofher dircdion,
'o'
'-tr-'- k is labehd 'j, Dfo',
any occurrcncc of
with
and
dcpcnds on whatcvcr k dcpcnds on. Notc ttat, lor <rrnvcniencc, we do not restrict applications of Dfo to cascs whcrc
'-'
'-tr-'
thc Iirst
bcing abbrcviatcd is thc main conof thc
'-r-l-'
ncctivc of thc formula: drll occurrcncc o1
may bc
'o'
abbrcviatcd, and any
may bc cxpiuded.
(iv) K has a spccial cxtcnsion of Scqucnt Introduction: wc arc
allowcd to usc thc scqucnt n(A -' B) F LrA ' nB (Examplc
2.4.2 on page 82) lbr SI.what makcs this cxtcnsion spccial is
that wc do nol prove thc scqucnt first, belbre using it in SI
later. I{ather, making this sequcnt avallablc to oursclvcs is
what !!cts thc systcm K off thc ground. Equivalently, wc can
use thc scqucnt F u(A - B) - (LrA - lfB) in any application
of TI- Thc prmisc,/conclusion vcrsion is slightly morc convcnicnt, but thc thcorcm vrrsion is morc common in prcscn-

in semantically
definedsystems 103
53:Deducibiliry
"

'

tations of this kind of approach to modal dcduction. In thc


thcorcm vcrsion thc formula'n(A .- B)* (nA - trB)'is callcd
an dxiom to indicatc that it is simply assumedto be corrcct,
without independent proof (in tJrc way Euclid assumes his
axioms of geometry to be conect without indepcndcnt
proof). We will refer to our two sequentsas axiom-seque.nts,
and we use whichever of them leads to the quicker proof,
labeling the line L, SI (K)', or'TI (K)'as appropriate.

These four items define the deductive system K and the deducibility
rclation L-K,
which coincidcswith tso.Thc proofs of this and subsequent
claims about which deducibiliw rclations match which scmanticcolrsequcncc rclations a-repostponcd until Chaptcr 5. But we will illustrate
the proof systemswith sampledeductions,such as the follow:ingproof
of a basic scqucnt of modal logic. In it, we use 'NK' as a labcl for any
stcp justificd by a scqucnt of non-modal scntential logic which does
not havc its own namc.
Example3.3.1:Showtr(A & B) FKtrA & nB
1 (1)
(2)
(3)
(4)
1 (s)
(t)
(7)
(8)
1 (9)
1 (10)

D(A& B)
(A&B)-A
tr(A & B) - A)
tr(A & B) * trA
trA
(A&B)-B
tr(A & B) - B)
tr(A & B) - uB
aB
trA & DB

Prcmisc
TI (NK)
2 Ncc
:ISIK
4 1 - F

TI (NK)
6 Nec
TSIK
8,r -E
5,9 &I '

Line (4) is by the substitlrtion-instanccof the K axiom scquentobtained


by putting 'A & B' for 'A' and 'A' for 'B'.
Proofs likc this are arrived at by some combination of luck, expcrienceand insight, since thcre are few reliable heuristics for finding the
right non-modalthcoremsand axiom-sequentsubslitulion instanccsto
work witr. But despite its hard-to-work-with nat1ue,the system K as
describedin (i)-(iv)aboveis in fact sound and complete for Gen, and
other proof systems which i e sound and complete for the va.rious
semantic consequencerelations of $2 can be obtained simply by aug'
menting K with futher axiom sequentswhich calbc uscd in SL

l04 Chapter
3: First-Order-Definable
ModalSystems
. Thc systcm KT, T for short, is obtaincd by adding thc axiom-sequent
r- rA * A, equivalentlyaA F A, to K. So a proof in KT is the same as a
proof in K cxccpt that lines may be justified by appeal to SIusing !A F
A or F trA - A; such a line will be labeled'TI (T)' or 'j SIO)'. KT is sound
and completewith respect to Rf; that is, FKrand trRtcoincide,Here is a
samplc proof in KT; noticc that though line :l contains a modal opcrator, its justification is non-modal:wc usc a modal formula in a substitution-instanceof a sequentof non-modallogic.
F;<ample.
l)-3-2:ShowA FKTaA"
1 (1)
(2)
I (3)
1 (4)

A
tr-A - -A
-tr-A
.A

Premise
TI (T}

2,r sr(cMT)
3 Dfo

The system KB is obtaincd by addjng to K the axiom-sequent oEA F


A (equivalently, F o[A - A), which we call B. As the reader pcrhaps
anticipatcs, KB is sound and complete with respect to Sym; that is, FKB
and er- coincide. In KB we have thc following proof.
Example 3.:1.3:Show A FKBtroA.

1 (r)
(2)
1
1
I
1

(3)
(4)
(5)
(6)

A
on-A - -A

Premise

-o!-A
-^!-D*A

2,1 SI (CMT)
4 Dfo
5DN
a
6 Dfo

tr-tr-A
roA

n (B)

'A'
'-A'.
Linc 2 uses the axiom-sequentB, replacing with
The system K4 is obtained by adding to K either the axiom-sequcnt
ooA F oA (equivalently,F ooA .* oA), or else the sequent !A F !trA
(cquivalently,F trA - !!A), known as 4. Wc fcaturcd thc 'o'vcrsion as
Exuunplc3.2.:l(pagc95),but in thc proof systcm for Trn it is morc com'o'vcrmon to usc the'tr'version as the axiom sequentand derive thc
sion, which is the approachwe shall adopt. K4 is sound and complete
with respect to Trn; that is, FK4and trr," coincide-In K4 we have the following proof.

definedsystems l05
S3:Deducibilityin semantically

Example 3.3.4:Show ooA FK4oA


I
|

I
I

(1)
(2)
(3)
(4)

ooA
-!--!-A
!-A * n!-A
!-A .' --D-A

(s)
(6)

tr(n-A ' --tr-A)


trtr-A - tr--tr-A

(7)
(8)
(9)

!-A * tr--!-A
-E-A
oA

Prcmisc
1 Dtlo

Tr(4)
TI (NK)
4 Ncc
5 SI(K)
3,6 Sr (NK)
7,2 SI (tfl)
a
B Dfo

Line ti uses K, putting 'D-A' for 'A' and '--E-A' for 'B'.
From this point it is straightforward to obtain sound and complcte
systemsof proof lbr RS,RT and Eq, simply by adding axiom-sequents
tO KT:
. The system KTB, usually known simply as B, is KT plus the
. axiom sequent B. (KT)Bis sound and complete for Rs.
. The system KT4, usually known as 54, is KT plus thc axiom
sequent4. 54 is sormd and completefor RT.
. The system KTs, whcre 5 is the axiom sequent oA F troA, is
usually called 55. 55 is sound and completc for Eq.
It is a consequenceof thes results that tJIesubsystemdiagram of thc
previous section could be dupucatedhere, with the appropriate proof
systems replacing the semantic systems in that diagram. We understand the semantic inclusions as resu.ltingfrom a restdction in the
classof interprctations: a semanticsystem.t is a subsystcmof a semantic system S' when the interpretations for S' are a subset of those for
S. On the derivation side, one proof systemS' includes another S when
the rules of S are available in S', and when all the axiom sequcnts and
definitions of S are availablcin .t'; for propcr inclusion, some axiomsequent or mle or definition must be available in S' and not in S.
Finalfy, therc is the qucstion of whether an essentiaw second-order
semantic system can have a matching proof systcm similar to thosc
just described"The answeris that this is possible,and the system G is
an example-The proof system KL js obtainedby addjng to K the axiom
sequentL, which is tr(trA - A) F trA or equivalcntly F tr(lA - A) - trA,
bascd on Lijb's Formula 'tr(trA - A) - trA'. KL is sound and completc
for G-

106 Chapter
3: First-Order-Definable
ModalSystems
. Wc cnd this discussionof dcducibility in various systcmsby remarking that lie unwieldy nature of this kind of approach to prool mcans
that thcoretically significant scquents can require proofs which even
those on the forefront of developmentof modal logic have some difficulty discovcring.Thc problem of showing that 4 is derivablein KL is
an cxampleof this phcnomcnon,but thc following proof was evcntlal
ly discovcrcd(indcpcndcntly)by Kripkc, dc Jongh and S.filbin. In it wc
assumethc sequent tr&, FKL(nA & trB) - tr(A & B);we have ahcady (ir
eilect) proved one half of this; the otler is an exerciseExample.1.:1.5:
S}J.ow
FKL!A - trlA
(1)
(2)
(3)
(4)
(s)
(6)
(7)
(8)
(9)

A * [(!A & BDA) - (A & trA)] rl (NK)


tr{A & trA) - (!A & nnA)
Tl (K, tr&)
*
(A
A [!(A & DA)
& nA)l
1,2 sI (NK)
*
(A
D(A [D(A & DA)
& nA)l) 3 Nec
!A - ![tr(A & !A) ' (A & !A)l 4 sI (K)
D(tr(A & !A) * (A & rA)) - tr(A & nA) TI (L)
DA * !(A & !A)
s,6 SI (NK)
(trA
trA
& trtrA)
2,7 SI (NK)
*
trA nrA
8 SI (NK) +

Iine I is a substitution instanceof FNKA - (B & C) - (A & B)),and line


6 results from replacing'A'in Ldb's Formula with'A & trA'. There are
oths thcorctically significant scquents whose proofs are even morc
obscurc.
To end with a caveat,lest the dcvclopmcnt of this scction lcavc an
oversimplified pictu.rein the reader'smind- Not every consistent combination of first or sccond-orderconditions on R can be matched with
a rcasonable deductjon system like the systems we have discussed
hcrc- In thc othcr dircction, thcrc arc dcduction systcms which corrcspond to no collection of first-order conditions, such as KL, and which
detemine second-orderconditions which are much less manageable
than the'no infinite R-chain' condition for C. Most interestingly, there
are also deduction systems.t'which do not match any semantic conscquence rlation which can be dcfincd by first or sccond-ordcr conditions on l?. So if S is any scmantic system rclativc to which such an .9'
is sould, there will be some semanticallycorrect sequent in S which
cannot be proved in the deduction system;that is, S' must be incomplete with respect to .S.Thus the whole subject of generalrelationships

in semantically
definedsystems l07
53:Deducibility
likc lhcsc betweensemanticsystemsand proof systcmsis rathcr complicatcd; we make a start on it in Chapter 5.
t-JExercises
I Show the follow.ing (in thcsc problems you may not use the natural
deducdon rules for n and o from Chapter 2):
(1)
(2)
(3)
(4)
(5)
(6)
*(7)
(8)
(9)
*(l0)
(11)
(12)
{l:l)
*(14)
(15)
(16)
(17)
(18)
(19)
(20)

nA & trB FKtr(A & B)lHint use r** A - (B - (A & B))]


tr(A - B) F-Ktr(-g - -A)
-oA iF,( !-A
-aA -rFx o-A
oA, tr(A * B) F k oB
o(A & B) FK.A & cB
oA v oB iF Ko(A v B) (only left to right in solutions)
D(A - B) F* .A - rB
F K Bo n A . l o A
F-Kr(DA * B) v (!B - A)
FKBtr(A ' trB) .- (oA * B)
!(oA * B)rx, A - cB
FKBaouA - trA (KB4 is the system K plus axioms B and 4)
t-'<uoA * loA
FKB4
tr(tr4 - B) v tr(trB - A)
!(A v .B) F s4!A v oB
loA -' oDA Fs ao(A * trA)
Fs5.trA - trA
trA - DB Fs5D(trA ' trB)
Fss!A * trlA

U We have formulated the system 55 in two djfferent ways- Let us use


S5Ax to mean the axiomatic system of tlfs scction. Show that for any
formula p, if p is fully modalized then F,ro, p - rtp.lHtnt: distinguish
a spccial kind of fully modalized tbrmula, from which all other fully
modalized formulae are constructed ^Show that if p is a fi.tlly modalized
formula of this special kind, then Fs5a,p * !p. Then show that for each
fully modalized formula p not of the special kind, if the main subfor
mulas of p imply their olrn ncccssitations in sAx, Fs5^,p - trp-l
III Lct rrs usc SsND to mean the natural deduction syslem givcn in
Chapter 2, 54. Show dircctly (wifhout appeal to scmantics) that pr,...,p,
rrr o qiff pr,...,p, Fs5n"4. [Hint: given a derivation of 4 fiom Fv...,p"in
SsAx, describe how to transform it into a dcdvation of q frtm pt,...,p,

l08

Chapter3: First-Order'Definable
ModalSystems

iJr S5ND; thcn do thc convcrsc, dcscribing how to Oansform a dcriva


tion in 55ND into a dcrivation in SsAx- lhis dirccl.ionis hardcr. llsc thc
r < r s u l t so 1 ' p r o b l c m 2 . 4 . 1 Tp, a g c U l l , a n d o l p r o b l c m l l i r n m ( : d i a t c l y
abovc.l

You might also like