See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/268536956

The logic of ”actually”

Article  in  Reports on Mathematical Logic · January 1977

CITATIONS READS

113 1,107

2 authors:

John Newsome Crossley Lloyd Humberstone

Monash University (Australia) Monash University (Australia)
158 PUBLICATIONS   724 CITATIONS    144 PUBLICATIONS   2,234 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Early colonial Philippines View project

Mathematical Logic View project

All content following this page was uploaded by Lloyd Humberstone on 08 May 2016.

The user has requested enhancement of the downloaded file.

REPORTS ON MATHEMATICAL LOGIC
8 (1977), 11-29

John N. CROSSLEY and Lloyd HUMBERSTONE

Monash University, Clayton, Australia

1HE L O GIC OF "ACTUALLY"1

1. Introduction

In this paper we shall be looking at the logical uses of the expressions

'actually' and 'in fact' - between which we shall not be concerned to distin­
guish - where, by 'logical uses' we intend merely to exclude the speaker's
rhetorical use of such p1rases in dispelling or forestalling a misapprehension on
the part of the hearer, as when one says ".Actually it was March, not .April,
when we bought the house ... ", and so on. To introduce discussion of these logical
uses, let us cite an example with a familiar .Aristotelian flavour, namely, the
sentence "It is possible that a man who is sitting down should be standing".
The obvious ambiguity between senses may be paraphrased as:
(1) It is possible that some man be both sitting down and standing,
on the one hand, and:
(2) Some man is sitting down and it is possible that he be standing, on the other.
The ambiguity would be unlikely to cause misunderstanding since the charitable
hearer would rule out the first reading, it being contradictory; but it is inter­
esting to note that instead of the strategy of disambiguation followed here, in
which natural language contorts itself to expose directly distinctions of scope,
there is another simple way of forcing the non-contradictory reading. That is,
by changing our original sentence to "It is possible that a man who is actually
(or 'in fact') sitting down should be standing". This suggests the following idea:

Received .August 25, 1976.

1 The first author acknowledges valuable conversations with D. C. Makinson, C. Rauszer,

S. J. Surma and A. Wronski. He also acknowledges with pleasure the hospitality afforded him
by the following institutions and their members during the writing of this paper: University
of Warsaw, Technical University of Wroclaw, Jagiellonian University of Cracow, P olish Aca­
demy of Sciences and Ateneo de Manila University, Philippines. The second author would like
to thank M. K. Davies for stimulating conversations on the topics of this paper.
12

that the logical use of 'actually' is to clarify matters of relative scope when
a modal operator is involved. According to this idea it would be quite wrong
to introduce an 'actually' opera.tor into a semantically lucid formal language:
the logical function of 'actually' is to effect scope-disambiguations, indicating
which of two alternative translations into such a language is appropriate, and
not to figure, itself, as an element in either.
The trouble with this account is that there turn out to be occurrences of l
logical 'actually' which cannot be treated as mere scopeclues in this way, but
have instead to be regarded as occurrences of a genuine scope-bearing operator,
so that there is a point in considering modal languages into which an actuality
operator has been introduced alongside necessity and possibility opera.tors.
Presently, we shall have a look at a purely propositional such language and its
associated logic, but, to :motivate that discussion, we should give at least one
example of why the interpretation of 'actually' criticised above will not do -
and the example to be given is itself quantificational. For this example, we make
the enormously simplifying assumption that the semantics for our modal lan­
guage, in terms of possible worlds, is such that t,he domain of individuals does
not vary from world to world (as in Hughes & Cresswell's semantics for SS
(see [2] chapter 10)). With such a language in mind consider how we might
render the ambiguity of the sentence:
(3) It is possible for every red thing to be shiny.
or equivalently,
(3') It is possible that every red thing sl10uld be shiny.
The ambiguity in question is brought out by asking whether (3) asserts the
existence of a possible world in which every red object (i. e., every object red
in that world) is shiny, or instead the existence of a possible world in which all
the objects which are red in the actual world are shiny. These are quite clearly
distinct senses, the latter of which could be isolated by reformulating (3) as:
(4) It is possible that everything which is actually red, should be shiny.
Yet equally clearly, the ambiguity in question cannot be rendered by transla­
tion into our conventional modal language. For the materials at our disposal
in such a language allow us to assemble at most the following three putative
translations:
(5) O('v'x) (Red(x) ::::> Shiny(x))
(6) ('v'x)O ( Red(x) ::::> Shiny(x))
(7) ('v'x) (Red(x) '.) O Shiny(x))
Now, while (5) adequately captures the first sense of (3), neither (6) nor (7) gets
anywhere near capturing the second sense, since these allow us to pick, for each
object, a different world in which it is, if red, shiny, in the case of (6), and, in
the case of (7), for every red object, a different world in which it is shiny. Thus
they capture at best the senses of such sentences as 'It is possible for any red
object to be shiny'. Wbat we want for the 'missing' sense of (3), however, is
 13

a single world in which every object red in our world is shiny: an 'every' reading,
rather than an 'any' reading. So the universal quantifier must appear within,
and not outside, the scope of the possibility operator. But the semantics for
conventional modal languages dictates that the truth-value of a 'possibly'
formula is determined for a given world exclusively by the truth-value of the
immediate subformula involved at some world possible relative to the given
world, and this itse:f is determined by how things are in the new world and in
yet further worlds possible relative to it. VVhat we require to capture the missing
sense is some device for referring back to how things are in the original world:
in the possible world envisaged in (4) the objects which are to be shiny are those
objects which, in our world, are red. The semantics to be given in the next
section for tbe propositional fragment of a modal language containing an
actuality operator are such as to give this back-referring function to the new
operator. Writing 'A' for 'it is actually the case that', the idea is that in the
enriched language, the missing sense of (3) is to be captured by a more-or-less
direct translation of (4):

(8) (Vx)[A(Red(x)):) Shiny(x)]

Hopefully, consideration of this example, which is only one of several sorts

that can be constructed, exhibits the expressive weakness of the conventional
languages in a rather striking way and thereby motivates the exploratfon of
an 'actually' enriched language, at the same time as disposing of the idea that
'actually' functions as no more than a scope-disambiguating clue in the process
of translating English into one of the usual modal languages.

2. A System of Modal Actuality Logic

Having established the case for an actuality operator, what are we going
to do about it? On the walk-before-you-run principle, we can make some fairly
obvious - at least, obvious in the light of the discussion so far - suggestions
about modal propositional logic with 'actually', and again in the interests of
simplicity, let us think of our underlying modal logic as SS. It will be recalled
that a semantics for S5 can be given in terms of what is sometimes called a model
W = ( W, V), where W is a non-empty set (of possible worlds) and V is a function
which assigns to each propositional variable and world a truth-value. V is then
extended to all formulae in the following way, where we write W l=wa for
'V(a, w ) = T':
(i) W l=w,...., a if and only if not W l=wa,
(ii) W l=wa :> /3 if and only if either not W l=wa or W l=w/3,
(iii) W l=wOa if and only if for all w' € w, W l=w,a.
(We may take possibility, and the remaining truth-functional connectives
as defined in terms of negation, implication, and necessity in any of the usual
14

ways). Now a formula of SS is valid if there is no world in any model constructed

in accordance with (i)-(iii) at which it is false. We may proceed to a semantics
for - let's call it 'SSA' our modal actuality logic, by revising the concept
-

of a model so as to take into account the idea of one world being singled out as
the actual world. In other words, let us say that a model, W, is to be a triple
(W, w*, V), where w* E W (we call w* the 'actual world' of the model). vVe@
extended V to all formulae by adding to (i)-(iii) above
(iv) W l=wAa if and only if W l=w* a.

.As usual, a formula is valid just in case it is assigned the truth-value Tin every
world in every model of SSA.
By obvious analogy with Kripke's use of the phrase 'rigid designator',
we could say that SSA, as here defined, gives 'actually' the logic of a 'rigid
operator'. Indeed, if one chooses to regard sentences as designating their truth­
-values, (iv) tells us that '.A.a' rigidly designates that truth-value that 'a' de­
signates in the actual world; so the parallel with proper names and rigidly used
descriptions is quite close. .Another general point worth making is that SSA pre­
serves for 'A' a property which its other non-truth-functional operators have,
namely, that formulae introduced by these operators do not change their truth­
-values - as purely truth-functional formulae may - from world to world. This
is in striking contrast, of course, to the situation which obtains in modal logics
weaker than SS, and constitutes a large part of our motive for consideiing the
'actually' extension of 85.
SSA can be axiomatised by subjoining to some axiomatisation of S5 with
necessitation as a rule of proof the following five axioms:
(.Al) A(Ap'.Jp)
(.A2) A (p '.J q) '.J (Ap '.J Aq)
(.A3) Ap = �A �p
(.A4) Op '.J Ap
(.A5) Ap '.J OAp
This axiom set brings out rather well the contrast between our own approach
and that of some others (e.g., Kaplan's treatment of 'actually' in [4], and
Kamp's treatment of 'now' in [3]), on which the analogue of 'Ap = p' is a theo­
rem. Evidently, in such approaches, the substitutivity of provable equivalents
fails to hold, so providing a contrast with the usual modal logics. In our system1
however, no such deviations are required: necessitation holds unrestrictedly
as a rule of proof, and one can show that provably equivalent formulae are
intersubstitutable salva provabilitate. The closest we get to 'Ap = p' as a theorem
is in 'A(Ap = p)' which is easily seen to be valid on the semantics given. No
doubt the idea that the former equivalence should be a theorem stems from the
observation that the outright assertion of '.Actually p' (or 'Now p') is tantamount
to the assertion of the simple 'p', but it is far from clear that any such relation
 15

a.s this deserves promotion to the ra.nk of 'va.lid equiva.lence'. Such promotion
a.rgua.bly involves a. confusion between the idea. of logica.l truth or logica.l neces­
sity, the pre-formal correla.te of va.lidity, on the one ha.nd, a.nd wha.t, in a. na.tura.l
extension of the terminology introduced by Ca.sta.iieda. in [1], ma.y be ca.lled
'la.lie necessity', on the other. No doubt such sentences a.s 'I a.m here' a.re la.lica.lly
necessa.ry in the sense tha.t they ca.nnot be falsely a.sserted, but to sa.y this is
certa.inly not to sa.y tha.t they cannot be false, a.nd some arguing would need to
be done by a.nyone proposing to a.ccord va.lidity to such contingencies. At a.ny
ra.te, wha.tever the worth of these considera.tions, it is of interest to explore
a.lterna.tives to these other a.pproa.ches, to see wha.t they look like when de­
veloped.
For a.11 this, one could of course introduce a distinction between, let us ca.11 it
general validity, the property tha.t a. formula has if it can be falsified at no world
in any SSA- model, and real- world validity, that property a formula has if it
can be falsified at no world designated the actual world in any SSA- model.
General validity is what our set of axioms is designed to capture, but we can
make a few points concerning real-world validity while we are on the subject.
In the first place, it is obvious that if a contains no occurrences of 'A', then a is
generally valid if, and only if, a is real-world valid. In the second place, a for­
mula f3 is real-world valid if, and only if, A/3 is generally valid.
The completeness of our axioms relative to this notion of general validity
can be shown by an adaptation of the completeness proof for propositional SS
given by Hughes & Cresswell (op. cit., 116-121) 2• Their proof adapts to the
case of SSA without any difficulties because our axioms suffice for the reduction
of any formula to a (:modified modal conjunctive) normal form, in the sense
that any formula is provably equivalent in SSA to a conjunction of formulae
of the form:
/3 V Dr1v...v Drn v Oov As ,
where the only occurrences of O, O, and A are those exhibited 3 Such a con­ •

junct is valid if and only if one of the following disjunctions is valid:

(c'lV/3), (c'Jvs), (c'JVy1), , (oYyn). The reasoning leading to this conclusion,
•••

a.nd from it to the corollary that SSA's axioms are complete, is that given
by Hughes& Cresswell for the case of S5, with a few obvious and minor changes.
(For instance, for the 'only if' part, we argue that if no such disjunction is valid
then we can construct for the conjunct in question a falsifying SSA model
(W, w*, V), where W= {w*, w0 , w11 , Wn} in which V (ovs , w*) =1!',
•••

V(ov{3, w0)=1!', and V(oVyi, wi)=J!' for l�i�n. Evidently, V (a, w0)=1!',
where a is the conjunct in question.)

2 This observation is due to M. K. Davies.

3 That these rather simple normal forms are available for SSA follows from our normal form
theorem (Theorem 21) for SSAR, below, in Section 5, together with the fact that A Oa and Da
are provably equivalent in SSA, as are .A(>a and (>a. (Note that both halves of both equi­
valences require SSA's axiom .A.4, which is missing from the axioms of SSAR.)
16

3. Questions of Redundancy

�Vl/e introduced the idea of an actuality operator as a non-redundant addition

to the vocabulary of modal languages by means of an example which involved
quantification in an essential way, which leaves open the possibility that for
the propositional fragment there is some kind of redundancy: a question t�
which we now devote some attention. There is an obvious distinction between
two ways in which an operator (or any other symbol) may be said to be redun­
dant in a language from the point of view of a logic. First, it could be directly
redundant, in the sense that the logic proves equivalent any formula and that
formula got from it by simply deleting all occurrences of the operator. This,
for instance, would be the case with a truth operator in a two-valued logic. On
the other hand, it could be indirectly redundant, in the sense that for any for­
mula of the language in which the operator occurs, there is some provably equi­
valent formula in which it does not. It is obvious that 'A' is not directly redun­
dant in SSA, as the invalidity, on our semantics, of 'Ap = p' shows. Neither
is 'A' indirectly redundant in SSA: provably equivalent formulae have the same
truth-value in every world, and how could 'Ap', for instance, have the same
truth-value in each non-actual world, as some A - free formula� But these no­
tions of non-redundancy are rather different from that involved in our earlier
quantificational example, for while they rely on failures of equivalence in non­
actual worlds, in the original example (about the red objects being shiny) we
showed what amounts to this: that even restricting our attention to actual, or
designated worlds, we could find for the formula '(Vx)[A(Red(x)) ::i Shiny(x)]',
no A-free equivalent. Such a restriction of attention gives a doubling up of our
distinction between direct and indirect redundancy. We introduced that di­
stinction on the assumption that the A- free formula to be found was to be
provably equivalent to the original formula: in other words, insisting that the
equivalence in question had to be, in our earlier terminology, generally valid.
V\Te may now consider the we8Jker requirement that the A- containing and the
A-free formulae be only real-world equivalent, i. e., that the equivalence between
them be only real-world valid. What this comes to, bearing in mind certain
other facts already mentioned, is, for the special case of indirect redundancy,
the following: 'A' is indirectly real-world redundant if any only if for any
formula a there is an A-free formula a' such that SSA I- A(a = a' ) .
Such a notion as that just introduced may strike the reader as somewhat
artificial as a notion of redundancy. But it might commend itself as a genuine
and interesting notion of redundancy in view of some such considerations as
the following. Being true, not in this world or that world, but being true tout
court, amounts to being true in the actual world. A symbol would surely be
a gratuitous addition to our language if whatever was true tout court could be
said without its aid. This is all very impressionistic, of course, and there is no
need to enter into the issue of relative versus absolute notions of truth here.
"Whatever the weight of such remarks may be, the point is this: it turns out
 17

(the proof being supplied in Section 5) 4 that the actuality operator of propo­
sitional SSA turns out to be, in the sense just given for the phrases, indirectly
real-world redundant. (It is not, by the way, directly real-world redundant:
we could not, for instance, prove ' OAp' real-world equivalent to 'Op'; i.e., we
do not have 'A( OAp = Op)' as a theorem in SSA.) The upshot of this discussion
of redundancy, then, is that an 'actually' operator of the sort we have been
looking at is more important, from the point of view of expressive power, as an
addition to a quantified, than to a purely propositional, alethic modal language.

4. Global Operators

The last topic to be covered in our informal discussion of SSA concerns not
so much the system itself, as the possibility of extending it in such a way as to
meet a likely objection. The anticipated objection is to our fifth axiom
'Ap ::> OAp' , which may indeed, if read literally, strike one as counter-intuitive.
It seems unlikely that objections would be raised to any of the other axioms,
but this one may be so unpalatable to some readers that they rather doubt the
value of SSA as even a candidate for a plausible modal actuality logic. Its validity
arises from the fact that it is one and the same world that counts as the actual
world, for every world in the model. So if something is true in that world it. is
certainly going to be true in every other world that it is true in that world. It
would appear, then, that there is no simple tinkering with our semantical
framework that would leave us with .Al-.A4 valid, while invalidating .A5, keeping,
of course, the whole of SS itself intact. It is, perhaps, somewhat startling that
such straightforward assumptions about necessity as are embodied in SS, when
combined with some equally simple-minded ideas about 'actually', lead us to
accept the offending axiom as valid. (In fact, we do not wish to commit ourselves
to saying that ther� is anything unacceptable about .A5, - only to provide
food for thought for those who have made up their minds against it.)
Faced with this situation, in which independently quite nice ideas about
'necessarily' conflict with independently quite nice ideas about 'actually' when
the two concepts are brougl1t into relation, many would no doubt say that the
'
less nice ideas are those about this young upstart, 'actually', and it is therefore
these tlmt should be scrapped if we want to get an acceptable modal actuality
logic. Since this would amount to infanticide, or at any rate to a culpable
reluctance to persevere, we make precisely the contrary proposal, wanting
a novel account of necessity which will mesh well with the account already
provided of 'actually'.

4 In fact this is proved there (as Theorem 18) for a proper subsystem of SSA which we call
SSAR, from which it fol lows a fortiori that the result holds for SSA.

2 - Repoirts on Mathematic•al Logic 8

18

To give this new account, we introduce the idea of what may be called global
operators. .A rough idea is given by saying that global operators stand to modal
operators in much the same relation as modal operators stand to truth-functional
operators. More precisely, we observe, in notations like that used in (i)-(iv)
above, that both the model superscript, and the world subscript, on the truth-at
turnstile lie dormant in the statement of the truth-conditions for truth-functional
compounds. The semantics for modal operators, on the other hand, requires
that the world variable come to life - in general that it be either quantified
or have some function applied to it (as, e.g., a successor function on the 'moment
of time' variable in giving the semantics of'next' in a tense logic). .Analogously,
we may define a global operator as one whose semantics bring the model place
itself to life 5• In particular, consider the relation 'is a variant of' on SSA models
holding between any two such models differing at most in what world is actual,
and in any consequences this may have (regarding the .A- containing formulae)
for V. In terms of this relation, which we symbolise '� ', we define what it is
for a formula a to be fixedly true (represented '.:F') at a world in a model:
(v) W Fw .:Fa if and only if for all W ' � W , W ' t=wa .
It is not hard to see how such an operator might come to have a role in the
representation of necessity, since to say that something is necessarily true comes
pretty close to saying that it would be true no matter which (possible) world
was the actual world. Variance between models being an equivalence relation,
we shall obviously have, as valid for '.:F', everything valid for' D' in SS. Whether
or not we have anything else will depend on a choice we must now make as to
the interpretation of propositional variables (a choice, it might be mentioned,
of a sort that regularly arises when one pursues the logic of global operators).
On the one hand, we could choose to read them in such a way that
(9) p:J.:Fp
comes out valid. This is not an unnatural reading, since we might be thinking
along the following lines: any propositional variable assigned truth at a world
in a model will clearly be assigned truth at that world in every variant model,
since it is only the truth-value of .A- containing formulae that is affected by
passage over to new variants. But the trouble with this reading is that uniform
substitution fails to preserve validity in the resultant logic, since
(10) .Ap :J .:F .Ap

is a substitution instance of (9), and (10) is invalid. Indeed, the whole point of
introducing the 'fixedly' operator is to invalidate (10), the analogue of our
originally controversial formula (.Axiom 5) relating necessity and actuality.
.Anyway, another way of reading propositional variables is to think of them as
bound by invisible outermost universal propositional quantifiers, ranging over

5 Montague gives a 'global' account of modal operators in [7].

 19

all propositions, or, to think of the matter substitutionally over all formulae of
the language. Trivially, such a reading enables uniform substitution to preserve
validity, and (9) turns out invalid, since not every formula implies its own
'fixation' (witness (10)). It does not much matter what we choose to do, since
the problem of the validity of something like (9) will come up in the end, say,
if we think of introducing our 'fixedly' operator into an applied first-order
language, whereupon we shall find such formulae as

(11) Red(a) :i 3' (Red(a))

coming out valid.

Indeed, the validity of (11) in such a language shows that '3'' by itself does
not represent necessity. But recall the idea that what is necessary is what would
be actual no matter which world were the actual world- anidea which suggests
that it is not '3'', but rather'3'A' that we should take as representing necessity
in this framework. If we look at the two axioms relating necessity and actua;lity
in our original formulation, we see that replacing' D' by':FA' in both has the
effect of changing the (arguably) dubious formula into a patent invalidity, while
keeping the originally acceptable formula valid, though now on somewhat
different grounds. If we wish to compare the two notions of necessity more
directly, in a language containing both (and whose semantics are given by
(i)-(v)), we find that they are equivalent for A -free formulae, though in
general 3'Aa implies Da but not conversely. We may conclude from this that
while either suffices for the representation of necessity in the familiar contexts,
when actuality is not involved, the globally defined operator, or something like
it, ought (if one objects to A5) to be brought in when actuality is under con­
sideration. More generally, the fact that any such avenues are open to us at all
shows that we should not be too ready to give in to the argument that any
world-designating treatment of 'actually' is doomed to failure because it
inevitably leads to the validation of 'If actually p, the necessarily actually p'.
We will not pause, however, to explore the intricacies of the global actuality
logic whose leading ideas have just been sketched, going on instead to fulfil
some promises made about the formalities of SS A , via the investigation of
a related, and from a technical point of view more interesting, system of modal
actuality logic.

5. Completeness and Decidability of a Modified Logic for 'Actually'

There are various directions in which we could extend our investigations of

the logic of 'actually'. Obviously, the addition of quantifiers to the language
would bring new and interesting problems. The exploration of'actually'-enriched
versions of modal logics weaker than SS, for example S4 and T, would also be
worthwhile. However, we choose instead to discuss a system which is perhaps
20

best described as an alternative actuality logic based on SS. The reason for this
is that the completeness problem for this alternative system is not so easily
settled as for SSA, in the absence of a correspondingly tidy normal form theorem.
(It will be seen from Theorem 21 below that the normal forms we eventually
get out cannot be guaranteed to be 'first-degree' formulae, whereas for SSA,
they are.) Hopefully, then, our discussion of this new system should provide
a good basis for any subsequent treatment of the 'actually' extensions of modal
logics weaker than SS, as well as yielding the results for SSA itself promised in
notes 3 and 4.
To see that there is room for two alternative S5-based actuality logics, recall
that the semantics for SS can be given in either of two ways. In Section 2,
above, we invoked that version of the semantics in which models are taken as
pairs (W , V), and by adding to the attendant clauses (i), (ii), and (iii), the
obvious clause for 'actually' (iv), we arrived at our semantics for SSA. The
other version of the semantics of SS, however, takes as its models triples of the
form (W, R, V) where Wand V are as before and R is some equivalence relation
on W, which enters into the clause for D (clauses (i) and (ii) remaining un"
changed);
(iii)' W l=w Oa if and only if for all w' such that wRw', W l=w,a •

.Although, as is well known, the two semantical accounts yield the same set
of valid formulae within the familiar modal language, they cease to do so when
we add a 'w*' to our models along with clause (iv), to deal with formulae con­
taining 'A'. That is, if we take wbat we may call a.n S5AR-model to be a structure
(W , w*, R, V) where w* e W, R is an equivalence relation on W, and V is a func�
tion from pairs of propositional variables and elements of W to truth-values,
extended to all formul�e in accordance with (i), (ii), (iii)', and (iv), and we define
SSAR-validity as unfalsifiability at any world in any SSAR-model, then we
find that the set of SSAR-valid formulae does not coincide with the set of
SSA-valid formulae. Our axfom .A4 of Section 2 above, for instance, is SSA-but
not SSAR-valid. However, if we delete that axiom from our basis for SSA.
and add a rule of 'actualization' ('.Act.', for short), analogous to necessitation,
we obtain a sound and complete axiomatisation of the SSAR-valid formulae.
In this section, then, we shall show, amongst other things, that the addition of
the following to any basis for SS with necessitation as a rule of proof gives an
axiomatisation of the SSAR-valid formulae:
(.Al) A(Ap '.) p)
(.A2) A(p :::i q) '.) (Ap '.) Aq)
(.A3) Ap = ,..,_A -p
(.A4) Ap :::i OAp
(.Act.) If a is a thesis, then .A.a is a thesis.

From this point on, we use '.A4' for the fourth axiom here, and not as in Section 2.
 21

Our strategy will be to give a completeness proof for S5AR in the style of
Makinson [5], and then a decision procedure based on a normal form theorem
in the style of Hughes& Cresswell [2]. The normal form theorem also yields an
alternative completeness proof.

Lemma 1. The axioms of S5AR are valid.

Proof.
(.Al) ?ID l#wA(Ap '.) p) if and only if ?ID l#w•AP '.) p if and only if (?ID l=w*AP
and ?ID l#w•P) if and only if (Ill l=w*P and ?ID 17"w•P)· Hence (.Al) is valid.
(.A2) ?ID 17"wA(p '.) q) '.) (Ap '.) Aq) if and only if ?ID l=w.(P '.) q) and ?ID 17"w(Ap '.) Aq)
if and only if (?ID l=w•P '.) q, ?ID l=w•P and ?ID 17"w•q). Hence (.A2) is valid.
(.A3) ?ID l=wAP if and only if ID} l=w•P if and only if ID3 17"w• "'p if and only if
?ID 17"wA "' p if and only if ?ID l=w "'A"" p.
(.A4) Suppose that for some ?ID= (W , w*, R, V) and some w E W, ?ID l=wAP
then ?ID l=w,Ap for all w' E W, hence in particular ?ID l=w,AP for all w' E W such
that wRw'. Hence ?ID l=w OAp. If no such w E W exists then trivially
?ID I= Ap '.) OAp. Hence (.A4) is valid.

Lemma 2. If a is valid then Aa is valid.

Proof. If a is valid then V(a, w)= T for all w and in particular for w = w*.

Theorem 3. (a) All theses of S5AR are valid.

(b) S5AR is consistent.

Proof. (a) by induction on the length of proofs.

(b) The formula "'(p '.) p) is not valid.

.An alternative proof of consistency may be given as follows. Suppose we

have a proof of a contra.diction in S5AR, a1, ... , an . Delete all occurrences of A
and modal operators, then all the ai become formulae of propositional calculus
and (subject to adding proofs of formulae of the form {J '.) {J), a0 . . , an becomes
.

a. proof of a contradiction in propositional calculus which is impossible.

We now observe that by (.A3) A commutes with "'.

Lemma 4. S5AR l- (Ap '.J Aq) '.) A(p '.J q).

Proof. 1. "'A(p '.) q) .Assumption

2. "'(p '.) q) '.) p Tautology
3. A( "'(p '.) q) '.J p) (.Act.) on 2
4. A( "'(P:Jq) :J p):J(A,..., (p :J q) '.) Ap) (.A2) and substitution
5. A "'(p:Jq) '.) Ap modus ponens
6. "'A(p'.Jq):JAp remark preceding this
lelnlna
22

7. Ap
8. "'(p'.)q)'.)"'q Tautology
9. A"' (p:) q)'.) A"'q cf. 3, 4, 5
10. "'Aq cf. 6, 7
11. Ap'.)("'Aq'.)"'(.Ap'.)Aq)) Tautology
12. "'(Ap'.)Aq) modus ponens 7, 10, 11.
Hence the result by taking contrapositives.

Lemma 5. A commutes with all propositional connectives.

Proof. Immediate by induction using Lemma 4, the remark preceding it

and Axiom (A2).

Lemma 6. (i) S5AR I-Ap = A.Ap; (ii) S5AR I- .A (p = Ap).

Proof. 1. A (Ap'.)p) (Al)
2. A (A"'p'.)"'p) Substitution
3. A("'Ap'.) "'P) Lemma 5
4. A(p'.) Ap) Lemma 5 and propositional calculus
5. Ap'.)A.Ap (A.2)
6. A.Ap'.)Ap 1, (A2) and modus ponens.

We shall see below that A does not commute with D (or O). As usual we
define a Lindenbaum set to be a maximal consistent set of formulae (of SSAR).

Lemma 7. Every consistent set of formulae can be extended to a maximal

consistent set. Moreover, if A is a Lindenbaum set, then
(i) every thesis of SSAR is in A ,
(ii) A is closed under modus ponens,
(iii) av {3 is in A if and only if a is in A or {3 is in A.
(iv) for every formitla a exactly one of a and "'a is in A.

Proof. As usual, see e.g. Mendelson [6].

Lemma 8. If A is a Lindenbaum set then A* = {a: A.a EA} is a Lindenbaum set.

Proof. Suppose a¢ A* then A.a¢ A. Hence "'Aa EA. Now "' commutes
with .A, so by Lemma 7(i) and (ii) A"' a EA. Hence "'a EA*. So either a or
"'a EA*.
Now suppose A* is not consistent. Then for some au ... , an EA*, SSAR l­
a1 '.) ( a2'.)...'.) ,..., (p '.) p) .. .) .By (Act.) and (A.2) we obtain SSAR l-Aa1'.)(A.a�'.) . . . '.)
'.) "' (.Ap '.) Ap) ... ) . But Aau ... , Aan EA so A is inconsistent, which is a contra­
diction. Thus A* is a Lindenbaum set.

In order to carry through the completeness proof we now need to prove

 23

Lemma 9. Let A be a Lindenbaum set. Let a be any formula such that Da€A. ,...,

Let Aa= {Aa: Aa€LI} u {,...,a} u {{3: D/3€LI}. Then Aa is consistent.

Proof. Suppose not; then there are formulae Aau ... ,Aam€A, {J1,... ,f3n
such that D/311 ... , Df3n €A and S5AR f- Aa1'.) (... '.) (Aam'.)({J1 '.) ... '.) (f3n '.)
'.)a) ... ) ) . .. ) . By necessitation and the S5 thesis D(y '.)a)'.) ( Dy'.) Do) we
obtain by repeated use of modus ponens
SSAR f- OAa1 '.J (... '.J (OAam '.J ( Df31 '.J . '.J (Df3n '.J Oa) ... )) . .. ) .
..

By .Axiom (.A4) and modus ponens using Lemma 7, if Aai €A,OAai €A. Further
0{31,... , DfJn €A. Hence by Lemma 7, Oa€A. This contradicts the con­
sistency of A.

Corollary. If A,a are as in the lemma there is a Lindenbaum set At 2 A.

Now we can proceed to the completeness proof. Let K be a consistent set of

formulae. Extend K to a Lindenbaum set A0• Let w* = A� (see Lemma 8).
Define W inductively by
(i) w* EW
(ii) A0 €W
(iii) if A is a Lindenbaum set in W and a is such that Oa EA then A; EW ,...,

where At is the Lindenbaum set given by Corollary 8 containing Aa.

Since A is countable, W is countable. Now define a relation R on W by

setting ARA' if and only if for every formula a, if Oa EA then a EA'. Let
V(p , A)= T if and only if p EA for each propositional letter p. Let
W= (W,w*,R,V).

Lemma 11. If a EA� then Aa EA for all A EW.

Proof. If a EA� then Aa EA0 by Lemma 8. By (A . 4) and Lemma 7,

OAa, DOAa,. .. are all in A0 • .Also AAa EA by Lemma 6 and Lemma 7.
Hence ..Aa EA� and hence OAa, DOAa, . .. are all in A�= w*. It follows now
by induction from the definition of W that Aa, DAa, DOAa, ... are in A for
all A €W.

Lemma 12. For each formula a, V(a, A)= T if and only if a€A.

Proof. By induction on the complexity of a. For propositional letters, ,...., , '.)

and D the proof is as usual. Now suppose a= A{3.
A{3 EA implies {3 Ew* since otherwise {3 Ew* and A{J EA by Lemma 11
,..., ,...,

and Lemma 7. By the induction hypothesis it follows that V({3,w*)= T, hence

V(A{3, A)= T by the definition of V.
Conversely, if V (A{3, A) = T then V ({3,w*) = T by definition of V, hence
{3 Ew* by the induction hypothesis and A/3 EA by definition of W.
24

We now have that if K is a consistent set of formulae then W is a model

for S5AR such that V(a, A0) T for all a € K. Thus we have proved:
=

Theorem 13. (Completeness of S5AR) If K is a consistent set of formulae

of SSAR then K has a model.

Corollary 14. a is a thesis of SSAR if and only if a is valid in every S51\R

model.

vve now give some examples of non-theses of SSAR.

Example 1. OAp) Op. Let W = {w* , w}, V(p, w*) = T, V (p , w) = If
and let wRw and w* Rw* be the only places where R holds. Then if
W (W, w*, R, V) , W loFw DP bub W l=w OAp. Hence OAp) Op is not a thesis
=

of S5AR by Theorem 13.

Example 2. OAp) A Op is not a thesis of S5AR. For consider the structure
W ({w* , w} , w*, R, V) where R {w*, w}2, i. e. R holds everywhere and
= =

V (p , w*) = T, V (p , w) = If. Then for u= w or w*, W l=uAp, hence W l=w DAp

but W /oFw• Op.
It follows from this latter example that A does not commute with o.
ExampJe 3. A Op) DP is not a thesis of S5AR. Let W = ({w*, w}, w*, R, V)
where only w* Rw* and wRw hold, V(p , w*) T and V(p, w) = If. Then
=

W l=wA Op but W /oFw Op. (Notice that, in contrast with our first two examples,
this formula, A Op) Op, is a thesis of SSA.)

Lemma 15. (i) SSAR I- OAp = Ap; (ii) S5AR I- <)Ap = Ap.

Proof. (i) .Ap) OAp (.A4)

Oq) q SS
OAp ) Ap Substitution.
(ii) <)Ap) ,.., D ,..,A p
) ,.., DA"' p (.A3)
) "'A"' p by (i)
) Ap (.A3)
) OAp by (i)
) <)Ap S5.

This lemma and the examples above will ultimately show that the normal
form we now set out to establish is irreducible in an obvious sense.
First we shall establish the indirect redundancy of A in S5AR (see section 3
above).

Lemma 16. (i) SSAR 1- A( DAa = a); (ii) S5AR I- A( <)Aa = a).

Proof of (i) OAa = Aa Lemma 15 (i)

A( OAa :sAa) (.Act. )
AOAa = AAa Lemma 5
 25

AAa = Aa L�mma 6(i)

AOAa = Aa propositional calculus
A( OAa = a) Lemma 5.
For (ii), replace D by <> in the proof of (i).

Lemma 17. (i) S5AR I-A( O(avAp) = Oavp),

(ii) SSAR I- A ( 0( avAp) = Oavp),
(iii) SSAR I- O(.Aavp) = Aav op,
(iv) S5ARI- O(Aavp) = Aavop.

Proof. (i) A( O(avAP))

A( O(av OAP)) Lemma 15 (i)
A( DavOA,0) S5
AOaVAOAP Lemma 5
AOavAp Lemma 15 (i), Lemma 6 (i)
A(OaVp) Lemma 5 .
(ii) Similarly.
(iii) O(AaVP) = D(OAavp) Lemma 15 (i)
= O Aav DP SS
= AaV DP Lemma 15 (i).
(iv) Similarly.

Theorem 18. There is an effective proceditre which assigns to each formitla a

a formitla a' not containing A such that SSARI- A(a= a').

Proof. We first put a into a form containing only the connectives ,...,, &, v,
O, <> and A. Next we move the negation signs as far into the formula as possible
using S5I- p= ,..., p, de Morgan's laws, the modal laws S5 I-
,...,
DP = <>,..., p
,...,

and SSI- <>P = D ,..., p and Lemma 5. In the formula so obtained, negation
,...,

signs can occur only immediately preceding propositional letters. Now we use
the modal reduction laws SS 1- <>P= D OP, SS 1- DP = O DP, S5 1- OP = OOP
and SS I- DP= D DP to reduce all iterated modalities to single modal operators.
Now we proceed by induction on the construction of the formula (which for
convenience we shall again call a) so obtained.
Case 1. If a is a propositional letter preceded by zero or more negation
signs, let a' = a.
Case 2. If a is of the form AP let a' = p then by Lemma 6, SSAR I- A (a = a').
Case 3.1. If a is of the forni op.
Subcase 3.11. If p is of the form ,..., y then y is a propositional variable so a
does not contain A, so let a' = a.
Subcase 3.12. If p is of the form Ay then by Lem:ma 16, S5ARI- A(D Ay = y)
so let a'= y'. By Lemma 5, S5ARI- A(a = y').
Subcase 3.13. p is of the form y& b. Then SSAR I- DP = Dy& Ob. By the
induction hypothesis there exist formulae y', b', not containing A such that
26

SSAR I- A ( Dy = y') and SSAR I- A ( Do � o'). Hence by Lemma 5, we easily

obtain SSAR 1- A( D,8 = y'& o'). So let a'= y'& o'.
Subcase 3.14. If ,8 is a disjunction of the form y v o.
Subcase 3.141. If y, o are possibly negated propositional letters, let a' = yVo.
Subcase 3.142. If either y or o is a conjunction, say y = e&C, then replace
yVo by (eva)&(Cvo). Now using the induction hypothesis, proceed ::iis
in 3.13.
Subcase 3.143. If either y or o begins with a modal operator then we use the
laws S5 I- D(yv De) = Dyv De or SSI- O(yv <)e) = DyV <)e and then use
the induction hypothesis.
Subcase 3.144. If y is of the form Ae then we let a' = e'v Do', then
by Lemma 17(i), S5AR 1-A(a' = D(Aevo)). Similarly if o is of the form Ae.
There are no other subcases, as a of the form D,8, where ,8 begins with a modal
operator, was eliminated in our initial description of a.
Case 3.2. If a is of the form <),8 we proceed dually to 3.1, replacing D by <)
and & by v and vice versa. This completes the proof.

Corollary 19.
There is an effective proceditre which assigns to each formula a
a formiila a' not containing A such that S5AR I- Aa = Aa'.
'Ne now proceed to our normal form theorem.
Given a formula a we perform the following reductions on it. By virtue of
Corollary 19 above and theorems of 85 it will follow at each stage that the
formula a* obtained is such that S5AR f- a = a*.
Step 1. \Vrite a so that its only connectives are , &, v, D, <) and A.
,...,

Step 2. Use the laws of negation and modalities and Lemma 5 to move
negation signs inwards as far as possible, so that they only precede propositional
letters and at the same time cancel double negations.
Step 3. For any well-formed part of a which is of the form A,8 and is not
properly contained in any other well formed part of a beginning with A use
Corollary 19 to obtain a formula ,B' such that S5AR I- A,8 = A,8' where ,B' does
not contain A. Replace A,8 by A,B'. The formula now obtained contains no
nested occurrences of A .
Step 4. Reduce all iterated modalities to single modal operators by the
reduction laws of the modal logic SS.
Step 5. If the combination DA or <)A occurs in a replace the combination
by A. The resulting formula is still equivalent to the original one by Lemma 15.
Step 6. If the resulting formula still contains a modal operator with an occur­
rence of a modal operator or A in its scope, apply step 6.1 or 6. 2 according as
a nested part is of the form D,8 or <),8.
Step 6.11. D,8 is of the form D(y& o). Then replace this by Dy& Do.
Then the resulting formula is still provably equivalent to a in S5AR.
Step 6.12. D,8 is of the form D(yVOo) where O is D or <); replace this
by OyvOo. Similarly for D(Oyvo). The resulting formula is still provably
equivalent to a in S5AR.
 27

Step 6.13. 0(3 is of the form (.AyVo). Then replace this part by .Ayv Do.
Similarly for (yv.Ao). Since SSAR f- D(.Ayvo) = D.Ayv Do = .AyvD o the
resulting formula is still provably equivalent to a in SSAR.
Step 6.14. DP is of the form yva where neither y nor a begins with .A, D
or <). Since yvo contains a modal operator either y or a must be a conjunction.
Suppose y= s&C then S5AR I- DP= D{(s&C)vb) = D{(svo)&(Cvo)).
So replace DP by D(svo)& D(Cvo). Similarly if o is a conjunction.
Step 6.2 . .As for 6.1 with <) replacing D and & replacing v and vice versa.
Step 7. Use Lemma 5 to move the operators .A as far in as possible p:;i.st all
propositional connectives except �.

Step 8. Repeat the above steps as many times as possible.

.After the above has been completed, we obtain a propositional combination
of formulae of the forms p, � p, A p, .A� p, DP, .A DP, <)p, .A<)p where p is
a propositioni;!il letter and p is a propositional formula.
Note that (3 contains no modal operator because
(i) steps 1, 2, 4, 6.11, 6.12, 6.14 and 6.21, 6.22, 6.24 include the reduction
procedure of Hughes& Cresswell, p. 53;
(ii) if a combination D.A or <)A occurs it is replaced by .A (step 5) a.nd
(iii) if an A occurs within the scope of Dor <)it will eventua.lly come lmder (ii)
by the iteration step 8.
Now we treat fornmla.e of the forms p, ,...., p, Ap, A� p, D(3, AD(3, <) (3 , A<)p
a,s single propositiona.l letters in a. new a.lphabet a.nd we trea.t the formula. a a.s
a propositiona.l formula. over that a.lpha.bet.
Step 9. Reduce a to conjunctive norma.l form.
Step 10. Ea.ch conjllllct of a is provably equivalent to a formula of the
following form by rearranging disjuncts;
ava'V D(31V . . . VDPmV ADy1V... vA D yn V<)(3�V . . . v<)(3;v .A <)y�V... VA<)y�
where a, p1, ••• , f3m, (3�, ... , p;, y17 ••• , Yn, y�, ... , y� contains neither A nor a.ny
modal operator and a' is a disjunction containing only formulae of the form Ap
o r �Ap. a' is therefore equivalent t o a formula o f the form Aa0 where a0 is
a propositional formula (in fact a disjunction of possibly negated propositional
letters). Replace a' by Aa0• Since S5 f- <)((31v(32) = <)(31v <)p2 we :may replace
<)(3�v...v<)(3� by a formula <)p0 where p0 i& a propositiona.l formula ((3�V...V (3;)
and similarly we may replace A<)y�v... vA<)y� by a formula A <)y0 where y0
is y�v... vy; and is a propositional formula (using Lemma 5).

Definition. 20. A formula is said to be in ANF (A normal form) if it is a con­

junction of formulae of the form
( *) avAa'v DP1 V . ..v Df3mv ADP�v... vA D(3�v <)y vA<)y'
where a, a', p1, ••• , Pm , {3�, . .. , p�, y, y' contain neither A nor any modal operator.

Since we can permute the order of disjuncts by propositional calculus we

have established the following theorem.
28

There is an effective procedure which, given a formitla a, yields

Theorem 21.
a formula a* in ANF such that SSAR f- a = a*.
Theorem 22. A formula a in ANF is valid if and only if in each conjunct of
the form (*) one of the formulae in the lists (**) (***) below is valid in proposi­
tional calculus.
(**) aVy, /31Vy , ... , fJmVy,
(***) a'vy', {3�Vy', ... , (J�Vy'.

Proof. If a formula in (**) is valid, say /3iVy, then PC f- (JiYY hence

S5 f- (JiV()y whence S5AR f- (*). If a formula in (***) is valid, say (J�vy'
then PC f- {3�Vy' whence S5AR f- {3�V()y'. But then S5AR f- A(/3�V()y')
by (A5) and S5AR f- A(J�vA()y' by Lemma 5, so S5AR f- (*).
Conversely, suppose none of the formulae in (**) or (***) is valid. We shall
construct an S5AR model in which (*) is not valid.
We construct an S5AR model W = (W, w*, R, V). VYe let W1 = {w 0 , , wm}, • • •

W2 = {w*, w�, .. . , w;,.} and W= W1 u W2• V\Te set wRw' for w, w' E W if and
only if w, w' E W1 or w , w' E W2• \-Ye define V as follows. By hypothesis there
are assignments V0 , ••• , Vm and V�, ... , V� making the formulae in (**) and (***)
false. Let V (p , wi )=V i (p), V ( p, w*) = V� ( p) and V ( p , w� ) = V� (p) for
0 <i <m, 1 <j<n .
Now we consider V( (*) , w0). By construction V ( avy , w0)=If so V ( a , w0)=F
and V (y , w0)=JJ'. Similarly V (y , wi)=F for 1 <i<m, but w0 , , Wm are all
• • •

the worlds R-related to w0 so V ( ()y , w0) =F. Since V ({h, wi) =F, V (DPi , wi)=F
for O<i<m. Hence V ( aVDP1V ... YDPmV()y , Wo)=F.
Similarly V (a'VDP�V... VDP�V()y', w*) =F. But

V (Aa'VADP�v...vAD(J� vA()y', Wo) = V(A(a'v 0/3�v...vD(J;,.V()y'), w o)

= V((a'v... v()y') , w*) =F.
Hence V {(*) , w0) =F, so (*) is not valid as required.

Theorem 23. There is an effective procedure for deciding whether a given

formula is a theorem of S5AR or not (or, alternatively, deciding whether it is valid).

Proof. Given a reduce a to ANF. Test each conjunct as described in the

proof of Theorem 21. Since the finding of models for formulae of propositional
calculus is uniformly effective, it follows that the above procedure is effective.
This completes the proof.

Corollary 24. There is an effective procedure for deciding whether a given

formula is a theorem of SSA (or, alternatively, deciding whether it is invalid). S5A
is consistent and complete.
Proof. This follows at once from Theorem 22 using the fact that SSA is an
extension of S5AR, S5A f- AD a= Da and SSA f- A()a = ()a.
 29

References

[l] Cast ane d a, H. N.: Some Nonformal "Logical" Relations, Phi los. Studies, Vol. 8 (1957),
89-92.
[2] H u gh es G. E. and Cress w e l l M. J.: An Introduction to Modal Logic, London, Methuen,
1968.
[3] Ka m p J. .A. W.: Formal Properties of 'Now', Theoria, Vol. 37 (1971), 227-273.
[4] Ka p l a n D.: The Logic of Demonstratives (Unpublished, Circa 1970).
[5] M a k i nson D. C.: On some completeness theorems in modal logic, Zeits. f. Math. Logik
u. Grundlagen d. Math. Vol. 12 (1966), 379-384.
[6] M e n d e lson E.: Introduction to Mathematical Logic, Princeton, van Nost rand, 1968.
[7] M o n t a g u e R.: Logical Necessity, Physical Necessity, Ethics, and Quantifiers, Inquiry,
Vol. 4 (1960), 259-269.

View publication stats