168 JOHAN VAN BENTHEM

--(1991b), Language in Action: Categories, Lambdas and Dynamic Logic, Studies

in Logic, 130 (Amsterdam: North-Holland).
--(1991c), 'Logic and the Flow of Information', Stanford University CSLI
Report 91-160; repro in D. Prawitz, B. Skyrms, and D. Westerstahl (eds.),
Proceedings of the Ninth International Congress of Logic, Methodology and
Philosophy of Science, Uppsala, 1991 (Amsterdam: Elseviers Science Publishers).
-- (1993), 'Programming Operations that are Safe for Bisimulation', Report
93-179, Center for the Study of Language and Information, Stanford University
(forthcoming in Proceedings Logic Colloquium. Clermont Ferrand 1993, special
issue of Studia Logica.)
--(1994a), 'Dynamic Arrow Logic', in J. van Eyck & A. Visser (eds.), Logic and
Information Flow (MIT Press, Cambridge, Mass.), 15-29.
--(1994b), 'Modal Foundations for Predicate Logic', Research Report
CSLI-94-191, Center for the Study of Language and Information, Stanford
University (forthcoming in Studia Logica Library, E. Orlowska (ed.), Memorial
Volume for Elena Rasiowa).
--(1995), Language in Action, second expanded edition (MIT Press, Cambridge,
Mass.).
--'Dynamic Logic and the Flow oflnformation', Studies in Logic, Language and
Information (forthcoming Cambridge University Press).
--VAN EYCK, J. and STEBLETSOVA, V. (1994), 'Modal Logic, Transition Systems
and Processes', Journal of Logic and Computation 4:5, 811-55.
VENEMA, Y. (1992), Many-Dimensional Modal Logics, dissertation, University of
Amsterdam, Institute for Logic, Language and Computation (Oxford: Oxford
University Press, forthcoming).
in: BJ. Copeland (ed.), Logic and Reality, Clarendon Press,
Oxford 1996,169-213.
8
Transfer Theorems for Multimodal Logics
KIT FINE and GERHARD SCHURZ
1. Introduction
Many of the modal logics that have been developed contain two or
more modal operators. A notable example is the tense logic of Prior,
which contains operators for both the past and the future. A more
recent example is the logic of programs, which contains infinitely many
operators, one for each program.
A multimodal logic will have various monomodal fragments; and in
the simplest case, it will be the join of these fragments-there will be no
interactive axioms. Our concern in the present chapter is to investigate
the question of when certain properties of the monomodal logics trans-
fer to their join. To answer this question, we develop a very general
proof method, which allows us to piece together models for different
logics. The resulting theorems provide very general answers to our ques-
tion, which are positive in most cases, but not in all.
Our investigation is a natural continuation of those begun by Prior.
For he was interested both in the development of multimodal logics and
in their relationship to monomodal logics. It is therefore with a keen
sense of his own contribution to the subject that we have pursued the
present line of research.
We let ;el and ;e2 be propositional mono modal languages with
respective modal operators 01 and 02; and let ;e12 be the propositional
bimodal language with operators 01 and 02. Let L C ;e12 be a normal
bimodal logic. We call L stratified if there are normal monomodallog-
ics LI C 51 and L2 C ;e2 such that L is the minimal normal bimodal
logic containing both LI and L2. We write L = LI EB L2 in this case and
Some of the initial ideas behind this chapter were contained in a letter from Fine to Schurz in
1990. The subsequent work has been joint, with Fine writing up sections I and 6 and Schurz
writing up the rest. The result on strong completeness transfer has been obtained indepen-
dantly by Valentin Goranko and Solomon Passy; the results on transfer of strong and weak
completeness. f.m.p., and of decidability (under the assumption of weak completeness) have
been obtained independently by Marcus Kracht and Frank Wolter. Our own proof of decid-
ability transfer is based upon ideas in their proof.
170 KIT FINE AND GERHARD SCHURZ
say also that L is the join of L\ and L2. Joins ofaxiomatizable logics
are axiomatizable in a simple way: if L\ is axiomatizable as K\X\ and
L2 as K2X2 , then L\ E9 L2 is axiomatizable as K\2(X\ u X2). (Here K;
is the minimal normal and Xi a set of additional axiom
schemata.)
We say a property transfers if, for any normal logics L\ and L2,
L\ EB L2 has the property whenever both L\ and L2 have it. In this chap-
ter we consider the question of which properties transfer. We base our
investigation on a generalized notion of completeness (with respect to
classes of frames) which is relativized to a formula space. We show that
this generalized notion of completeness transfers. This general result
will yield immediately the transfer of strong and weak completeness and
of some intermediate notions of completeness. The proof underlying
this result will then establish the transfer of various other properties,
such as the finite model property (f.m.p.) and decidability and interpo-
lation (though under the assumption of weak completeness). Finally, we
show that all of these results generalize to stratified normal multimodal
logics (with slight exceptions in the case of Lm.p. and decidability), and
that certain analogues of these results hold also for stratified classical
logics.
Our transfer results make it unnecessary to establish completeness
and other properties separately for stratified multimodal logics, as long
as these properties are known to hold for their monomodal compo-
nents. Thus the results have applications in all areas in which several
modal operators which do not interact logically are used.
One area of application is the is-ought problem, i.e. the investigation
of Hume's claim that normative statements are never derivable from
descriptive statements. In arguments put forward by ethical naturalists,
where nonnative statements are apparently derived from descriptive
premisses, these premisses involve statements about necessary features
of human nature, society, or natural order. Thus the logical study of the
is-ought problem requires a bimodal logic with non-interacting opera-
tors for necessity and obligation, i.e. a stratified alcthic-deontic logic
(Schurz 1991, 1996). Now any comprehensive study of the is-ought
problem should investigate, not one particular alethic-deontic logic, but
all the possible joins of the philosophically relevant alethic and deontic
logics (Schurz 1991: 44, 1996, ch. 1). Moreover, it turns out that the
investigation of the is-ought problem requires not only knowledge of
completeness, but also of such properties as interpolation and Hallden-
completeness (cf. Schurz 1991: 74-83). These properties are known for
.... n .. ... t n"\nnnn'"vhl lou;rc;: hut not for their bimodal ioins. While
TRANSFER THEOREMS FOR MULTI MODAL LOGICS
171
with the traditional methods of proof, one would have to establish these
properties separately for all joins, our transfer theorems give the gen-
eral answer at once. To be more exact, if La:= KaXa is an alethic logic
for (W,R) (R being the alethic accessibility relation)
satlsfymg condItIons Ca, and Ld := a deontic logic complete for
(W,S) (S being the deontic ideality relation) satisfying con-
ditIOns Cd, then their join La EB Ld = KadXa Y d is complete for all
frames for which (W,R) satisfies C
a
and (W,S) satisfies Cd;
moreover, if both La and Ld have in addition f.m.p., or are decidable
or have interpolation, then La E9 Ld has this property too. '
In a broader framework, one would wish to investigate the is-ought
for join a deontic logic and a bimodal tense logic (con-
talnlng.lnteractlve. aXIOms fOf the past- and the future-tense operators).
One might also Wish to add operators for action and belief. Since our
results are fully general, they will also hold in these cases.
A second area of application is to erotetic logic. According to one
approach, the question 'Is it raining?' is taken to mean 'Make it that
I either know that it rains or that I know that it does not' (formally
O(Kp v K-,p)). The appropriate logic for 0 and K would appear to be
a . logic, as in Aqvist's system PIE (1965).
And agam, one may Jom vanous systems of epistemic logics with vari-
systems o,f deontic logics. A final example of non-interaction is pro-
Vided by the operators for belief and ethical value in a stratified
logic. These systems arise from the study of the the-
SIS of the value neutrality of science, where it is crucial that a statement
like 'It is believed that something is ethically good' is not itself an eth-
ical but a descriptive statement (cf. the study of the 'Max Weber thesis'
in Schurz 1996, ch. 7.1).
Of course it would be desirable to investigate the question of trans-
fer also for the case in which certain standard interactive axioms are
added to the join. An example is again provided by tense logic since
many such logics arise from adding Prior's axioms p -7 GPp and
p -7 HFp to the separate logics for the past and the future. Although
our results do not directly apply to multimodal logics with interactive
axioms, they can be helpful for the study of non-stratified multi modal
logics in two ways. First, it is often simpler to establish completeness
(and other properties) for such a logic L if this is already known for its
sUblogic (i.e. the maximal L' L which is stratified). Second,
also happen that our method of proof can be applied in estab-
hshlng transfer In the presence of a standard set of interactive axioms.
We shall give an example of each sort. The examples are informally
172
KIT FINE AND GERHARD SCHURZ
discussed in the rest of the introduction; a more fom1al treatment is
given in the appendix. .' . .
The first example is dynamic logic DL. This mfimtely many
box operators [a] corresponding to programs a. these
grams correspond to relations Ra over a set J'!' of possible states,
an 'arrow' in Ra corresponding to an executlOn of program a
starts from the one state and terminates with the other. Accordmgly,
[alA is taken to mean that every execution of a
leads to a state in which A is true. The basIc system of DL contams no
formal constraints for the atomic programs beyond the axioms and rules
for normal logics (cf. Harel 1984: 512-13; .19.87: 88).
However, for certain applications, both within dynamIc logic I:self (cf.
Harel1984: 522' Gargov and Passy 1990) and to the logic of actlOns (cf.
Segerberg 1980:' 292), it is natural to impose additional axioms ?n cer-
tain classes of atomic programs. For instance, a class of atomIC pro-
grams Pd may be taken to be deterministic (upon executing a
in Pd, each input state leads to at most one output state). One wIll then
add the axiom (Deta): (a)A [alA for all a E Pd. Another neces-
sarily disjoint) class of atomic programs Pt may be .by the
condition that they always terminate. The correspondmg ax lOrn IS then
(Da): [alA (a)A for all a E Pt. A third class is the 'complete random
programs' P which at any given input state, randomly choose any pos-
sible outpu[ state.' They correspond to the (S5a)-axioms: [alA A,
[alA [a][a.]A, and (a)(a]A A, for any aE P
r
.. .
DL when extended by axioms of this sort, looks hke an ordmary
But it is important for dynamic logic that atomic pro-
grams can be combined into more complex programs. One standardly
has the program of executing first a and then {3, the program a {3
of executing a or (non-deterministically), the program ?A of tes,h?g
whether A is true, and the program a* of iterating a a non-determmls-
tically chosen number of times. These complex p,rograms are
ized by certain additional interactive axioms, which are explamed m the
appendix. , .' .
The non-interactive fragment of a dynamic logic DL IS called a quasl-
dynamic logic, QDL. Although our transfer directly
only to QDLs, they have some straightf0rw.ard ImphcatlO?S the
interactive part. The superposition of QDL with the mterachve aXlOms
for the operations of sequence (;), union (u), and test (?) has, been
loop-free dynamic logic, for short and has. been a of mde-
pendent interest (see Bowen 1980). It IS shown III the appendix that all
of the transfer theorems apply also to the FDL-extension of a QDL.
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 173
The problem is much more difficult when the star operator * is included
and we have no general results.
The second example concerns a certain class of interactive axioms
which may be imposed on a bi- or multimodal logic. Assume LI and
LJX are complete OJ-logics (where X is an additional axiom schema
and LJX the smallest Dl-logic containing LI and all instances of
X), L2 is a complete D2-logic, and consider the bimodal logic
L := LIEB L2 + D2k OJ (X), for a fixed k;::: 1; where D2k OJ (X) :=
{D2k DJnA I A an instance of X, nEw}. This logic lies between the two
joins LJ EB L2 and LJX EB L2; it coincides with the latter for k = O. With
slight modifications, the proof underlying our theorems goes through in
this case, giving the result that if Ll, LJX, and L2 are complete (or have
f.m.p. or interpolation under the proviso of weak completeness), then
also L has the respective property.
Note that in the special case where Ll already contains DIX (but not
necessarily X), L coincides with LJ EB L2 + D2kX. A situation of this
kind arises within doxastic deontic logic, i.e, the logic of rational belief
B and obligation. For though BA A is certainly too strong for
belief, B(BA A) is acceptable under certain interpretations, as is
O(BA A). A similar situation arises when rational desire D is com-
bined with deontic logic [D(DA A); OeDA A)]; or when rational
desire is cOI1}bined with rational belief [D(BA A)]. A related exam-
ple-though within the realm of classical rather than normal logics-
arises from Aqvist's discussion of his epistemic-deontic logic PIE
(explained above). He argues that one of the shortcomings of PIE is
that it implies logical omniscience, whence he suggests replacing the
basic axiom and rule for normal epistemic logics, namely K(A B)
(KA KB) and A/KA, by the weaker versions O(K(A B) (KA
KB)) and A/OKA (1976: 30-5).
2. Preliminaries
Before we tackle our main problem, let us record some basic terminol-
ogy. ; will always denote a propositional modal language, which is
identified with the set of its well-formed formulae. These are built up
from a given infinite set of propositional variables IP C 5. A, B E ; are
formulae and r, LI ; sets of formulae. For a given index set T, let 51
be a propositional multimodallanguage with modal operators 0; for all
i E T. (In particular, 5{I), 5{2), and 5{I.2} are abbreviated by 5}, 52,
and ;12, respectively.) As usual. we call a subset L 5[ a normal modal
logic of 5[ iff it is closed under substitution, modus ponens (MP). and
174
KIT FINE AND GERHARD SCHURZ
the necessitation rule (Nj): AIOjA for all i E I, and if it contains all tau-
tologies (Taut) and the axiom scheme (Kj): o,{p q) (DiP 0iq) for
all i E 1. The deduction relation for L is defined by r A iff
A) E L for some finite r. In what follows, except in
section 6 on classical logics, L will always stand for a normal modal
logic.
We base our investigation on the relational semantics for modal log-
ics. Frames for a multimodal language :/ have the general fonn
(W,(Rj)iE/), where the R;'s are binary relations on the non-empty set W.
In particular, frames for!1 (:2, !12) may be supposed to have the fonn
(W,Rl) W,R2)' (W,RhR2)' respectively). w, v, u, ... denote possible
worlds. A model based on the frame (W,(Ri)jE]) has the fonn
(W,(Rj)iEJ, <p), with <p: IJl> Pow(w) (,Pow' for power set). Particular
modelsl frames are denoted by italic letters MI F and classes of modelsl
frames by roman letters M/ F. The clauses for 'Fonnula A is true at
world w in model M' are the standard ones. A fonnula A is valid in a
model M iff it is true at all worlds in M; it is valid on a frame F iff it is
valid in all models based on F. A set of formulae Ll is true at w in
M Iva lid in M Ivalid on F iff this holds for every A E Ll. A model M is
a modelJor L iff L is valid in M; a frame F is a frame Jor L iff L is valid
on F. L is weakly/strongly complete Jor a class F oj Jrames iff every
L-consistent A E !ILl C! is true at a world in a model based on a
frame in F. Lis weaklylstrongly complete iff there exists a class ofJrames
for L for which L is complete.
We first record two basic facts. For L !], let F(L) denote the class
of all !I-frames for L. If FI is a class of !I-frames, and F2 a class of
then we define the join of FI and F2, denoted by FI EB F2,
as {(W,RhR2) I (W,RI) E FI and (W,R2) E F2}.
PROPOSITION 1. For all LI 9;\, L2 :2 : F(L\ EB L2) = F(Lt) EB F(L2).
(Proof Straightforward.) If a logic L is frame-complete at all, it is cer-
tainly complete for F(L). This together with Proposition 1 implies that
the problem of completeness transfer reduces to the following: given
that L\ and L2 are frame-complete, is then LI EB L2 complete for
F(Ll) EB F(L2)? This reduction will be helpful in the proof of complete-
ness transfer.
The next proposition is of independent interest. A multimodal logic
L !] is a conservative extension of Lj (i E l) if for every A E !j: A E
Li iff A E L.
PROPOSITION 2. Provided Ll and L2 is consistent: L = Ll EB L2 is a con-
Qvt"nc;n" of T.I :lOci of L.,.
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 175
One way to prove Proposition 2 is by a well-known result of Makinson
(1971t every consistent nonnal monomodallogic L,.,. is valid either on
the smgleton:frame.({w},{(w,w)}) or on the singleton frame ({w},0). A
powerful proof which generalizes to joins of countably infinite classical
multimodal logics has been given by Thomason (1980: 143-4). As a
corollary, Proposition 2 gives as a first result on transfer:
COROLLARY 1. Consistency transfers. (Proof Straightforward.)
We turn to the transfer of completeness. The ensuing proof technique
rests on the fact that a formula of the bimodal language A E !12 can
be c?nsidered also as a formula of the monomodal language !\ by
treatmg outermost occurrences of 02-subformulae of A as atomic. Here
occurrence of 02B in A is called outennost if it does not properly
m the scope of some 02. Similarly for !2 and outermost occurrences of
0l-subfonnulae. One way to implement this technique would be to
all occurrences of 02B-subformulae by special propo-
sl.tlOnal vanables of !\. We would have then to deal with the three
different languages !t. !2, and !12, between which we would 'move'
appropriate We will pursue here a slightly
dlf,(erent strategy, which avoids the use of different languages and
replacement functions treating the bimodal language !12 as ambigu-
ous. Thus, we can consider a formula A of !12 as a formula of either a
a 02-logic, or a bimodal logic, depending on whether we con-
outermost ol-subformulae, or only propositional
vanables as atomic. These different 'views' on 9;-formulae get cashed
out in terms of different kinds of model.
Let us introduce the basic notions. We work with the bimodal lan-
guage !12, which we now write as !. !'s formulae are built up from
the set IJl> of propositional variables by the formation rules of bimodal
logics: IJl> C!, and if A, BE!, then -,A, A vB, 0IA, and 02A E 9;.
To simplify our. speech, we introduce the variable rr for the type of the
monomodalloglc; rr ranges over the two values 1 and 2. rr* denotes the
complementary type of rr in {I,2}, i.e. rr = 112 iff rr* = 2/1. Put Orr :=
{o,.,.A I A E :}. Then ! can equally be considered as the set of wffs of
an built up from the atomic variables IJl> u Orr by the for-
mation rule of an !,.,.-language; i.e. IJl> u orr* ! and if A, B E :, then
-,A, A v B, and o,.,.A E!. A fonnula which is atomic in! viewed as a
rr-language, i.e. every formula in IJl> u orr*, is called a ;r-constituent. 1:rr
= IJl> V denotes the set of all rr-constituents. A constituent is any l-
or 2-constltuent; 1: = El u E2 is the set of all constituents in :; i.e. the
set of all !-formulae which are not truth-functionally compound.
176
KIT FINE AND GERHARD SCHURZ
We represent the different views on : semantically in terms of dif-
ferent notions of frame and model. A n-model for :, based on the n-
frame (W,R.r), is a triple of the form M = (W,Rmq>,,) with 'P,,: Err-:
Pow( W). (If M is a rr-model, its components are also denoted by W ,
R"M, and q>"M.) A 12-model for :, based on the 12-frame (W,RI,R2), is
a quadruple (W,R!,R2,'P) with '1': I? -t Pow(W). Let ,.,. E {1,2,12!.
'(M,w) \= A' stands for 'A is true at world w in ,.,.-m?del M' and IS
defined according to the recursive clauses for : viewed either as an :1-,
an :2-, or an :12-language according to whether,.,. is 1, 2, or 12. (In
particular, the atomic clauses for,.,. = 112/12 are: If A E ElIn/l?, then
(M,w) \= A iff wE 'P1(A)/'P2(A)/q>(A).) Validity of formulae in a ,.,.-model
and on a ,.,.-frame.is defined in the usual way. .
L" : is a n:-Iogic over :t iff it contains (Taut), (K,,), IS closed
(MP), (N,,) and under of for elem.ents m Err
(observe the modified substitutlOn A 12-I?glc over :t,
L c: is defined in the standard way (m particular, L IS closed only
under for elements in I?). Note that if L :" is a rr-Iogic
in the monomodallanguage :", then the corresponding rr-Iogic L" over
: is just the closure of L under substitution of :-formulae. for
sitional variables. Weak and strong completeness of a ,.,.-loglc over : IS
defined as before but with respect to ,.,.-frames [,.,. E {1,2,12}]. Clearly,
a normalrr-Iogic 'L over : is complete for a class of rr-frames F iff its
restriction L (l :" to :" is complete w.r.t. F.
For A E :, SF(A) is the set of all subformulae of A viewed as an :-
formula. SF ,,(A) is the set of all rr-subformulae of A v!ewed an :,,-
formula' these are all formulae having an occurrence m A which does
not lie in the scope of a otTo-operator. C,,(A) is t?e set of all n-
constituents of A' these are the rr-subformulae of A which are rr-con-
stituents; so C,,(/f) = SFiA) n Err. Obviously, A E: is built up from
C,,(A) by the formation rules of :" alone, and the of A at
a world w in a rr-model (W,R".,q>,,) depends only on the restnctlOn of '1'"
to the elements of C,,(A).
The set of all subconstituents of A, SqA), is defined by SqA) =
SF(A) n E. So a subconstituent of A is any, which has the
form of a constituent. A frequently occurnng notlOn IS the set of all so-
called n:-subconstituents of A, S,,(A); this is the set of all subconstituents
of rr-constituents of A; so S,,(A) = SqC,,(A. (Observe that the notion
of rr-subconstituents S,,(A) is different from the notion of subcon-
stituents which are rr-constituents, SqA) (l Err; however S,,(A) could be
equivalently defined by S,,(A) = To give an
pYRmo1e. for and SCt assume A = 02Q /\ 01-,0\02]1. Then CI(A)
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 177
= {02Q, O2]1}, C2(A) = {q, OI-,0102P}, SI(A) = {02Q, 02]1, Q,p}, S2(A) =
{q, 01 -,012]1, 0102]1, 02]1, p} and SqA) = {02q, 01-,012]1, q, 0102]1,
02]1, pl.
The set of truth-functional constituents TqA) of a given formula A is
the set of all constituents of A which are truth-functional components
of A. All the above notions apply in the obvious way also to formula
sets .d: if 'r// is any of 'SF, SF", C,,' SC, S"., TC', then .p(.d):=
{.p(A) I A E .d}. Finally, B(.d) is the set of all Boolean, or truth-func-
tional, compounds of formulae in LI.
An important notion is the extended n-degree of a formula A viewed
as an :t-formula, drr(A) , in distinction from the narrow rr-degree of
A viewed as an :t,,-formula, dOrr(A). drr(A) is the number of nested 0,,-
occurrences in A, independently of whether they lie in or outside
o,,*-scopes. The inductive definition of d1T(A) is: (i) for pEl?,
d1T(P) = 0, (ii) d1T( -, A) = d1T(A), (iii) d1T(A v B) = max( {d1T(A),drr(B)}),
(iv) d1T(O".A) = d1T(A) + 1, (v) d1T(O,,+A) = drr(A). For a set of formulae
.d, d1T(.d) shall be the ordinal supremum of {d1T(A) I A E .d}: if {drr(A) I
A E .d} has a maximum, then drr(.d) is this maximum; otherwise it is w
if .d is non-empty, and zero if .d is empty. In distinction, the narrow
1T-degree d01T(A) of A counts only the number of nested o".-occurrences
outside o".-scopes and is inductively defined by replacing (i) by 'if
A E E1T, dOrr(A) = 0' and omitting (v).
Given a frame F and relation R, a sequence vo,v\, ... , Vn-I,V
n
(n ;:::: 0)
of worlds in WE is called an R-path in F from w to u, with length n, if
Vo = IV, Vn = u, and VjRVi+l for 0 :5 i < n (so every world leads to itself
by an R-path of length zero). We write IV"m if F contains some R".-
path from IV to u. The 1T-distance dist,.(w,u) is the minimal n 0 such
that there exists an R,,-path in F from IV to u with length n; if there exists
no R".-path connecting wand u we put dist,,(w,u) = w. If F is a 1T-frame,
we simply write dist".(w,u) = dist(w,u). The semantical relevance of the
narrow rr-degree is obvious (the truth of A at w in a 1T-model M depends
only on worlds v E WM with dist(w, v) :5 d01T(A) ). The semantical rele-
vance of the extended 1T-degree depends upon the fact (later to emerge)
that any stratified 12-logics can be characterized by 12-frames which are
weakly 12-separated in the sense that, for all worlds w, u with w '* u, if
w -;;7 u then there exists no v '* w with w 7 v and v -;;7 u. Take a 12-model
M based on a weakly 12-separated 12-frame making A true at w, and
let WI(M) be the set of all worlds u in M with w. u. Assuming the
worlds in Mare R2-refiexive, the truth of A at w may then also depend
on the truth of A-subformulae in 02-scopes at worlds v in WI(M) with
distl(w,v) > dO}(A). For example, if A = 01020 1P and WRIURIV, then p
178
KIT FINE AND GERHARD SCHURZ
must be true at v. However, the truth of A at w can only depend on
those worlds u among the worlds in WI(M) for which distl(w,u) S
d I(A) holds. . .
For .1 k !t, DC,,(.1) is the degree-conserving o,,-closure of .1. ThiS IS
the set of all formulae of the form o,,"B, where BELl and drr(O"nB) S
drr(A) (here n ~ 0, o,,"A := on ... o"A n times and on
oA
:= A). Observe
that this definition is equivalent to the following inductive definition: (i)
.1 k DC,,(.1), and (ii) if A E DCiA) and drr(A) < drr(B) for some
BELl, then o"A E DCiA).
3 .. ,Generalized Transfer of Completeness
We say a logic L is complete w.r. t. a formula set r iff every L-consis-
tent subset A C r is true at a world in a model based on a frame for L.
All standard ;'tions of completeness are instances of this more general
notion; for instance, strong completeness is completeness w.r.t. ; and
weak completeness is completeness w.r.t. all finite formula sets. Also
observe that, by truth-functional logic, completeness w.r.t. A reduc.es to
completeness w.r.t. the set of un negated or negated truth-functIOnal
constituents of A.
A formula set closed under subformulae and under truth-functional
compounds is called a formula space.
THEOREM I. Let LI and L2 be a 1- and a 2-logic over !t and e a for-
mula space. Then: If LI is complete w.r.t. Deice) and L2 is complete
w.r.t. D
C
2(e), then L := LI ffi L2 is complete w.r.t. e.
First we give an informal explanation of our method of proof (the
systematic proof starts with Def. 1). Take some r k e which is L-con-
sistent. To prove our theorem we must show that r is true at a world
in a 12-model based on a 12-frame. Our proof will show that this model
can be obtained from certain 1- and 2-models which are based on 1- and
2-frames for Ll and L2 and which make true certain LI- and L2-consis-
tent formula sets in DCI(e) and DC2(e), respectively.
Because r is L-consistent, it is both LI- and L2-consistent. Since r k
e k DCn(e), there exists a 7T-model, for 7T = 1 or 2, which makes ~ true
at some world and which is based on a frame for LI or L2, respectively.
We call this model our initial model M* and the world which makes r
true the initial world w*. The world w* plays a special role in M*, which
we express by calling w* the base world of M*. For reasons of genera-
lity. we do not require M* to be w*-generated (although we c o u l ~ ) .
A \1 of the rr-models we use will be labelled. A labelled rr-model IS a 7T-
,
. l'
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 179
model M together with a base world WM and a constituent set E(M)
which is closed under subconstituents. We assign to each world w in M
its constituent set EMCw) by the following inductive definition: EMCwM)
= E(M); and if uRrrv and O"A E EMCu), then SqA) k L'MCv). The con-
stituent set of the initial model M* is sqn. The requirement of clo-
sure of constituent sets under all subconstituents is made in order to
guarantee that for all wand u in M* with w ~ U, EMo(U) contains all
subconstituents whose truth-value (at u) is relevant for the truth-value
of constituents in EMo(W) (at w)-not only in the initial model M*, but
also in later steps of the construction, when reflexive R"o-relations may
be added. To give an example: if DI D2DIP E EM*(W) and WRIURI v, then
o2olP must be in L'M*(U), and-given the possibility of uR,,*u-also DIP
must be in EMO(U), whence P must be in EMO(V). The SC-closure require-
ment is the simplest way to take care of this; it may yield more sub-
constituents than strictly necessary; but this is harmless.
Consider now, for any u E WMo, the set S,,(L'M*(U of rr-subcon-
stituents of EMo(U). The sets S,,(L'M*(U figure as agreement sets, because
we want to continue the construction by 'hanging off 7T*-models at these
worlds u which agree with M* on the truth-values of constituents in
S,,(L'M.(U. In order to do this, we must guarantee that the agreement
diagrams-the sets of those elements in S,,(EMo(u-the set of those ele-
ments in S,,(L'Mo(U or their negations-which are true in M* at u-are
themselves L-consistent. We will ensure this by requiring that beyond the
truth of r at w* in M*, the so-called lr-theory of E(M*), Trr(E(M*,
must be made true in M* at w*. This is the set of all D"Il-closures of
L-theorems which are truth-functional compounds of formulae in
SrrCE(M*, where n is limited by the maximal distance of a world
v from w* with non-empty constituent set. Because r is L-consistent
and T,,(E(M* k L, rv TiE(M* is L-consistent. Moreover,
rv T,,(E(M* k DCie); and so the requirement that M* makes
r v T rrCE(M*) true at w* can be satisfied.
The operation of 'picking out' only rr-subconstituents in the defini-
tion of the 7T-theories and the agreement sets (and diagrams) has an
important function: although it is not necessary for completeness trans-
fer, it will enable the proof of further transfer properties by induction
on 'alternation degree'-which is, roughly speaking, the number of
D1f-D".-o1T-nestings. The limitation in the definition of the 7T-theories will
enable the transfer of weak completeness though ensuring that when r
is finite, the rr-theories can be represented by finite formula sets. (For
strong completeness transfer it would be sufficient to replace T ,,(E(M*)
by L.)
180 KIT FINE AND GERHARD SCHURZ
For any u E WM, the agreement diagram of u in M* is L-consistent
(and therefore L,,-consistent) and it is contained in e, whence there
exists a 1T*-model N based on a frame for L,,* which makes this agree-
ment diagram true at some world. We identify this world with u, declare
it as N's base world (u = WN) and assume that WN is the only world N
shares with M*. We identify N's constituent set with the agreement set
of u in M*. In addition we require, for the same reasons as above, that
N makes the 1T*-theory of (N) true at WN. If all these conditions are
satisfied we say that N hangs off M* at WN. We proceed by hanging off
M* disjoint 1T*-models N at every world u in M*. We can iter-
ate the construction, by hanging off 1T-models at all worlds in these
models which are not base worlds. And so on, ad infinitum. It IS
important that with the exception of the 11*-model hanging off M* at
w*, we hang 11i11*-models off 11*/11-models only at worlds which are not
base worlds of the latter. This guarantees that the frame of our later
'limiting' 12-model constructed out of all these labelled 1- and 2-mod-
els will be a frame for L.
By hanging off models in the described way we step by step,
an increasing set of 'hanging' 1- and 2-models which, If fitted together
by an appropriate union, will yield a step. by
step, formulae in r true at W* with a successively alternation
degree. Instead of constructing this infinite mo?el stepwise, we prefer a
simpler and more abstract approach. We c?nslder all sets ?f 1- and 2-
models hanging off each other in the descnbed way. We call these
sets sproutings. Their elements will be taken from a set of large
cardinality. Zorn's lemma tells us that the set of these ha.s a
maximal element. The union of the elements of a maXimal sproutmg
will give us the desired limiting ..
The labelled 1T-models in a sprouting will have to satisfy different con-
ditions for different purposes. Our general definition of labelled 11-
model and sprouting will not include the assumption the 11-mo.del
is based on a frame for L" because for some purposes thiS assumptIOn
will have to be dropped. We will rather assume an abstract model selec-
tion function which assigns to an L,,-consistent formula set il a non-
empty set of pairs (M,w) closed under where M a
11-model for L" making il true at w. By varymg thiS model selectIOn
function, the general construction can be adapted to special transfer
purposes. ..
After these informal explanations we give the systematic proof.
TRANSFER THEOREMS FOR MUL TIMODAL LOGICS
181
DEFINITION 1.
(1.1) For any formula set il ;;e, its n-theory is defined by:
T,,(il) = {D.,nB I B E B(Sn(Ll fl Land d11(D",nB) ::;; d11(Ll)}.
(1.2) A labelled n-model is a triple <M,wm,E(M where
(i) M is a 11-model for L,,;
(ii) WM E WM;
(iii) E(M) is a set of constituents closed under subconstituents'
(iv) Tn(E(M is true at WM in M. '
(1.3) Given a labelled 11-model (M,WM,E(M, for each world wE
WM a constituent set EMW) is inductively defined as follows:
(i) EMWM) = E(M), and
(ii) if uRn
v
(u,v E WM) and DnA E EMU), then SC(A) E EM(v).
WM is called the base world of M, E(M) the constituent set of M, and
EMU) the constituent set of u in M. For convenience we will use 'M' to
denote the triple (M,WM,E(M. Def. 1.3 has some immediate implica-
First, each EM(u) is closed under SC, and EM(u) E(M) (induc-
tion on dist(wM,u. Next:
LEMMA 1. Assume M is a labelled 11-model and U E WM. Then:
(1.1) If E<u) is non-empty, d11(EM(u = (d1T(E) - dist(wM,u;
(1.2) For u '* WM, .EMU) is empty iff dist(wM,u) > d11(EMU.
Proof (Ll) by induction on dist(WM,U); (1.2) follows. QED.
. Provided d1T(E) is finite, d11(EMU decreases with increasing
dISt(WM,U), and EMU) becomes empty. In the special case
d11(E) = w, the EM(u)\ s will of course decrease neither in d
11
nor in size.
(1.2) excludes the special case of L'M(wM) = 0, when d11(L'M(wM =
dist(wM,wM) = 0 holds.
DE:INITION 2. Let M, N be labelled models, M of type 11. N hangs off
M Iff (i) N is of type 11*; (ii) WM (') WN = {WN}, and either M = M* or
WN is n?t M's base world; (iii) E(N) = Sn(EMWN; (iv) for all A E
E(N): A is true at WN in M iff it is true at WN in N.
The .. 'hang-off is obviously irreflexive and asymmetric (by
Der.2(1) and (11. If N hangs off M and W is N's base world, we say that
N hangs off M at w. For each 11-model M and W E WM, we call the set
Sn(L'MW the agreement set of w in M; if N hangs off M at W then it
is identified with E(N) by Def.2(iii), and M and N agree at on the
elements of this set by Def.2(iv). (This definition allows disagreement of
182 KIT FINE AND GERHARD SCHURZ
M and N on constituents not in the agreement set; but they will be
ignored in the definition of the final 12-model.) For any LlT-consistent
set A, 'IT-model M, and world w in M, the diagram of A in M at w,
denoted by :tA, is the set {A IA E A and A is true at w in M} u
{ -, A I A E A and A is false at w in M}. The agreement diagram of w in
M is the diagram of S,,(Eu(w in M at wand is denoted by Du(w).
Obviously, condition Def.2(iv) is equivalent to requiring that DM<.wN)
be true at WN in N. The important feature of labelled 'IT-models is this:
LEMMA 2. For every labelled 'IT-model M and u E WM: DM<.u) is L-
consistent.
Proof Assume,- on the contrary, that DM<.u) is L-inconsistent for some
u E WM. So there exists a finite and non-empty set A DA{{u) with
-,,Ad E L (where ALl is the conjunction of all fonnulae in A). Assume
dist(wM,u) = k. Every A E A is of the fonn B or -,B with BE
SlT(EM(u). So B E S,,(E(M and thus Ll B(S,,(I(M) (by Iu(M)
I(M; and d'IT(B) :$ dn(E(M - k and thus dn(LJ) =:;; dn(E(M - k (by
Lemma 1.1; EMCu) :1= 0 because Ll :1= 0). It follows that 0lTk..,ALJ E
T,,(E(M (Def.l.l), whence 0,/-,1\L1 is true at WM in M (Def.1.2(iv.
This implies that -,1\:1 is true at u in M, contradicting the fact that also
AA is true at u in M. QED.
Let M be a set of labelled 1- and 2-models, and M, N E M. The rela-
tion 'N indirectly hangs off M' is inductively defined as follows. (1) If N
hangs ofT M, then N indirectly hangs off M. (2) If, for some 0 E M, 0
hangs off M and N indirectly hangs off 0, then N indirectly hangs ofT
M.
DEFINITION 3. A sprouting of M* (the initial model) is any set S of
labelled 1- and 2-models such that (i) M* E S, and M* does not hang
off anything in S; (ii) every n-model M E S different from M* indirectly
hangs off M*; (iii) two distinct models in S have a world in common
only when one hangs ofT the other at that world.
The following are important properties of sproutings. (3.l)-{3.S) specify
purely 'structural' properties, while (3.6) specifies 'content' properties.
In (3.4) we use the notion of a hang-off chain for M, which is a finite
sequence (Mi Ii=:;; n, n 1) of models in S starting with M* and end-
ing with M such that for every 1 <i =:;; n, Mi hangs ofT Mi-I.
LEMMA 3. Assume S is a sprouting of M*. Then:
(3.1) For every ME Sand wE WM, at most one N E S hangs off
A'{ <It IAJ
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 183
(3.2) For every M E S different from M*: M hangs off exactly one
NES.
(3.3) AlIn-models in S have mutually disjoint world sets.
(3.4) For every M E S there exists exactly one hang-off chain, and it
contains pairwise distinct elements.
(3.5) If N E S hangs off M E S at w =/: w*, then w is not M's base
world.
(3.6) For every n-model ME Sand wE WM: (a) if M hangs off N,
then I(M) E(N); (b) E(M) E(M*) e; (c) DM<.w) e; Cd)
T,,(E(M DC,,(e).
Proof. (3.1) By Defs.2(i) and (ii) and 3(iii). (3.2) By Defs.3(ii) and (iii)
and 2(1). (3.3) By Defs.3(iii) and 2(i). (3.4) By the inductive def. of indi-
rectly hanging off and Def.3.ii., there exists such a hang-off chain. By
Lemma 3.2, every mode] in the chain distinct from M* has exactly one
predecessor, and by Def.3(i), M* has no predecessor; so there exists
only one such chain. Assume now the chain contains N at two distinct
places. Since every element detennines its predecessor, this would lead
to an sequence ... PNQ ... PNQ '" M never arriving at
M*. ThiS impOSSible because of Def.3(ii). (3.5) By Def.2(ii). (3.6) (a)
By Def.2(1ll) and because EMCw) E(M). (b) By Lemma 3.4 there exists
a for M, M* P ... QM. From this and (a) it follows by
an easy mductlon that E(M) E(M*), which proves the claim because
E(M*) = sqI') e (recall that e is closed under SF). (c) follows from
(b) and the facts that Du(u) B(EMCu and that e is closed under B.
(d) By Def.l(iv), TiI(M DClT(B(L'(M). Because (b) holds and e is
closed under B, DC,,(B(I(M) DC,,(e). QED.
T.he properties under (3.2) and (3.4), together with Def.3(i), may be
by saying that the structure (S, hang) is an
trreftexlVe tree With root M* (where 'hang' is the inverse of 'hang-orr).
A hang-off chain for M is a branch from M* to M.
A model selection function m is a function which assigns to each nand
L,,-consistent fonnula set Ll DC,,(e) a non-empty set mlT(Ll) of pairs
(M,w) such that M is an L,,-model making LJ true at w. That such a
function m always exists fol1ows from ordinary model-completeness of
n-Iogics (i.e. every L,,-consistent L1 is true at a world in a model for L1T)'
Given a model selection function m, we define Mm:= {M I (M,w) E
m1T(L1), L1 : is L,,-consistent, n = 1 or 2}, SETm := U {WM 1M E Mm}
and Xm = max({Xo, sup({IWMII M E Mm})}). A given model selection
function m is called regular if it satisfies two additional requirements: (i)
SETm is a set with cardinality ISETml strictly greater than X
m
, and (ii) for
184 KIT FINE AND GERHARD SCHURZ
each L-consistent Ll C;;;; DC".( e), m,,(Ll) is closed under isomorphism in
SETm. A regular model selection function always exists (for, given any
model selection function m"', a regular counterpart m is definable by tak-
ing any set SETm with ISETml > Xm and letting m assign to each Ll all SET
m
-
isomorphic copies of the elements of m"'(Ll. In what follows we assume
the variable m is ranging only over regular model selection functions.
Given an m, we assume <M"',w"'> E m,,(ruT,,(L'(M*) where
E)M*) = SC(r) and take the worlds of M*-sproutings to be elements of
SET
m
. This will guarantee that sproutings may increase indefinitely,
because condition (i) implies that if Ws is the set of worlds of a sprout-
ing S, then I Wsl < ISETml and thus I Wsl < ISET
m
- Wsl, i.e. there will be
always enough remaining objects in SETm .to enable further increase in
S. A sprouting ~ f M* is called an m-sprouting of M'" if for each 1T-model
M contained in it, (M,WM) E m( L'(M) u T,,(L'(M). For any m,
SPROUT
m
is the set of all m-sproutings of M"'.
LEMMA 4. (For any m:) SPROUT
m
has a maximal element.
Proof By Zorn's lemma, namely if every chain in a set a (w.r.t. a par-
tial ordering R of a) has an upper bound in a, then a has a maximal ele-
ment (w.r.t. R). Consider SPROUT
m
, partially ordered by C;;;;. Let C be a
chain in SPROUT
m
(w.r.t. C;;;;). UC is an upper bound of C because
S C;;;; UC for all sproutings SEC. It is easily shown that UC is in
SPROUT
m
(by showing that UC satisfies the conditions of Def.3 of a
sprouting of M*, and moreover that of an m-sprouting). SO SPROUT m has
a maximal element by Zorn's lemma. QED.
The important property of maximal elements in SPROUT
m
is this:
LEMMA 5. Let S+ be a maximal element of SPROUT
m
. Then for every
M E S+ and world win M, which is a non-base world of M if M =1= M"':
there exists an N E S+ hanging off M at w.
Proof Assume ME S+ is a 1T-model, wE WM is a non-base world of
M if M =1= M*, and there exists no N E S+ hanging off M at w. The
agreement diagram of win M, DM(W), is L-consistent by Lemma 2. Put
S,,(L'MCw:= L'N. Because T,,*(L'N) is a set of L-theorems, DM(W) u
T,,*(L'N) is L-consistent and therefore L"..-consistent. Because
DMCw) v T"o(L'(M C;;;; DC,,*(e) (by Lemma 3.6), DM(W) u T,,*(L'N) is
true at a world u in a labelled 1T*-model N such that (N,u) E
m"o(DM(W) v T,,*(L'N. So WN C;;;; SETm and IWNI ::; X
m
. Because m is
closed under sETm-isomorphism, we can identify u with w; we declare w
as N's base world WN and identify L'(N) with L'N (thus DM(W) = L'(N.
Because S+ E SPROUT
m
, every model in S+ has at most cardinality Xm;
TRANSFER THEOREMS FOR MUL TlMODAL LOGICS 185
it then follows from the tree structure of S+ that S+ contains at most
(Xm)n-l = Xm models with hang-off chains of length n. So S + is a count-
able union of model sets with cardinality not greater than X
m
, whence
IS+I ::; Xm. Therefore, IU{ WM I M E S+}I ::; Xm < ISETml; in addition
I WN I ::; Xm < ISETml. Because m,,*(L'(N) v T TTO(L'(N) is closed under
sETm-isomorphism, we can assume that all non-base worlds of N are
taken from SETm-U {WM 1M E S+}. Now N satisfies all conditions to
be a labelled 1T*-model hanging off Mat w. N indirectly hangs off M*
because M does so. N's non-base worlds are disjoint from
U{ WM I M E S+}, and N cannot share W with any model =1= M in S+
(by Def.3(iii) and the fact that no model in S+ hangs off Mat w). Thus
S+ u {N} is an m-sprouting which properly contains S+, i.e. S+ is not
maximal. A contradiction. QED.
DEFINITION 4. For every maximal element S+ in SPROUT
m
, we define
its corresponding 12-model M(S+) as follows: (i) WM(S+) =
U{WMIMES+}; (ii)R"M(s+)=U{R
TT
M
IMES+ and M is a 1T-
model}; (iii) (j)M(S+)(P) = U{(j)TTM(P) 1M E S+, 1T is I or 2}, for p E \PI.
We call M(S+) a maximal sPRouTm-model.
LEMMA 6. Let M(S+) = (W,RJ,R2,(j) be a maximal SPRouTm-model.
Then:
(6.1) Every wE W lies in exactly one I-model of S+, denoted by
MI(lv), and in exactly one 2-model of S+, denoted by M2(w),
one hanging off the other at w.
(6.2) (W,R,,) is the disjoint sum of all the frames (WM,R"M) for M E
S+.
Proof (6.1). By definition, wE Wlies in some 1T-model M1T(W). We dis-
tinguish two cases: (i) w = w*, or W =1= W* and w is not M1T(W)'S base
world. Then W lies also in some 1T*-model M1T*(W) which hangs off
M1T(W) at w, by Lemma 5. (ii) w =1= w* and w is M1T(W)'S base world (so
M1T(W) =1= M*). Then w lies in a 1T*-model M1T*(W) such that M1T(W)
hangs off M1T"'(W) at w, by Lemma 3.2. In both cases, w E W lies in at
least one MI(w) and in at least one M2(w), one hanging off the other
at IV. That it lies in at most one Ml(lV) and in at most one M2(w) fol-
lows from Lemma 3.3. (6.2). By Lemma 6.1, W = U{ WM 1M E S+ is a
1T-model} holds for 1T = I and 1T = 2; and by Lemma 3.3, all world sets
of 1T-models are mutually disjoint. QED.
The central results concerning M(S+) are the agreement lemma
and the frame lemma. The agreement lemma tells us that the 'big'
model M(S+) agrees at all worlds w with the 'small' models M1T(W) on
186
KIT FINE AND GERHARD SCHURZ
all truth-functional compounds of constituents in EM,,(w)(W). (In what
follows we abbreviate M7T(W) by M7T whenever the index w is redundant;
e.g. we write L'M,,(W) for L'M,,(w)(W).)
LEMMA 7. Agreement Lemma.
Let M = (W,R),R2,<P) be a maximal SPRouTm-model. Then, for every
W E Wand A E B(L'M7T(w: A is true at W in the 7T-model M7T(W) iff A
is true at W in the 12-model M.
Proof By induction on A. (1) A = pEP. Direction (M7T,W) \=
P W E <P7T
M
,,(P) W E <pCp) holds by DefA. Direction <==: If
(M,w) \= p, i.e. wE <pCp), then w E <P7T
N
(P), i.e. (N,w) \= p for some 17-
model N in S+ -by DefA, which must be the model Ml(w) or the model
M2(w) by Lemma 6.1. Either Ml(w) hangs. off M2(w) at w or M2(w)
hangs off Ml(w) at w. In either case, both p E EM\(W) and p E EM2(W)
follows from Def.2(iii). By Def.2(iv), p is true at w in MI(w) iff p is true
at w in M2(w). So (M7T,W) 1= P holds for 17 = I and 17 = 2. (2) A = ---.B
and A = B v C. Straightforward. (3) A = D7TB. (M7T,W) \= D7TB iff for all
v with WRM"V, (M7T, v) \= B; and this holds iff for all v with wR7T v, (M7T,V)
\= B (by Lemma 6.2). Now M7T = M7T(W) = M7T(V), when WRM"V, since
wand v lie in the same 7T-model and Lemma 6.1; and B E B(EM7T(V
because TecB) k L'M7T(V) (following in the case v =1= w from Def.1.3(ii),
and in the case v = w from 1.3(i) and 1.2(iii. So we can apply the
induction hypothesis and proceed: ... iff for all v with wR7T v, (M,v)
\= B, which is irue iff (M,w) 1= D7TB. (4) A = 0rr$B. O,,-B E EMrrCW) by
assumption. Either M7T hangs off M7T* or M7T* hangs off M7T at w. In
both cases, D7T+B E L'M7T'{W) follows from Def.2(iii) (in the first case by
Lemma 3.6(a), in the second because Drr*B is a 7T-constituent). By
Def.2(iv), (D7T,W) \= D
7T
*B iff (M7T*,W) \= D,,+B, and the latter is true iff
(M,w) 1= D7T+B, proved as above, except that 17 is replaced by 17*. QED.
Let us say, m satisfies the frame condition if for each 17 and L,,-con-
sistent 41 k DC
7T
(e), the models in mi4l) are based on frames for L7T.
Clearly, an m which satisfies the frame condition exists whenever both
Ll and L2 are complete w.r.t. e.
LEMMA 8. Frame Lemma.
Let M = (W,R),R2,<P) be a maximal SPRouTm-model where In satisfies
the frame condition. Then (W,RI,R2) is a 12-frame for L.
Proof By the well-known disjoint sum lemma, the disjoint sum of L-
frames is itself an L-frame. From this and (6.2) it follows that (W,R\)
TRANSFER THEOREMS FOR MUL TIMODAL LOGICS 187
is a frame for L\ and (W,R2) a frame for L2. By L = LI EB L2 and
Proposition 1, the claim follows. QED.
Since r k B(EM+(W* is true at w* in M* = MI(w*), Lemma 7
entails that r is true at w* in the 12-model M. Because LI and L2 are
complete, we can assume that the model selection function m satisfies
the frame condition; whence the frame of M is a frame for L by Lemma
8. This concludes the proof of Theorem I.
4. Applications of Generalized Transfer
THEOREM 2. Strong completeness transfers.
Proof Strong completeness coincides with completeness W.r.t. the max-
imal formula Identify e with ;;e in Theorem I. Because DC1(;;e)
= = ;;e, Theorem 2 follows. QED.
THEOREM 3. Weak completeness transfers.
Proof Weak completeness coincides with completeness w.r.t. to every
{A} with A E ;;e. Take some A and put e = B(SF(A in Theorem 1.
Because SF(A) is finite, there exist only finitely many formulae in
B(SF(A = e which are mutually non-equivalent by truth-functional
logic. Therefore there exist only finitely many formulae in DCie)
which are mutually non-equivalent in L
rr
. So every subset Ll DC,,( e)
is L7T-equivalent with a finite formula set Ll/(which may be obtained, for
example, by representing each L7T-equivalence class in Ll by one of its
members), and thus it is L7T-equivalent to a single formula (e.g. I\Ll/)'
Therefore, for each 17 E {l,2), LlT is complete w.r.t. DCie); whence by
Theorem 1, L is complete w.r.t. e and thus w.r.t. {A}. QED.
It is also possible to define certain degree-relativized notions of com-
pleteness that lie between weak and strong completeness. These also
transfer. Say L is (a,f3)-complete iff L is complete W.r.t. every formula
set Ll such that dl(Ll) < a and d2(Ll) < {J, where a, {J w + 1. As spe-
cial cases, we say L is a-complete iff it is (a,a)-complete; and it is fini-
tary degree complete iff it is w-complete. A monomodal logic L1I' is
a-complete iff it is complete w.r.t. every formula set Ll such that
d
0
7T(Ll) < a w + 1]. (Recall, d
0
7T(A) is the narrow 7T-degree of A and
d7T(A) the extended degree; the difference between them disappears if A
is a monomodal formula.)
188
KIT FINE AND GERHARD SCHURZ
THEOREM 4.
(4.1) If LJ is a-complete and L2 is t3-complete, then LJ EB L2 is (a,m-
complete.
(4.2) a-completeness and finitary degree completeness transfer.
Proof For (4.1): (a,t3)-completeness coincides with completeness w.r.t.
every r such that dl(T) < a and d2(T) < 13. Take some such r, and put
B(SF(T = e in Theorem 1. Clearly, d7T(T) = d7T(e) = d7T(DC,,(e ;;::::
d07T(DC.,(e. So LJ and L2 are complete w.r.t. DC\(e) and DC2(e),
respectively, by assumption;- whence L is complete w.r.t. e and hence
W.r.t. r, by Theorem 1. (4.2) follows from (4.1). QED.
A maximal SPRouTm-model has certain properties. from whic.h we .can
read off some further results concerning stratified
the frame of M is constructed solely from isomorphic Images and diS-
joint unions of the 1- and 2-frames of the given 'small' models. Second,
the frame of M is weakly 12-separated, in the explained sense (for all
worlds w, u with w =1= u, if w u then there exists no v =1= w with W;;v v
and v u); we write Sepl2 for the class of weakly 12-separated frames.
This gives the following transfer result.
COROLLARY 2. Assume that for each 7T, L" is strongly/weakly
for a class of 7T-frames F" which is closed under disjoint unions and ISO-
morphic image. Then:
(2.1) Ll EB L2 is strongly/weakly complete for F\ EB F2
(2.2) LJ EB L2 is strongly/weakly complete for FJ EB F2 n Sepl2.
Proof For (2.1). By the completeness and because. is
closed under isomorphic images, we can Impose on n: the additional
condition that each 7T-model in Mm is based on a frame m F". The
of M(S+) then lies in F\ EB F2 because its 7T-subframes are dlsJomt
unions of 1T-frames in F ". (2.2) follows from (2.1) and the fact that
M(S+) is weakly 12-separated.
QED.
Note that the soundness-counterpart of Corollary 2.1 is contained in
Proposition 1, which entails that if, for each 7T, F" is a class of frames
for L , then F\ EB F2 is a class of frames for L\ EB L2.
Co;ollary 2.1 indicates a natural limitation of transfer of
ness for a special frame class F, for if F is either not closed under diS-
joint unions or not closed under isomorphism, then completeness for F
need not transfer. Counter-examples are readily found; for example
LI and L., are both S5-logics. Then L" is complete for the class
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 189
of all universal frames V", which is closed under isomorphism, but not
under disjoint unions; and also, L" is complete for the one element class
N" = {(w, w X w)}, which is closed under disjoint unions (for trivial
reasons), but not under isomorphism. But L\ EB L2 is neither complete
for VI EB V2 nor for N\ EB N2, because the formula o\A 02A, which
is valid in every frame of these classes, is not a theorem of LI EB L2.
We say a class F of 12-frames is closed under joins of frame-reducts
if (W,R\,R2), (W,RI"',R2"') E F implies (W,RI,R2"') E F (and thus also
(W,RJ '" ,R2) E F). If F is a class of p.-frames (j..t E {l ,2, 12}), then L(F),
the j..t-logic ofF, is the set of all valid on all frames in F.
(Since frame-validity is preserved under substitution and all rules, this
set is a p.-logic.) We obtain the following characterization result for
stratified 12-logics in terms of their 12-frame classes.
COROLLARY 3. Assume F is a class of 12-frames closed under isomor-
phic images, disjoint unions and joins of frame-reducts. Then:
L(F) is stratified, and L(F) = L(P) EB L(F2),
where p1:= {(W,R\) I (W,R\,R2) E F} and F2:= {(W,R2) I (W,R"R2)
E F}.
Proof Because F is closed under joins of frame-reducts, F = F' EB F2.
FI and P are closed under disjoint unions and isomorphic images,
because F is closed under these operations. So we can apply Corollary
1.1: since L(F") is weakly complete for F" (1T = 1,2), L(F') EB L(F2) is
weakly complete for F = P EB P. Therefore L(F) = L(FI) EB L(P),
whence L is stratified. QED.
If the frame condition on m is dropped, the sprouting construction
can be used to prove something about transfer for arbitrary logics,
whether complete or not. For example, the transfer of model-complete-
ness and of completeness with respect to general frames can be proved
in this way-a trivial result, of course, since by the well-known facts
about canonical models, every multimodal logic is complete in this
sense. However, often a modal logic is characterized by a special class
of general frames. An example is the incomplete tense logic of S. K.
Thomason (1972), characterized by a general frame where the valuation
is restricted to finite and cofinite world sets. For such cases, a non-triv-
ial transfer result can be obtained.
A general 7r-frame is a pair G = (F,P) where F is an ordinary 1T-frame
and P k Pow( WE) is a non-empty collection of 'allowable' world sets
satisfying: (i) U E P => (W - U) E P, (ii) U, V E P => (Uu V) E P,
and (iii,,) U E P => {w E WI 'fIv E W(wR"v v E U)} E P. Similarly,
190 KIT FINE AND GERHARD SCHURZ
a general 12-frame is of the form (F,P) with Fan ordinary 12-frame and
P satisfying (i), (ii), (iiit), and (iii2)' Awmodel M (p. E {1,2,12}) is based
on the general p.-frame (F,P) if M is based on F (in the ordinary sense)
and q>M(P) E P for eachp E I? Validity is defined for general frames as
for ordinary frames.
(P ,P"') is a w-generated sub/rame of a general p.-frame (F,P) (where
w E Wl) if WI"" is the set of all worlds reachable from w by an R,..-path
(where RI2:= RI u R2), R"F" = R"F () (WF" X WF"), and P'" =
{ U () WF" I U E P}. Given a set {(Fj,Pj) liE /} of general p.-frames with
pairwise disjoint world sets, their disjoint sum is the general frame (F,P)
where F is the disjoint sum of the F;'s (i E l) in the ordinary sense and
P = {V!;;;; WF J. V () WFi E Pj for all i E l}. As is well known, the disjoint
sum of general frames is itself a general frame, and validity on general
frames is preserved under disjoint sums and under w-generated sub-
frames.
Assume Ow is a class of general1T-frames (for 1T = 1 and 2). The join
GI EB G2 of 01 and G2 is defined as {W,RI,R2), P) I W,RI),P) E GI
and W,R2),P) E G2}. We first note that Proposition I generalizes to
general frames (the proof is similar). So, G(L, EB L2) = G(LI) EB G(L2)
holds for all 1- and 2-logics. This entails, as a soundness transfer, that
Gl EB G2 is a class of general frames for Ll ffi L2 whenever, for each 1T,
G
w
is a class of general frames for Lw'
For any given 1T-model (p.-model) M, a minimal general frame GM
underlying M can be defined by putting GM: = (FM,PM) and pM: =
{IIAIIM I A E 9;}, where IIAIlM = {w E WM I A true at win M} is the truth
set of A in M. As is evident, pM is the smallest subset of Pow(M) con-
taining {q>M(P) I pEl?} and satisfying (i), (ii), and (iiiw) iiit) and (iih
above; and any formula set .1 closed under substitution which is valid
in M is also valid on GM' A given class of general 1T-(p.)frames G is
closed under subalgebras if (F,P) E G implies (F,P*) E G for every non-
empty P'" P satisfying conditions (i), (ii), and (iiiTT) iii,) and (iii2.
Let GSepl2 be the class of general frames based on frames in Sepl2.
Proposition 3 is a counterpart of Corollary 2 for general frames, but
restricted to strong completeness.
PROPOSITION 3. Assume that for each 1T, L" is strongly complete for a
class of general1T-frames G
TT
which is closed under isomorphic images,
generated subframes, disjoint unions, and subalgebras. Then L =
LI EB L2 is strongly complete for GI EB 02 and for Gl ffi G2 n GSep12.
Proof Given an L-consistent formula set .1, we identify the initial set r
with .1 u L (which also is L-consistent). Thus E(M*) = E, and the
TRANSFER THEOREMS FOR MUL TlMODAL LOGICS 191
formula space e equals 9;. There exists a maximal m-sprouting S+ of
M"', where m can be required to satisfy the following additional condi-
tions: (a) the minimal general1T-frame GM of each 1T-model Min Mm lies
in G
TT
(by completeness assumptions and closure of G
TT
under isomor-
phism and subalgebras), and (b) GM is wM-generated (by the closure of
GTT under generated subframes and the fact that the truth of formulae
at WM in M is preserved under the passage to the wM-generated sub-
model of M). The corresponding maximal 12-model M(S+) makes r
true at w*, by Lemma 7. It remains to show that M(S+) is based on a
general frame in G\ EB G2. Let G
TT
+ = W+, R
TT
+), P
TT
+) be the disjoint
sum of all the minimal general1T-frames GM of 1T-models Min S+. So
PTT + = OIAIIMTT + I A E 9;}, where MTT + is the disjoint sum of al11T-mod-
els in S+. G
TT
+ lies in G
m
by closure under disjoint sums. Moreover, it
holds that EM(w) = E for each ME S+ and w E M (because E(M*) =
E; and given E(M) = E, EM(w) = E holds for each win M since Mis
wwgenerated, and E(N) = E holds for each N hanging off Mat w). It
follows, by Lemma 7, that M\+, M2+, and M(S+) agree on all
mulae, whence p\+ = P2+ = PM(S+). Therefore the minimal general 12-
frame GM(s+) = W+,R\+,R2+), PM(S+ lies in GI ffi G2; and it is
readily verified, just as before, that it lies in GSep12 as well. QED.
We turn to the transfer of properties distinct from completeness. L
has the jini,e model property iff every L-consistent formula A is true
at a world w in a model of L with a finite world set. As is well known,
f.m.p. is equivalent to the finite frame property; i.e. we can assume
that the finite modei making A true at w is based on a frame for L. To
prove the transfer of f.m.p. we introduce the notion of the n:-alternation
degree ad.,,(A) of a formula A. This is, informally speaking, the maximal
number of alternations of nested 01- and 02-operators in A, where an
'invisible' Orr-operator is put in front of A if the B E TC(A) with maxi-
mal adwCB) is a 0TT*-formula. The inductive definition is as follows
(for 1T = 1 and 2): (1) For pEl?, ad,,(p) = O. (2) adi -,A) = adiA). (3)
adwCA v B) = max({adrr(A), adiB)}). (4) adTT(orrA) = ad.,,(A). (5) ad.,,(OTT.A)
= adTTo(A) + I (here, the inductive definition 'switches' between 1T and
1T*). Some examples: adTT(r:Jt,p) = 0, ad.,,(r:Jt,.p) = 1, = I,
= 2, = 2, = 3 (etc.), for
k, I, m > O. For a formula set .1, ad
TT
(.1) is the supremum of
{adwCA) I A E .1}.
Some important properties of ad
1T
are stated in Lemma 9. (9.1) tells
us that the operation C." reduces the 'switched' alternation degree of for-
mula sets. If M is a 1T-model, we identify M's alternation degree, adeM),
with ad,,(E(M. If, furthermore, M is an element of a sprouting S of
192
KIT FINE AND GERHARD SCHURZ
M*, then dist(M) denotes its distance from M*: dist(M) = n when
(M* = No, NI, ... , Nn-I, Mn = M) (n 2: 0) is the hang-off chain of M in
S defined before Lemma 3. Lemma 9.2 concludes that adeM) decreases
with increasing dist(M), with the effect that all ME S with dist(M) =
ad(M*) + 1 have a constituent set contained in I?, and all ME S with
dist(M) > ad(M*) + 1 have an empty constituent set.
LEMMA 9.
(9.1) For any Ll ; with ad
1T
CLl) > 0: ad1T.(C1T(Ll = ad,,(.1) - 1.
(9.2) Assume S is a sprouting of M* and ME S. Then: (i) if
adeM) Cf: I?, then adeM) ::5 ad(M*) - dist(M); (ii) if dist(M) =
ad(M*) + 1, then E(M) k I?; (iii) if dist(M) > ad(M*) + 1,
then r(M) = 0.
Proof (9.1) If ad
1T
(.1) = w, the claim is trivial; assume ad(.1) is finite.
For every D
1T
oA E C,,(.1) (there exist some because of ad1T(.1) > 0),
ad
1T
.(D."oA) = ad"(D
1T
oA) - 1 by def., and ad
1T
(D
1T
*A) ::5 ad,,(.1). In reverse,
for every B E .1 with ad" (B) > 0 there exists a D1T oA E C,,(B) with
ad
1T
o(D
1T
oA) = ad
1T
(B) - 1; in particular for aBE Ll with adnCB) =
adnC.1)
(9.2) (i) Induction on dist(M). If dist(M) = 0, then M = M* and the
claim holds trivially. Assume dist(M) > 0, and let M be a 7T-model.
Then M hangs off exactly one 7T*-model N (Lemma 3.4), with dist(N)
= dist(M) - 1. Since E(M) Cf: I?, adeN) > 0. Moreover, adeM) =
ad(E(M = ad
1T
(S".(EN(wM) (where S,,( -) = SC(C,,(-); and so
adeM) = ad,,*(EMWM - 1 by Lemma 9.1. Since ENCwM) k E(N),
adeM) < adeN). But ad(N)::5 ad(M*) - dist(N) holds by indo hyp.;
which gives ad(M)::5 ad(M*) - dist(M). (ii) follows from (i) (ad(M)
cannot become negative) and (iii) from (ii) (every N indirectly hanging
off some M with E(M) I? has empty E(N). QED.
Lemma 9.2 makes possible a proof of transfer for f.m.p.
THEOREM 5. Finite model property transfers.
Proof We put the following additional conditions on the model selec-
tion function m. For every finite L,,-consistent subset Ll k DC,,(e) and
(M,w) E m
1T
(.1), (a) M is based on a finite frame for Lrro and (b) if
TC(.1) I? (i.e. E(M) I?), then M is based on a singleton 7T-frame for
Lrro with the valuation function satisfying the requirement that for all
p E TC(.1), 'P".(P) = {w} if p E .1 and 'P".(P) = 0 if -.p E .1. Condition
(a) can be satisfied because LI and L2 have f.m.p. Condition (b) can be
satisfied by Makinson's result mentioned below Proposition 2, and
,.
i.
TRANSFER THEOREMS FOR MUL TIMODAL LOGICS 193
because 'Pm defined in this way, makes the singleton model M a model
for .1 = E(M) u T ".(E(M) at w, since E(M) I? and T ,,(E(M) con-
sists of tautologies.
Assume A E ;;e is L-consistent. E(M*) = SC(A) is finite, so ad(M*) =
n for some n E N. Because L\ and L2 have f.m.p., they are weakly com-
plete. So there exists the 12-model M+ of a maximal element S+ in
SPROUT
m
which makes A true at w'" and satisfies the additional condi-
tions (a) and (b) for m. Given (a) and Lemma 8, M+ is based on a 12-
frame for L. Given (a) and Lemma 9.2, every model ME S+ with
dist(M) = ad(M*) + 1 is based on a singleton frame, off which no
model can hang. So there exist no models N in S+ with dist(N) >
ad(M"') + 1. Because every model in S+ has a finite world set (condi-
tion (a and a finite distance ::5 n + 1, there exist only finitely many
models in S+ and thus only finitely many worlds in M+. QED.
As is well known, if a logic L is recursively enumerable and has
f.m.p., then L is decidable, i.e. there exists an effective algorithm deter-
mining whether or not an arbitrary formula A is in L. So Theorem 5
implies that L\ EB Lz is decidable whenever LI and L2 are recursively
enumerable and have f.m.p. However, under the condition of weak
completeness of L\ and L2 it is possible to establish a direct transfer
result for decidability. This is proved by means of a reduction of
L\ EB (theoremhood) to L,,-consistency (theoremhood).
Recall that for any finite formula set .1, T,,(.1) can L,,-equivalently be
represented by a finite set, which we denote by T "f(.1).
THEOREM 6. Reduction Theorem.
For all weakly complete L\, L2 and 7T E {1,2}: For any .1 k 5: Ll is
L\ EB L2-consistent iff .1 uT ,,(.1) is L,,-consistent.
COROLLARY 4. For any A E;: A E L\ EB L2 iff (I\T"f({A}) ---7 A) E
L
7T

Proof Theorem 6: Direction =*: If .1 u T ,,(Ll) is L

7T
-i'nconsistent, then
-; I\(r\ u r2) E L7T for some finite r, .1 and finite r2 T ,,(.1)
L, EB L2; whence -.I\r\ E L\ EB L2; so .1 is LI EB L2-inconsistent.
Direction :::: If .1 u T ,,(.1) is L
7T
-consistent, there exists a labelled 7T-
model M* based on an L,,-frame making .1 and T ,,(.1) true at w*, where
T ".(.1) = T ,,(E(M*, E(M*) = SC(Ll). This implies, by the proof of
Theorem 1, that a maximal sPRouTm-model exists which is based on a
frame for L, EB L2 and makes .1 true at IV*. SO .d is L\ EB L2-consistent.
Corollary 4 follows straightforwardly from Theorem 6. QED.
194 KIT FINE AND GERHARD SCHURZ
We say, a property .p transfers under the condition C if. LI $ L2 has
.p whenever both LI and L2 have .p and satisfy C. If C transfers
(in the unconditioned sense) and .p transfers under condition C, then the
conjunction of.p and C transfers (in the unconditioned sense).
THEOREM 7. Decidability transfers under the condition of weak com-
pleteness.
Proof If A is finite, Bf(A) denotes afinite
of B(..::!), containing for each A E B(A) a truth-functIOnally equivalent
formula. Consider the following decision procedure for L =
LI $ L2. Given A E :: (1) Construct Bf(S,,(A. (2) For all
B E Bf(S,,(A, apply to test whether BEL. (3) Constructl:!
on the basis ofthe result of step 2. (4) Test whether A E L by testmg
whether T "f( {A}) --+ A E L". Steps 1 and 3 terminate since the sets
Bf(S,,(A and DC,,(A) are finite and can be constructed. The non-
trivial step is step 2, which is recursive. For each BE Bf(S,,(A,
ad".(B) < ad,,(A) by Lemma 9.1; so step 2 is well-founded, reducing
'switched' alternation degree. Step 4 terminates by Theorem 6 and
assumption. So is an algorithm for decidability in L, terminating
in step 4. QED.
We turn to interpolation. For .p a formula or formula set in :, :,,(.p)
is the n-Ianguage of .p, that is the set of all wffs constructible from the
1T-constituents C,,(.p) with the help of the logical symbols of L".
Correspondingly, the 12-language of.p, :n(.p), is the set of all wITs con-
structible from the propositional variabies of .p. Let p. E {1 ,2, i2}. A
logic L has the interpolation property if for every (A --+ B) E L,. there
exists C E : (A) II :,.(B) with (A --+ C) E L,. and (C --+ B) E L,..
To take care of th: case..., A E L,. or BEL,., it is assumed that the log-
ical symbols of L contain the logical constant T. Call a pair (Ll),A2) of
:-formula sets L -inseparable if there exists no A E :,.(LlI) II :,.(Ll2)
with A I" A and J2 "L ..., A. (Observe that L,.-inseparability of (A \,A2)
L,. ,. . b (0 h
entails that both A I and A2 are L,.-conslstent, ecause.L con tams t e
constant T.) A logic L has the joint consistency property if for every
L -inseparable pair (A(,A2), the union AI U A2 is L,.-consistent. As is
:en known, interpolation is equivalent to the joint consistency prop-
erty. (As before, we simply write :/L for :dL\2.) For any formula set
A, let A's logic L(A) be defined by: L(A) = L II :(A). We will prove
joint consistency transfer under the condition of weak completeness by
adding these logics to two given L-inseparable sets Lli and .12. It follows
that a maximal SPRouTm-model exists for AI U .12. For any formula set
A, let .1 + be the set A u L(A).
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 195
LEMMA 10. If (.1\,A2) is L-inseparable, then (A\+,A2+) is LI- and L2-
inseparable.
Proof If (AI+.A2+) is not L,,-inseparable, then there exists an A E
:,,(A I +) n :,.(A2+) with .11 + Iy", A and A2 + Iy", ..., A. It follows that A E
:(AI) II :(.12), and further that AI fJ. A and A2 fJ. ...,A (because L(.1;)
L for i = 1,2). So (A\,A2) is not L-inseparable. Contradiction. QED.
Lemma 10 implies, given (AI,A2) is L-inseparable and the joint con-
sistency property of L", that L:l!+ U Ll2+ is L".-consistent and thus true
at w'" in an initial 1T-model M'" for L". We identify E(M*) with
SqA I u A2)' For any 1T-model M in a sprouting of M*, we split up its
constituent set and its agreement diagrams through the two language
components of :(AI) and :(A2). So we put Ei(M) = E(M) n :(.1;) and
DiMCw) = DMCw) II :(Aj) (i = 1,2). The role of the theorem sets
T ,,(E(M is now taken over by the logics L(Ei(M; more precisely,
Def.l.2(iv) has to be replaced by the condition that M makes
L(EI(M u L(1:2(M true at WM (everything else remains unchanged).
A labelled model satisfying this modified condition is called a bilabelled
model. Thus, we consider sproutings of M* with help of bilabelled mod-
els (Defs.2-4 apply without change). In order to continue the sprouting
construction we must guarantee that for each bilabelled 1T-model M,
(DIMCw), D2MCw is L-inseparable. Together with Lemma 10 this
implies that 'Dlu(w)+ u D2MCw)+ is L,,+-consistent, provided L7T+ has
the joint cons.istetlcy property.
LEMMA II. Assume M is a bilabelled 1T-model
El(M)+ U 1:2(M)+ true at WM. Then for each IV
(DIMCw), D2MCw is L-inseparable.
making
in M:
Proof Assume there exists an A E :(DIMCw II :(D2MCw with
DIM(w) fJ. A and D2MCw) fJ. ...,A. So there exist finite sets III DIMCw)
and Il2 DIMCw) with (I\IlI --+ A) ELand (I\Il2 --+ ...,A) E L. Now
DiMCw) :(Ei(M (i = 1,2) (because DiM(W) :(Ll;) and DiM(W)
E(M, whence D7T
n
(I\IlI --+ A) E :(El(M and D"n(I\Il2 --+ ...,A) E
:(1:2(M. So D"n(I\IlI --+ A) E L7T(El(M and D7T
n
(I\Il2 --+ ...,A) E
L,,(1:2(M for n = dist(wM,w); whence I\Il\ --+ A and I\Il2 --+ ...,A are
true at w. Because both III and Il2 are true at w, both A and ..., A must
be true at win M. A contradiction. QED.
THEOREM 8. The interpolation property and joint consistency property
transfer under the condition of weak completeness.
196 KIT FINE AND GERHARD SCHURZ
Proof Given an L\ E9 L2-inseparable pair (LI\,Ll2) of formula sets,
Lemma 10 guarantees that LI\+ u Ll2+ is true at w* in a bilabelled 7T-
model M* for L"., and Lemma II together with Lemma I 0 guarantees
that, for each bilabelled 7T-model M and win M, DIM(W) U D2M(W) and
consequently DIM(w)+ u D2M(w)+ is L-consistent and hence L
1T
+-
consistent; and so M* can sprout indefinitely. Therefore there exists a
maximal SPRouTm-model, where m satisfies the frame condition by
assumption (Lemmas 3-6 apply as before). It is a 12-model based on a
frame for L\ E9 L2 (Lemma 8) and makes LI\+ U Ll2+ and consequently
LI\ u Ll2 true at w* (Lemma 7). So Lll U Ll2 is Ll E9 L2-consistent.
QED.
Let 11- E {I,2,12}. L,. is Hal/den-complete iff for all A, BE;t,. with
;t,.(A) n ;t,.(B) = ;t(T), (A v B) E L,. implies A E L,. or BEL,.
(Hallden 1951). (If T is not a primitive of ;t, ;t(T) has to be replaced
by 0.) The equivalent consistency version of Hallden-completeness is
this: if Lll,Ll2 C ;t,. are L,.=consistent and ;t,.(Lll) n ;t,.(Ll2) = then
LI\ U Ll2 is L,.-consistent. Hallden-completeness follows from interpola-
tion if L1T (L) contains -,0
1T
.L (-, 01.L, -,02.L). A direct transfer of this
property is possible in a way similar to the proof of joint consistency
transfer.
PROPOSITION 4. Hallden-completeness transfers under the condition of
weak completeness. (Proof omitted.)
Transfer of decidability, interpolation, and Hallden-completeness has
been proved only under the condition of weak completeness. It would be
interesting to know whether these results hold also for incomplete logics.
We finally mention two simple results which do not rely on the con-
struction used so far but which are nevertheless naturally classified as
transfer results. A logic L,. (11- E {1,2,12}) is c-canonical, for c a cardi-
nal, if the frame of its canonical model M in :t,. is a frame for L,., where
:t,. contains c many propositional variables, or atoms, respectively.
Note that if a logic is c-canonical, it is c*-canonical for every c* < c (the
other direction does not generally hold). A logic L,. is called natural if
every natural model for L,. is based on a frame for L,.. (For the defini-
tion of a canonical and a natural model see Fine 1975: 18 and 23-4.)
We let X range over infinite cardinals.
PROPOSITION 5.
(5.1) If L\ is X-canonical and L2 is X*-canonical, then L\ E9 L2 is
min(X,X*)-canonical.
TRANSFER THEOREMS FOR MULTI MODAL LOGICS 197
(5.2) X-canonicity transfers.
(5.3) Naturalness transfers.
We the proofs because they are straightforward. The proof of
transfer has to be performed in the ambiguous language !
With mfimtely many atoms. Therefore it applies only to canonicity with
respect to infinite cardinals. We choose a set I? with II?I = min(X,X*).
The proof rests on the fact that the frame of the canonical model for
LI ffi L2 in ! consists of generated 7T-subframes of the frame of the
canonical model for L1T in :t. The transfer of naturalness follows from
the fact that a natural 12-model for L\ ffi L2 is a join of natural 7T-mod-
els for L1T in the ambiguous language :t.
5. Generalizations for Stratified Multimodal Logics
!o o.btain generality. we co?sider (possibly infinite) joins of (possibly
mfimte) multlmodal logiCS which are themselves not necessarily strati-
fied. In what follows, Q is always a non-empty set of pairwise disjoint
non-empty index sets IE Q (/, J, K are used as index sets in Q). !I is the
languag: with modal operators OJ for each i E I. = !Un is the lan-
guage With modal operators OJ for each i E UQ. The notion of a normal
(multimodal) logic in :tK, K k: UQ, is as explained in Section 1. Given
normal logics LI k: !J, their join E9{LJ I IE Q} is the minimal normal
logic in :tn which contains LI for each IE Q. A logic LJ k: ;;eJ is called
stratified if there exists an Q with IQI > 1, J = UQ, and for each IE Q
a normal logic LJ such that LJ = E9{LJ I IE Q}; the logic LJ is called
monostratified if III = 1 for each IE Q. Frames for :tl, called I-frames,
have the form (W,R/) with W=I= 0 and RI an indexedset of relations
R; k: JVl for each i E I; frames for ;;en, called Q-frames, have the form
(W,Rn) with Rn an indexed set of relations Rj for each i E UQ. Rn iI
is the I-reduct of Rn defined by Rn i I = (Rj E Rn liE I). If Flare
classes of I-frames for each IE Q, then ffi{FJ I IE Q} =
I (W,Rn i I) E FJ for all IE Q}. Again, for any logic L in ;;e,
F.(L) IS the class of :t-frames for L. The other notions of Section 1 apply
dIrectly.
We say that a property tP generally transfers (under condition C) if
whenever tP (and C) holds for all LJ (/ E Q), tP holds for EB{L/ I IE Q}.
In what follows we state the generalizations of the results of the
previous chapters for stratified multimodal logics; we denote these
198 KIT FINE AND GERHARD SCHURZ
generalizations just by adding '''''; so 'Theorem 1 "" is the generalization
of Theorem 1 for bimodal logics, etc. Proofs are omitted when obvious
or parallel to the bimodal case.
Proposition 1 '" says that F(E9{L/I IE .o}) = E9{F(L/) I IE .o}.
Proposition 2'" tells us that if all LI are consistent, then E9{L/I I E .o}
is a conservative extension of LI for each IE.o; however, R. H.
Thomason's proof (1980) establishes this result only under the restric-
tion that all I E .0 are countable. The same restriction then applies to
Corollary 1'" saying that consistency generally transfers.
Turning to Theorem 1 '" we treat !en as an ambiguous language,
abbreviated by !e. The set of I-constituents of !en, J:I, is the set
IP> u {OjB E !e If.E (U.o - 1)}. The set of all constituents, J:, is
U{J:III E .o}. An I-model based on an I-frame (W,R[) has the valua-
tion function <PI: J:I Pow(W). A .o-model based on an .o-frame
(W,Rn) has the valuation function <P: IP> Pow(W). L[ : is an I-logic
over ; if it contains (Taut), (Ki) for all i E I, and is closed under MP,
(Ni) for all i E I, and under substitution of !e-formulae for elements in
J:I. An .o-logic Lover !e is defined in the standard way. SF(A) is the set
of all subformulae of A (viewed as an !e-formula); SF ICA) is the set of
. all I-subformulae of A, i.e. all subformulae of A which do not properly
lie in the scope of some 0roperator with j E (U.o - 1). CICA) = SF ICA)
nJ:I is the set of all I-constituents of A; SqA) = SF(A) n J: the set of
all subconstituents of A. All the notions apply to sets in the obvious way.
TqA) and B(LI) is defined as in Section 2.
An additional notion is needed in the case of infinite sums. A }-atom
is any formula which is a propositional variable or of the form OjA with
j E J; AJ is the set of all J-atoms in !e. C/J(A) = SF ICA) n AJ is the set
of IJ-constituents of A; i.e. all l-subformulae which are J-atoms, and
SIJ(A) = SC(C[J(A the set of all IJ-subconstituents of A. The parallel
with the bimodal case is this. In the infinite case, we will hang off at any
(non-base) world W in an I-model M a J-model N for each J =1= I. So
corresponding to the agreement sets S1T(J:MCw in a rr-model M are the
sets SIJ(J:M(w of IJ-constituents for each J =1= I; we call them the J-
agreement sets. The extended I-degree of a formula A, dI(A), is the num-
ber of nested in A with i E I; the inductive definition
consists of clauses (i)-{iii) of Section 2 and (iv): for i E I: dI(OjA) =
dI(A) + 1, (v): for j E J, J =1= I: dI(OjA) = dI(A). dI(LI) is the supremum
of {dI(A) I A ELI}. cJOI(A) is the narrow I-degree of A, defined by drop-
ping (v) and replacing (i) by dOI(A) = 0 for A E J:I.
0K
n
stands for any string of n operators Ojl ... Ojn with i ..... , ill E K. Say
an index i occurs in a formula set LI if a OJ-operator occurs in some
..
"
'\'
'1:.
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 199
formula in LI; an index set I occurs in LI if some i E I occurs in LI. For
any LI !e, put IIJ :={i E I I i occurs in LI} and .oIJ := {I E .0 I I occurs
in LI}. The I-degree-conserving closure of LI, DCICLI}, is the set
{O/dnB I BELl and dI(O/"nB)::; dI(LI)}.
Theorem 1'" asserts that for any formula space e ;;E, if L/ is com-
plete w.r.t. DeICe) for each IE .0, then L = E9{L/I I E .o} is complete
w.r.t. e.
Let us give the relevant definitions from the proof.
Definition 1 *: (1 *.1) For any LI :, LI's set of IJ-theorems TI.l(LI) is
defined by TIJ(LI) = {O/lB I B E B(S/J(LI n L, dI(O/dnB)::; dI(LI)}.
(1 *.2) A labelled I-model is a triple (M,WM,J:(M (abbreviated by M)
where (i) M is an I-model for L/, (ii) W M E WM, (iii) J:(M) is a set of
constituents closed under subconstituents, and (iv) for each J E
.oE(M) - {I}, M makes T/J(.E(M true at WM. (1*.3) For each world w
in a labelled I-model M, a constituent set .EMCw) is inductively defined
by: if i E I, uRjv (u,v E WM) and DjA E .EMCu), then SqA) .EM(V).
If l:(M) is finite, then .oE(M) as well as h(M) is finite. So the union of
all theorem sets which have to be made true by Mat WM according to
condition (iv) can be Lrequivalently represented by a finite formula
set. An I-path in M (which leads from Wo to W
II
and has length n 2: 0)
is a sequence Wo, ... ,WII of worlds in M where for each 1 ::; i < n,
WjRkWi+l holds for some k E I. We define the distance dist(w,v)
between two' worlds wand v in an I-model as the minimal length of an
IE(M)-path in M leading from W to v, provided such a path exists; oth-
erwise dist(w,v) = w. Observe that only relations Rj with i E h(M>
'count' in this notion of dist(w, v), which is the counterpart of the
olz"(.\l)n-closure in the definition of TIJ(.E(M. The inductive definition
(1.3"') implies that J:MCu) is empty for any u with dist(wM,u) = w (even
if u is reachable from W M by some I-path).
It follows that: Lemma 1 *: for every labelled I-model M and u in M,
(i) dI(J:MCu = dI(.E(M - dist(wM,u) if .EMCu) is non-empty, and (ii)
for u =1= WM, EMCu) is empty iff dist(wM,u) > dIC.EMCu.
Definition 2* A labelled I-model M hangs off a labelled J-model N (at
WN) if: (i) J =1= I, (ii) (as in Def.2(ii, (iii) .E(N) = S/J(.EMCWN) , and (iv)
(as in Def.2(iv. The sets S/J(.EM(W are called J-agreement sets, their
diagrams DJMCw) are called J-agreement diagrams.
Lemma 2* For any labelled I-model M, W in M and J =1= I, the J-
agreement diagram is L-consistent. This follows for J E .oE(M) as in the
bimodal case and holds for J E (.0 - .oE(M for the trivial reason that
then DJMCw) = 0 holds.
Definition 3* is of a sprouting S of M*. Conditions (i) and (ii) are the
200 KIT FINE AND GERHARD SCHURZ
same as in the bimodal case, but condition (iii) has to be modified,
because models hanging off the same model may share a world. It says
now that two models M and N in S have a world w in common in only
two cases: either one hangs off the other at w, or they are models of dif-
ferent I-type both hanging off a common model PES at w.
Lemma 3* says that any sprouting S of M* satisfies the following con-
ditions: (3*.1) For every I-model ME S, wE WM and J E Q - {l}, at
most one N E S hangs off M at w. (3*.2)-(3*.5) are as in the bimodal
case, except that in (3.3*), 'n' has to be replaced by 'I'. (3*.6) For every
I-model ME Sand w E WM: (a) if M hangs off N, E(M) k E(N);
(b) E(M) k E(M*) k e; (c) for each J =1= I, DJM(w) k e; (d) for each
J =1= I, T/J(E(M k DCI(e). (In the proof of (3*.2) one has to take care
of the two cases of Def.3*(iii); the rest is as in the bimodal case.)
A regular model selection function m is defined as in the bimodal
case, with 'n' replaced by 'l', except that we must put
Xm = max({I.oI, Xo, sup({IWMII M E Mm})}) (where Mm
{M I (M,w) E mICLl), Ll k : is Lrconsistent, I E Q}) to ensure that SETm
contains enough elements. S is an m-sprouting if for each I-model
ME S, (M,WM) E mI(E(M) u U{T[J(E(M I J E .oE(M) - {l}});
SPROUT
m
is the set of all m-sproutings of M*. Lemma 4* tells that
SPROUT
m
has a maximal element. Lemma 5* says that for every maximal
element S+ in SPROUT
m
, every I-model Min S+ and world w in M which
is a non-base world of M if M =1= M*, and for every J E (Q - {l}): there
exists an J-model N E S+ hanging off M at w.
Definition 4*. We define, for every maximal element S+ in SPROUT, its
D-model M(S+) by: (i) WM(S+) = U{ WM I M E S+}; (ii) for each
I E Q: RnM(S+)j I = U {Rfli M E S + and M is a I-model}; (iii)
fJq>M(S+)(p) = U{q>fI(P) I M E S+ is an I-model, IE Q} for pEP.
Lemma 6* tells us that for every maximal SPRouTm-model M(S+) =
(W,Rn,q the following holds: (6*.1) Every w E W lies exactly in one 1-
model of S+ for each lED, denoted by MI(w), and there exists exactly
one K E Q such that for all J =1= K, MJ(w) hangs off MK(w) at w. (6*.2)
(W,Rn) is the disjoint sum of all the frames (WM,RIM) for all ME S+.
Assume M = (W,Rn,q is a maximal SPRouTm-model. Then:
Lemma 7*. Agreement Lemma.
For every lED, wE Wand A E B(EMICw: A is true at w in the 1-
model MI(w) iff A is true at w in the Q-model M.
Lemma 8*. Frame Lemma.
If m satisfies the frame condition, then (W,Rn) is an D-frame for L.
The proof is analogous to the bimodal case. But one remark on
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 201
the induction step A = 0kB with k E K and K =1= I in Lemma 7* is
appropriate. By Lemma 6*.1 there are three cases: (a) MI(w) and
MK(w) hang off some J-model Nat w (J =1= I, J =1= K); (b) MI(w) hangs
off MK(w) at w; or (c) MK(w) hangs off MI(w) at w. In all three cases
0kB E EMKCw) follows, thus the induction step can be performed in the
model MK(w). This concludes the proof sketch of Theorem 1*.
L is (aliI E Q)-complete [a-complete] if it is complete w.r.t. every
formula set Ll with dI(Ll):S al(:S w + 1) for each IE Q [with dI(Ll):S a
(:s w + 1)]. LI is a-complete if it is complete W.r.t. every Ll such that
dOI(Ll) :s a (:s w + I). We have:
Theorems 2*-4*. Strong, weak, a-, and finitary degree completeness
generally transfer. (4*.1. If for each lED, LI is aI-complete, then
E9{LIII E Q} is (aI II E D)-complete.) Similarly, the analogues of
Corollaries 2 and 3 can be stated and proved.
The notion of a general frame is defined as in the monomodal case.
The join E9{GII IE Q} of the classes GI of general frames is defined as
{W,Rn),P) I W,RntI),P) E GI}. The other notions generalize
directly to the multimodal case. Proposition 1 * generalizes to general
frames in the obvious way.
Proposition 3*. If for each IE Q, LI is strongly complete for a class
of general I-frames Gf closed under isomorphic images, generated sub-
frames, disjoint unions, and subalgebras, then E9{Lfl IE Q} is strongly
complete for' E9{GIII E Q} and for E9 {GIll E.o} n GSepg.
The distance dist(M) of a model M in SPROUT
m
is defined as in the
bimodal case. The inductive definition of the I-alternation degree adICA)
of a formula A has clauses 1-3 of the bimodal case (but with I instead
of n), and (4): If-tE I, then adICojA) = adICA); (5): if j E J =1= I, then
adICOjA) = adJCA) + 1. If Mis anlmodel, adeM) = adICE(M.
Lemma 9*.1: For any Ll k: with adICLl) > 0 and J =1= I,
adJCCIJCLl :s adICLl) - I holds. It is proved as in the bimodal case;
except that ':s' stands instead of '=', because a formula A ELl with
adICA) = adICLl) need not have /J-constituents B with adJCB) =
adICA) - 1 for every J; only for some J. Lemma 9*.2 says the same as
in the bimodal case.
The proof of Theorem 5 in the bimodal case was based on the fact
that every monomodal logic is valid on a singleton frame; this enabled
us to terminate hang-off chains in sproutings at models M with
E(M) k P. By Proposition I"', this is also true for every monostratified
multimodal logic. However, it does not hold in general for unstratified
multimodal logics LI. In fact it is possible to formulate interactive
axioms implying a certain lower bound for the cardinalities of their
202 KIT FINE AND GERHARD SCHURZ
models (and hence of their frames). Consider, as an example, the
bimodal logic L n contammg as its sole axiom scheme:
02(OlnT 1\ 0In+1-1). The following facts about Ln are readily verified:
Fact 1. Any 12-model for L n has more than n worlds (Proof-sketch.
M 1= L n implies that (M,u) 1= OInT 1\ 0ln+ 1-1 for some u in M, which in
turn implies that M must contain an RI-path u = Wo, ... , Wn consisting
of n + 1 RI-irreflexive worlds.)
Fact 2. L" is valid on the 12-frame F" : = 0 Rt 1 Rt2, ... , n-l Jr(n,
where '0', ... , 'n' are Rt-irreflexive worlds, and R2 is universal. (Proof-
sketch. M 1= 02( OinT 1\ 0ln+ 1-1) because (M,O) 1= OinT 1\ 0ln+ 1-1.)
As S. K. Thomason (1972: 153-4) has shown, there are even multi-
modal logics which are valid on no frame. Also observe that the valid-
ity of a multimodal logic L on singleton frames has an interesting
equivalence in terms of stratification: L is valid on a singleton frame iff
it has a monostratified extension L"'. (Direction => holds because the
logic of a singleton frame for L is monostratified; direction ~ holds
because every monostratified logic is valid on a singleton frame.)
To prove Theorem 5* in the general multimodal case, we will proceed
differently. We assume that the sprouting grows only for IE Qr (r:=
{A} is the initial formula set). We will then 'multiply' the sprouting con-
struction finitely many times to the point that the number of worlds w
in I-models M with J:(M) ~ I? is sufficient to accommodate a frame for
LJ for each J E (Qr - {I}) and is, in addition, sufficient to accommo-
date a frame for each IE (Q - Qr). However, this can be done only
under the additional assumption that the frame cardinalities of
{L/I IE Q} are finitely bounded, i.e. there exists an n E N such that
each L/ is valid on an I-frame with cardinality n/ :5 n.
Theorem 5*. The finite model property generally transfers, provided
the frame cardinalities of {L/ II E Q} are finitely bounded.
Proof Take an L-consistent A E ;e, put r;= {A}, and consider an
m-sprouting S of M* (making A true at w"'), where m satisfies the addi-
tional condition that for every IE Qr, Ll ~ DCICe) and (M,w) E m(Ll),
M is based on a finite frame for L/. S shall satisfy the condition that if
a J-model N hangs off an I-model in S, then J E (Qr - {I}). S shall
grow up to models with dist(M) = ad(M*) + 1. So for each I-model
ME S with dist(M) :::; ad(M*) and IV in M (IV a non-base world of M
if M *" M"') there exists a J-model hanging off M at IV for each J E
(Qr - {I}) (argued as in Lemma 5"'); while J:(M) ~ I? holds for all
models M with dist(M) = ad(M"') + I, and J:M(IV) = 0 holds for all of
their non-base worlds. S has only a finite number of models (because
Qr is finite and each world set is finite) and so only a finite number of
. ,.
. I;
TRANSFER THEOREMS FOR MUL TIMODAL LOGICS
203
worlds. Note that if L/ has a finite frame of cardinality n/, then it also
has a finite frame with cardinality n[.k for every kEN (by formation
of disjoint sums). For each IE Qr, let k/ be the number of all non-base
worlds in I-models ME S with dist(M) = ad(M"') + 1; we can these
worlds {-ultimate. Let k be the number of all worlds in S. Let Sf be the
disjoint union of nl copies ofS, where n is the finite bound of frame-car-
dinalities for {L/ I IE Q}. For each IE Qr, there exist n!.k/ I-ultimate
worlds in Sf; and there exist n!.k worlds in Sf. Since n!.k[ and n!.k are
integer multiples of nJ for each J E Q, there exists a frame for LJ with
cardinality n!.k/ and one with cardinality n!.k. Let M(Sf) be the Q-model
corresponding to Sf (by Def.4*). We impose on the set all I-ultimate
worlds in M(Sf) a frame for LJ for each J E (Qr - {I}). Moreover, we
impose on the set of all worlds in M(Sf) a frame for LK for each K E
(Q - Q,..). Call the reSUlting model M(Sf)"t. It is readily shown that this
model conforms to the agreement lemma. Also, its frame has the prop-
erty that each of its I-reducts is a disjoint union of frames for L/; and
so it is a frame for L. Moreover, it is finite, and so the claim is proved.
QED.
F.m.p. does not generally transfer in the absence of the condition that
the frame cardinalities of {L/ I IE Q} are finitely bounded. A counter-
example is given by considering the logic L(Fn) of the frame Fn in the
example above. For each n E N, L(Fn) has f.m.p. with n + I as a lower
bound of its frame cardinality (following from L n ~ L(F") and Fact 1).
But the infinite join E9{L(Fn) In E N} does not have f.m.p. since the
frame cardinalities of {L(Fn) I n E N} are not finitely bounded. (In the
infinite join it is assumed, of course, that all the logics L(F") (n E N)
have mutually disjoint index sets, e.g. {(I ,n),(2,n)}.)
For the purpose of the reduction theorem, note that for every finite
Ll ~ 5 and IE Q, U{T[J(Ll) I J E (QJ - {I})} can L/-equivalently be
represented by a finite formula set, denoted by T /f (Ll).
Theorem 6* Reduction Theorem.
Given that for all I E Q, L/ is weakly complete, then: For any
Ll ~ ;e: Ll is E9{L/ II E Q}-consistent iff Ll u T /f (Ll) is L/-consistent.
Corollary 4"'. For any A E 5: A E E9{L/II E Q} iff
(J\T/I({A}) ~ A) E L/.
General decidability transfer is only possible under two additional
restrictions.
Theorem 7*. Decidability generally transfers under the condition of
weak completeness, provided that (i) UQ is recursive, and (ii) there
exists a recursive functionftaking each i E UQ into the IE Q to which
it belongs.
204 KIT FINE AND GERHARD SCHURZ
The decision procedure 1fR(I) for ffi{L/ I I E Q} is as follows. Given
A E 9:,: (1) For all J E (Q{A} - {I}), construct Bf(SIJ(A; (2) For all
BE Bf(SIJ(A, apply 1fR(J) to test whether BEL; (3) construct T/f (A)
on the basis of the result of step (2); and (4) test whether A T/f( {A})
A E L/. Condition (ii) in Theorem 7* is required to construct the finite
set Q{A} from the formula A; condition (i) is required because the decid-
ability of ffi{L/ I IE Q} presupposes the recursiveness of UQ, which is
not implied by the recursiveness of each IE Q. (Thereby, an index
set I is called recursive if there exists an effective procedure mapping I
bijectively into a recursive set of natural numbers.)
The remaining counterpart results are proved as in the bimodal case.
The I-language of A, 9:,ICA), is the set of all wffs constructible from the
I-constituents of A 'with help of the logical symbols of L/. 9:,(A) is as in
Section 3. For LI L(LI) = L n 9:,(LI), and LI+ = LI u L(LI) (where
L = ffi{L/ I IE Q}).
Theorem 8* and Proposition 4*. Interpolation (joint consistency prop-
erty) and Hallden-completeness generally transfer under the condition
of weak completeness.
Proposition 5 *. If L/ is for aliI E Q, then ffi{ LII IE Q}
is for = I IE Q}). and natural-
ness generally transfer.
6. Transfer for Classical Logics
We consider the question of transfer for classical logics. We have only
been able to prove certain analogues of the transfer results for normal
logics, not the results themselves. We concentrate on the bimodal case,
since the extension for arbitrary joins of multimodallogics offers no dif-
ficulty in principle, as seen from Section 5.
The bimodal language Y; for classical logics is the same as for normal
logics. A subset L ::e is said to be a classical (bimodal) logic over Y; if
it contains all tautologous fonriulae and is closed under the rules of sub-
stitution for variables in P, modus ponens, and replacement of equiva-
lents (E,,): A H B/DTrA H D"B, for 7T = 1 and 2. A frame for 9:, is a triple
(W,R),R2) where W is a non-empty set, and each RTr is a subset of
W X Pow(W) (Le. R" is a relation between worlds and sets of worlds);
such a frame is called a neighbourhood frame. A model is a frame sup-
plemented with a valuation function <po Let the truth set IIAIIM of A in
M be {v in W I A is true at v in M}. Then relative to a model M, the
clause for D"A in the truth-definition is this:
TRANSFER THEOREMS FOR MUL TIMODAL LOGICS 205
(M,w) 1= D".A iff wR".IIAIiM.
We follow the previous conventions for defining the models and frames
etc. of different type 1, 2, or 12; and but for the change in the defini-
tion of frame and in the truth-clause, we define the various syntactic and
semantic notions in the same way as before.
In extending our original transfer results to classical logics there is a
basic problem. For in the case of normal logics there is a formula,
namely DnA, whose truth at a world w will guarantee that A is true at
all nth successors of w; and hence there is a theorem set whose truth at
the base world of a 7T-model will verify that portion of the logic which
enables us to hang off an appropriate 7T*-model from worlds in that 7T-
model. However, in classical logic there is no formula capable of per-
forming the role played by on; and consequently, some other way of
appropriately grounding the hang-offs must be found.
We solve this problem by taking for granted, in the assumption of
completeness for the component logics Ll and L2, that the appropriate
formulae can be made true at each of the worlds. Let us introduce the
relevant concepts. We say that a formula or formula set is (universally)
true in a model M if it is true at each world of M, and that a pair (LI,n
of formula sets is true at the world w in M if LI is universally true in M
and r is true at IV in M. We say that A is a consequence of (LI,n in the
frame F if A i's true at any world of any model based on F in which
(LI,T) is true; we say that the pair (LI,n is satisfiable in a frame F if it is
true at some world of some model based on F. A special case of conse-
quence is that in which r is empty; we then say that A is a super-con-
sequence of LI.
A formula A is said to be derivable from the pair (LI,n in the logic L
if there is a derivation in L of A from LI and r in which the application
of the replacement of equivalents is restricted to those premisses which
do not depend upon any assumption in r. The pair (LI,n is said to be
L-consistent if there is no derivation of .L from it. A special case of
derivability is that in which r is empty; we then say that A is super-
derivable from LI. (Derivation from a pair (LI,T) reduces to super-deriva-
tion from LI; for A is derivable from the pair just in case there is a
conjunction C of finitely many formulae of r such that C A is super-
derivable from LI. However, consequence from a pair need not, in the
same way, reduce to super-consequence from a set.)
Say that L is super-complete W.r.t. the formula space e if every L-con-
sistent pair of sets of formulae in e is satisfiable in a frame for L. Strong
super-completeness is then super-completeness W.Lt. the space of all
206
KIT FINE AND GERHARD SCHURZ
formulae; and weak super-completeness is super-completeness w.r.t.
finite formula spaces. It should be clear that super-completeness (10
both the strong and the weak case) includes ordinary completeness as
the special case in which J = 0. .
Just as for normal logics, it will be important for certam of our proofs
of transfer to make use of completeness W.r.t. models rather than
frames. In the present case, this result takes the following form:
THEOREM 9. Model Super-completeness.
Let L be a classical monomodallogic. Then the pair (J,l) is L-consis-
tent iff there is some model for L in which it is satisfiable.
Proof The soundness (right-to-Ieft) direction is
Suppose now that (J,l) is L-consistent. Define a model
(W,R,tp) by: (i) W =, {w I w is a maximal set for whIch (J,w)
is L-consistent}; (ii) wRV, iff V = IBI for some DB 10 w, .where IBI =
{v E WI B E v}; (iii) tp(P) = {w E Wlp E w}. We show 10 the usual
way: ,(
LINDENBAUM'S LEMMA. If (J,l) is L-consistent for some set of formulae
r, then (J,r+) is L-consistent for some maximal extension r+ of r. We
can then establish:
TRUTH-LEMMA. For any formula A, A is true at w in the canonical model
M iff A E w.
Proof The proof is by induction on A and is analogous to standar? '
proof of completeness (Segerberg 1971, ch. 1.3). The cr.ltlcal case IS
when A = DB. Now DB is true at win Miff wRIlBlIM, which holds (by
indo hyp.) iff wRIBI, which holds (by def. of R) iff IBI = Iq for some D.C
in W. Thus we need to show that DB is in w iff IBI = Iq for some DC 10
W. The left-to-right direction is trivial. To establish the
direction we show tusing Lindenbaum's Lemma) that IBI = Iq Impbes
that B C is super-derivable from J. Given B H C is super-derivable
from J, so is DB H DC; and so given DC is in w, DB is in W.
From these lemmas it is straightforward to establish model super-
completeness. There is also no difficulty in extending the proof to mul-
timodallogics. QED.
The theorem provides a simple corollary, on
i.e. super-completeness for frames. Say that a model .. M. IS If
it is (i) differentiated (as defined in Fine 1975: 23) and (ll) tight, I.e. wRV
implies that, for some formula B, DB is true at wand V = liB 11M A
TRANSFER THEOREMS FOR MUL TIMODAL LOGICS 207
classical logic is natural if any natural model for the logic is based on a
frame for the logic. Then:
COROLLARY 5. Any natural logic is strongly super-complete.
Proof By the proof of the theorem above, since the canonical model
constructed there is natural. QED.
This result comprehends most of the standard classical logics, since
most of them can be shown to be natural.
Let us now establish the transfer of super-completeness.
THEOREM to. For every formula space e: Super-completeness w.r.t. e
transfers.
For the purpose of proving this result, let us suppose the logics Ll and
L2 (of the respective types 1 and 2) are super-complete w.r.t. e and that
the pair (J,l) of subsets of e is consistent in L = Ll EB L2. We wish to
show that the pair (J,l) is satisfiable in some frame for L.
For any formula set r, the 'IT-theory of r, T,,(J'), is the set of all truth-
functional compounds of formulae in S,,(l') v SC(J) which are super-
derivable in L from J. (Thus Til') is relative to the given set J.) A
labelled n-model is then a triple (M,WM,E(M where (i) M is a 'IT-model
for L", (ii) w M E WM, (iii) E(M) is a set of constituents closed under
subconstituents, and (iv) T ,,(E(M is universally true in M. We also
designate the theory T ,,(E(M of clause (iv) as T(M).
An initia/labelled model (M*,w*, E(M* is picked as follows. We
select a value for 'IT; we let E(M*) be the set of subconstituents of r; and
we choose a 'IT-model M* and the world w* of M* so that the pair
(T(M*),l) is true at w* in M*. We know that an initial model exists.
For (J,l) is L-consistent; so (T(M*),I/ is L-consistent and thus L,,-con-
sistent; and so, by the super-completeness of L" w.r.t. e, (T(M*),l) is
true at a world w* of a 'IT-model M*.
Given a labelled 'IT-model M and a labelled model N, we say that N
hangs off M (at WN) if: (i) N is of type 'IT*; (ii) WM fl WN = {WN}, and
either M = M* or WN'* WM; (iii) E(N) = SiE(M; (iv) the same con-
stituents of E(N) and of SC(J) are true at WN in both Nand M. Note
that in clause (iv), agreement on the subconstituents of J, as well as on
the 'IT-subconstituents in E(N), is required. This is because we want J to
be universally true in the resulting model.
Given the initial model and the relation of hanging-off, a sprouting
and an m-sprouting is defined in the same way as before. To show that
a maximal m-sprouting exists, the basic problem is to establish the
208 KIT FINE AND GERHARD SCHURZ
appropriate form of consistency at the world from which one attempts
to hang off a model. Suppose M is a 'IT-model and w a world in M and
IT is the diagram of SnC.E(M v SqA) at win M. Then we need to show
that (A,Il) is L-consistent; for T ".(IT) is super-derivable in L from A,
and so it will then be that (T ".(IT),II) is L-consistent and hence L"o-con-
sistent. But if (A,II) were L-inconsistent, then from A would be super-
derivable the negation of a conjunction C of formulae in IT. So -, C
would belong to T(M) and consequently be true at w in M, which is
impossible.
Given a maximal m-sprouting S+, the corresponding 12-model M+ =
(W+,Rl+,R2+<P+) is defined as follows: (i) W+ = U{WM 1M E S+}; (ii)
wR,,+Viff (\ WAf), where M = M'IT(w) (the 'IT-model in S+ con-
taining w); and (iii) <p(P) = U{<pM(P) I ME S+}. We call M+ the big
model and the members of S+ the small models. We refer, likewise, to
the big and small frames.
The proof that the big frame is a frame for L depends upon an ana-
logue of the generated submodel lemma for normal logics. Given a
neighbourhood frame F = (W,R), say that the subset Vof W is closed
if, for each v in V and each subset U of W, vRU iff vR( U (\ V). Let the
restriction !I a neighbWIrhood frame F and of a neighbourhood model
M to V, Fl V, and MI V, respectively, be defined in the obvious way.
Then the fundamental result on V-closure is:
LEMMA 12. Suppose that the frame of the model M = (W,B,<p) is V-
closed for some V C W. Then for any A E ;;e and v E V: (MT V, v) 1= A
iff (M, v) 1= A.
Proof By induction fn A. The critical case is when A = DB. But v
makes DB true in M V iff vCR i V)(I\BIIMi v), i.e. by indo hypo and the
def. of restriction, iff vR(IIBlIM (\ V). Likewise, v makes A true in Miff
vR IIBIIM. But by V-closure, vR(IIBI\M (\ V) iff vRIIBI\M. QED.
Say that the frame F = (W,R) is the disjoint union of the frames
(Fi \ i E 1) if the world-domains Wi are pairwise disjoint and if, for any
member wand subset Vof W = U{ Wi liE l}, wRY iff wRiV n Wj),
where Fj is the frame to which w belongs. Then it follows from the
lemma that any disjoint union F = (W,R) of the frames (Fi liE l) for
L is also a frame for L. For suppose that some theorem A of L is false
at the world w of a model M = (W,R,<p) based on F. Let (Mi liE l) be
the corresponding models based on the frames (F; liE I), and let Mj be
the model to which the world w. belongs. thx frame of M is
closed; and so by the lemma, A IS false at w 10 Ml Wj = Mj.
Now the bi!! frame F+ is the join of a big I-frame G and a big 2-frame
....
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 209
H. By construction, G is the disjoint union of the small I-frames; and
likewise H is the disjoint union of the small 2-frames. But since each
small 'IT-frame is a frame for L'Il" it follows from the observation above
that the big 'IT-frame is a frame for L,,; and since the logic L = Ll EB L2
is stratified, it follows that the big frame is a frame for L.
Let us now consider the question of agreement. Suppose that M is a
small model. Let A be a truth-functional compound of the constituents
in .E(M) v sq.1). Then we may show that A is true at the world w in
M iff it is true at the world w in M+. The critical case is when A = o"B.
There are two subcases, according to whether M is of type 'IT or 'IT*. (i)
M is of type 'IT. o.,.B is true at w in Miff wRMIIBIIM, i.e. by indo hypo and
the construction, iff wR+(I\BII.w n WAf). 0wB is true at w in M+ iff
wR+IIBII.w. But since the frame of M+ is WM-closed wR+(IIBII.w (\ WAf)
iff wR+IIBIIM+. (ii) Suppose M is of type 'IT*. Then some model Neither
hangs off Mat w or has M hang off it at W. In either case, 0wB will be
a common member of .E(M) v SqA) and .E(N) v sq.1); and so, since
M and N will agree at w in the truth-values of those common members,
the present subcase will reduce to the previous one.
From agreement, it readily follows that the pair of formula sets (A,1/
is true at w* in M+; and so the proof is complete.
Let us define 'super' analogues of various standard conditions on log-
ics. L has if each L-consistent pair of formulae is satisfiable
in a finite model for the logic; L is super-decidable if the relation of
super-derivability between formulae is decidable; L has super-interpola-
tion if whenever C is super-derivable from A in L there is a formula B
in their common language such that C is super-derivable from Band B
is super-derivable from A; L has the super joint consistency property if
(A v .1', r v T') is L-inconsistent whenever, for some formula A in the
common language of A v rand .1' v T', A is L-derivable from (.1,1/
and -,A is L-derivable from (A',T'); L has super Halldbt-completeness
if (.1 v A', ruT') is L-consistent whenever (A ,1/ and (.1' ,r') are L-
consistent and their common non-logical language is empty.
By using the above methods of proof we are able to establish the
transfer of the following conditions: weak super-completeness, strong
super-completeness, and (under the assumption of weak completeness)
super-interpolation, super joint consistency, and super Hallden-
completeness. The proof of conditional transfer for super-decidability
breaks down; for the complexity of A and thus the complexity of
the theorem set TiT) C B(Sw(T) u SqA does not reduce in the
proof. However, we are able to show that if L\ and L2 are weakly com-
plete and super-decidable, then L = Ll EB L2 is decidable; because
210 KIT FINE AND GERHARD SCHURZ
derivability in L for r amounts to super-derivability from (0,Ij; in this
case T'IT(I) reduces in complexity. For the same reason, the proof of
transfer for super-f.m.p. breaks down, though we are able to show that
if L\ and L2 have super-f.m.p. then L has f.m.p. Similar 'super' trans-
fer results can be proved for normal logics by similar methods.
It would be of interest, in the light of these results, to know whether
a classical (or normal) logic can have a given property without having
the corresponding super-property and, should this be possible in the
classical case, whether one can have transfer for the super-property
without having transfer of the given property.
7. Appendix. Applications to Multimodal Logics with Interactive
Axioms
7.1. Dynamic Logics
First some terminology, following the explanations in the introduction.
The language !e(DL) of dynamic logic is built up from a set P of atomic
program terms and a set n:n of proposition variables by the formation
rules: PC PROG; if a, fJ E P and A E !(DL), then a;fJ, aU fJ, ?A and
0.* E PROG (where PROG is the set of program terms; a, fJ, ... E
PROG); n:n C !e(DL), and if A, BE !e(DL), then .. A, A v B, and [alA
E !e(DL), for all a. E PROG. !e(QDL), the language of quasi-dynamic
logic, is the restriction of !e(DL) to formulae without non-atomic pro=
gram terms; and !e(FDL), the language of loop-free dynamic logic, is
the restriction of !e(DL) to formulae without the star operator. Any
normal multimodal logic in !e(QDL) is called a quasi-dynamic logic.
QDL, FDL, and DL denote the class of all quasi-dynamic, loop-free
dynamic, and dynamic logics, respectively. For sake of simplicity, we
restrict our attention to monostratified quasi-dynamic logics L (the
more general case is analogous); i.e. L = 61{La I a E P}, where each La
is a normal monomodal logic in the single atomic program a.
Proposition 1* tells us that the class F(L) of all frames (W, {Ra I a E P})
for L is identical with the join 61{F(La) I a E P} of the classes F(La) of
all La-frames, for all a E P.
Given any monostratified L E QDL, let L + denote its extension to a
loop-free dynamic logic, which is uniquely determined by adding all
instances of the following interactive axiom schemes:
TRANSFER THEOREMS FOR MULTIMODAL LOGICS 211
(C) [a;fJ]A H [a][fJ]A;
(U) [a U fJ]A H [alA 1\ [fJ]A;
(T) [?A]B H (A 4 B)
and under MP and the rules Na for all programs a in PROG
no operator. Similarly, given any frame F E F(L), let F+
denote expanSIOn to a frame for L +, which is uniquely determined
by settIng R?;p := Ra'Rp (the relational product) and Raup := Ra U Rp.
Moreover, gIven any model M E M(L), let M+ denote its expansion to
a model for L + by setting in addition R?A : = {(u,u) I (M,u) 1= A} for all
E !(FDL). (Since the relations R?A depend on the valuation func-
tion, we include them in the models and not in the frames.) We put
F(L +) := {F+ I FE F(L)} M(L {M+ 1M E M(L)}. Finally,
for A E let At be ItS famIlIar translation in the language
!(qDL), determIned by the following rules: (P)t = p; ( .. A)I = .,(At);
(A V B)t, = (At V Bt); ([a]A)t = [a](At) whenever a is atomic; ([a;fJ]A)t =
[a][fJ](A) ; ([a U fJ]A)t = [a](A)t 1\ [fJ](A)t; and ([?A]B)' = (At 4 Bt).
The above translation enables us to transfer weak and strong com-
pleteness, f.m.p., canonicity, and Hallden-completeness from L =
61{La.1 a P} to its loop-free extension L +, and hence enables us, in
combInation with our other results, to transfer these properties (the last
three under the proviso of weak completeness) from the component log
ics La to L +. I -
Th.e extension. to full logic raises difficulties. The star oper-
IS charact:r!zed semantIcally by the condition that R
a
* is the reflex-
Ive and tranSItIVe closure of Ra; and syntactically by the axioms
[a*]A.4 A 1\ [a][a*]A and [a*](A 4 [alA) 4 (A 4 [a*]A). It is undefin-
able In FDL. Thus the above method will not work' and nor is it clear
what to put in its place. '
7.2. Multimodal Logics with D2kDj (X)
again the simplest case of a bimodal logic with one inter-
actIve aXIOm (the more general case is analogous). Recall the definition
of L\X and of D2kof (X) given in Section 1.
THEOREM. Let L\ and L\X be I-logics over !e and L2 a 2-logic over !e,
and e be space. Then: If L\ and L\X are complete W.r.t.
and L2IS complete w.r.t. DC2(e), then L:= (LI 61 L2)D2
ko
'f(X)
IS complete w.r.t. e, for any k ;::: O.
Sketch: Li.ke that of Theorem 1, but with the following changes.
GIven the L-conslstent set r e, we start with a 2-model M* based on
212 KIT FINE AND GERHARD SCHURZ
a 2-frame verifying r at w*. The definition (1.1) of a 2-theory has to
be modified to: T2(.d) := {D2nB ELI B E B(S2(.d and d2(D2nB):;;'
d2(.1)}. It now follows that D2"D
1
mXwill be in T2(.1) when Dim X E S2(.1)
and k:;;. d2(.1). In Lemma 2 we have to prove now that, for every
labelled 2-model M and u E WM, two claims hold: (a) that DM<.u) is
always L-consistent and hence Ll-consistent, and (b) that whenever
there exists an R2-path from WM to u with length ~ k, then DM(U) is
L1X-consistent. While (a)'s proof is as before, (b)'s proof rests on the
following:
LEMMA. If A E L1X, then D2"Of(A) C L.
Proof The proof is by a straightforward induction on the length of the
proof of A in LIX.
Now assume, contrary to claim (b), that there exists an R2-path from
WM to u with length ~ k, and DM<.u) is LIX-inconsistent. Then there
exists ..,A..1 E LIX (with finite .1 ~ DM<.u, whence D2
k
..,M E L (by
the lemma), and .., ALl E B(S2(.E(M) (arguing as in Lemma 2).
Therefore D2
k
..,AA E T2(.l'(M, whence ..,1vJ must be true at u in M.
A contradiction.
We require from the selection function m that (i) each 2-model M2 in
our sprouting is generated, and (ii) a I-model Ml for DM(U) u TI(l'(M
based on a frame for LIX hangs off each world u in M2 which is reach-
able from some world by an R2-path with length ~ k. Such a model
always exists by claim (b). It is then readily verified that the frame of
the resulting maximal 12-model is a frame for L. QED.
From this theorem, it follows as before that the properties of weak
and strong completeness, the finite model property, interpolation, and
Hallden-completeness transfer from LJ, LIX, and L2 to L. However, the
proof of decidability does not go through because in the procedure
explained in the proof of Theorem 7, we now have to test whether 02nB
is in L, for all BE Bf(S2(A such that d2(D2n) ::; d2(A); but adl(02nB)
need not be smaller than ad2(A).
REFERENCES
AQVlST, L. (1965), A New Approach to the Logical Theory of Interrogatives (Uppsala:
Almqvist and Wiksell).
BOWEN, K. A. (1980), 'Interpolation in Loop-free Logic', Studia Logica, 39:
297-310.
FINE, K. (1975), 'Some Connections between Elementary and Modal Logic', in
,
"
TRANSFER THEOREMS FOR MULTIMODAL LOGICS
213
S. Kanger (ed.), Proceedings of the Third Scandinavian Logic Symposium
(Amsterdam: North-Holland).
GARGOV, G., and PASSY, S. (1990), 'A Note on Boolean Modal Logic', in P. Petkov
(ed.), Mathematical Logic (New York: Plenum Press).
GOLDBLATT, R. (1987), Logics of Time and Computation, CSLI Lecture Notes, 7
(Stanford, Calif.).
HALLDEN, S. (1951), 'On the Semantic Non-completeness of Certain Lewis Calculi',
The Journal of Symbolic Logic, 1512: 127-9.
HAREL, D. (1984), 'Dynamic Logic', in D. Gabbay, and F. Guenthner (eds.),
Handbook of Philosophical Logic, ii (Dordrecht: Reidel).
MAKINSON, D. C. (1971), 'Some Embedding Theorems for Modal Logic', Notre
Dame Journal of Formal Logic, 12: 252-4.
SCHURZ, G. (1991), 'How Far can Hume's Is-Ought Thesis be Generalized:
Investigations in Alethic--Deontic Modal Predicate Logic', Journal of
Philosophical Logic, 20: 37-95.
--(1996), The Is-Ought Problem: An Investigation in Philosophical Logic (forth-
coming).
SEGllRBERG, K. (1971), An Essay in Classical Modal Logic (Uppsala: Filosofiska
Studier).
- (1980), 'Applying Modal Logic', Studia Logica, 39: 275-95.
THOMASON, R. H. (1980), 'Independent Modal Logics', Studia Logica, 39: 143-4.
THOMASON, S. K. (1972), 'Semantic Analysis of Tense Logic', Journal oj Symbolic
Logic, 37: 150-8.