Claudio Pizzi Necessity and Relative

Contingency

Abstract. The paper introduces a contingential language extended with a propositional

constant τ axiomatized in a system named KΔτ , which receives a semantical analysis via
relational models. A definition of the necessity operator in terms of Δ and τ allows proving
(i) that KΔτ is equivalent to a modal system named Kτ (ii) that both KΔτ and Kτ are
tableau-decidable and complete with respect to the defined relational semantics (iii) that
the modal τ -free fragment of KΔτ is exactly the deontic system KD. In §4 it is proved that
the modal τ -free fragment of a system KΔτ w weaker than KΔτ is exactly the minimal
normal system K.

Keywords: Contingency, relative necessity, propositional constants, semantic tableaux.

§0. The problem of defining necessity in terms of contingency has been

solved in systems of contingency logic which are at least as strong as KTΔ —
the contingency version of KT — but has received0a substantially negative
answer with respect to KDΔ — the contingency version of KD — and to
normal systems weaker than KDΔ1 .
A different approach to the question consists in taking into consideration
contingential systems whose language is more extended than the standard
one for containing one and only one propositional constant.
The language LΔτ which will be used in what follows has the following
primitive symbols:
0) parentheses ( , )
1) ⊥, ⊃, Δ
2) a countable set of propositional variables p, q, r . . .
3) the propositional constant τ .
The rules of formation and of elimination of parentheses are submitted
to usual conventions. The auxiliary functors ¬, ∧, ∨, ≡, T are introduced

1
As stated in Humberstone [2002], “where S is a -based modal logic, SΔ is to be
its Δ + Boolean connectives fragment, thinking of Δ as having been introduced into S
by the definition of ΔA as A ∨ ¬A”(p.123). For the above quoted negative result see
especially Cresswell [1988]. Cresswell proved that, with the exception of some anomalous
normal systems (as for instance K +Verum) systems weaker than KT cannot in general
grant the definability of  in terms of Δ.

Presented by Daniele Mundici; Received December 20, 2005

Studia Logica (2007) 85: 395–410

DOI: 10.1007/s11225-007-9044-y © Springer 2007
404
396 C. Pizzi

in standard way. A further auxiliary symbol is introduced by the definition

(Def ∇) ∇A =Df ¬ΔA.

∇A may be read “contingently A” and ΔA “absolutely A”.

The intuitive meaning that can be associated to τ is context-dependent:
τ may be read for instance as “the moral laws are respected” or “general rela-
tivity is accepted by the scientific community”, but every consistent stock of
information which is not analytically true will do as a suitable interpretation
of τ .
The system which here will be named KΔτ is axiomatized in the following
way:

KΔ0. All the tautologies of the classical propositional calculus PC

KΔ1. Δp ≡ Δ¬p
KΔ2. (Δp ∧Δq) ⊃ Δ(p ∧ q)
KΔ3. (Δp ∧∇(p ∨ r )) ⊃ Δ(p ∨ q)
KΔ4. ∇τ
Rules: Modus Ponens (MP), Uniform Substitution (US)
(ΔNec) A →  ΔA
(ΔEq)  A ≡ B →  ΔA ≡ ΔB
Let us now add to KΔτ the following two definitions:
[Def O] OA = Df Δ(τ ⊃ A)
[Def ] A = Df ΔA ∧ OA.

Thanks to the standard definition of ♦ and to the definition

[Def P] PA =Df ¬O¬A
we obtain the useful equivalences
(P) PA ≡ ∇(τ ∧ A)
and
(♦) ♦A ≡ ∇A ∨ ∇(τ ∧ A)
We will state without proof that Replacement of Proved Material Equiv-
alents (Eq) is a valid KΔτ -rule.
It may be proved that KΔτ is sound with respect to the class of KΔτ -
models, where a KΔτ -model is a 4-ple W, Wτ , R,V which is so defined:

(i) W = ∅
(ii) Wτ ⊆ W
Necessity and Relative Contingency 405
397

(iii) R is a τ -forked relation over W, or in other words

∀x∃y1 ∃y2 ((xRy1 ∧ y1 ∈ W τ ) ∧ (xRy2 ∧ y2 ∈

/ W τ ))

(iv) V is defined as usual for truth-functional wffs and for the other
wffs by the following two clauses:
(iv1) V(τ, x) = 1 iff x ∈ Wτ
(iv2) V(ΔA, x) = 1 iff for every y such that xRy and every y  s.t. xRy  ,
V(A,y) = V(A,y  ).
Obviously in place of (iv2) we could have
(iv2’) V(∇A,x) = 1 iff for some y such that xRy and some y  s.t. xRy  ,
V(A,y) = V(A,y  ).

If we call worlds the elements of any set W belonging to a KΔτ -model,

KΔτ -validity is the truth in all worlds of all KΔτ -models.
It is useful to remark that the result of removing KΔ4 from the set of
axioms of KΔτ is the minimal contingential system KΔ. Such a system has
been proved by S.Kuhn in [1995] to be complete with respect to the class of
all contingential models, i.e. of all 3-ples W, R, V where W is as in (i), V
is as in (iv) and (iv2), while R is an arbitrary two-place relation over W2 .
It may proved then:

Theorem 1. For every A, if A is a thesis of KΔτ , A is KΔτ -valid.

Proof. Standard induction on the length of proofs. We first show that

all axioms are KΔτ -valid. As KΔ0-KΔ3 is concerned, the result follows
from the mentioned completeness theorem for KΔ, i.e. from the fact that
KΔ0-KΔ3 hold in all contingential models, so a fortiori in all KΔτ -models.
As to KΔ4, suppose that V(∇τ, x) = 0 in some world x of some KΔτ -
model. Then V(Δτ, x) = 1, and by (iv2) this means that for every y such
that xRy and every y  s.t. xRy  , V(τ, y) = V(τ, y  ), which is incompatible
with condition (iii) stating that R is τ -forked.
It is trivial to show that rules MP, US, ΔNec, ΔEq are KΔτ -validity
preserving.

2
After removing axiom KΔ4 we lack an axiom governing τ , so τ plays the same role of
a variable and does not add new theorem schemata to system KΔ. A KΔ-model can then
be redefined as a 4-ple W, Wτ , R, V where W, R, V are as in traditional KΔ-models,
Wτ ⊆ W but τ is not bound to the interpretation described by the condition (iv1):
V(τ, x) = 1 iff x ∈ Wτ . Every KΔτ -model may be then be considered a 4-ple which is a
special case of a KΔ-model.
406
398 C. Pizzi

§1. As far as completeness with respect to the defined semantics is con-

cerned, a proof via canonical models is technically unproblematic3 . However,
a constructive completeness proof may be may be derived from a practical
decision procedure which may be defined for KΔτ . A step towards estab-
lishing the decidability of KΔτ consists in translating it into an equivalent
modal system Kτ .
Let us call now call Lτ a language which is like the language LΔτ with
the difference that  takes the place of Δ. Kτ will be a system, based on
Lτ , whose axioms are (i) an axiom expressing the normality of :
K. (p ⊃ q) ⊃ (p ⊃ q)
and (ii) a second axiom expressing the so-called Bilateral Possibility of τ :
(Bτ ) ♦τ ∧ ♦¬τ
Rules: Modus Ponens, US, Nec: A →  A
We will state without proof that Replacement of Proved Material Equiv-
alents (Eq), is a valid Kτ -rule.
Kτ turns out to be decidable by a suitable extension of the semantic
tableau method for K. A K-tableau T is a set of diagrams dx, dy. . . con-
taining modal wffs related by arrows representing the accessibility relation
between diagrams and built by applying a set of rules which is essentially
the same as formulated by Hughes and Cresswell in [1968]. Familiarity with
this text is assumed in what follows. Any tableau for a wff A (the input of
the tableau) will be said to be closed when the Reductio supposition that
A has value 0 in the first diagram dx leads, after evaluating all the possible
alternatives, to some inconsistent assignment to some subformula of A in dx
or other diagrams. It will be said to open when at least one of the alter-
natives does not lead to some inconsistent assignment to some subformula
of A.
We now define a Kτ -tableau as a K-tableau having a wff of Lτ as
input and built by applying, beyond the rules for K-tableaux, the following
two additional rules, the second of which is a Stop Rule introduced to prevent
the existence of infinite tableaux :
(R1) Given any diagram dx containing wffs with consistent assignments,
draw two arrows from dx toward two new accessible diagrams dy’, dy”;
insert in dy’ τ with assigment 1 and in dy’ τ with assigment 0. Reproduce
then in both dy’ and dy” all the arguments of the universal modal operators
occurring in dx (i.e.  with value 1 or ♦ with value 0).

3
For this approach to the semantics of KΔτ see Pizzi [2006].
Necessity and Relative Contingency 407
399

(SR) Let D be a chain of diagrams dxRdy. Rdz such that dz is identical to

some dk occurring earlier in D (two diagrams being identical if and only if
they contain the same set of wffs with the same assignment). Identify then
dz and dk, and send to dk all the arrows which go to dz.
If dzRdk and dz contains one or more modal wffs, the modal degree of
dk is always lower than the modal degree of dz. This implies that if two
diagrams dz and dk are identical and belong to the same chain D, they do
not contain modal formulas. The Stop Rule is then applied only to diagrams
containing non-modal wffs.
The following graph gives an instance of a tableau shortened by an ap-
plication of SR.

p
0
@
@
@
@
@
R
@
-
'  $
τ τ
0 1
& %
@
I
@ @ @
I
@ @
@ @ @ @
@ @ @ @
@ @ @ @
@ @@
R @ @@
R
'-  $
τ τ
1 0
& %

The proof of the following three theorems is easy and will be omitted:

Theorem 2. Every tableau T built by application of the rules for K-tableaux,

R1 and SR (so every Kτ -tableau) is a KΔτ -model.
408
400 C. Pizzi

Theorem 3. Let T’ be a tableau for some input formula A which is like

a Kτ tableau T for A with the only difference that T’ is built with no
application of SR. Then T is closed only if T’ is closed and T is open only
if T’ is open.
Theorem 4. The tableau method for Kτ provides a decision procedure for
KΔτ -validity.
In order to prove the completeness of Kτ with respect to the class of
KΔτ -models we rely on the notion of a characteristic formula of a diagram
as introduced by Hughes and Cresswell [1968]. Let us suppose that di-1 and
di are two diagrams such that di-1 R di. Then, supposing that ♦ has been
everywhere eliminated by definition, the characteristic formula C(di) of a
diagram di is a disjunction
C(di) = E1 ∨ ¬U1 ∨ ¬U2 ∨. . . ∨ ¬Un
where E1 is the wff which has been introduced in di by the occurrence in
di-1 of E1 with value 0, while U1 . . . Un that are wffs that receive value 1
in di due to the fact that U1 . . . . Un receive value 1 in di-1.
Given C(di) as before defined, obviously in Kτ a characteristic wff
C(di) is a thesis of KΔτ iff (♦τ ∧ ♦¬τ ) ⊃ A is such.
Completeness may now be proved by a suitable adaptation of the com-
pleteness theorem for K. If a diagram di of a closed Kτ -tableau for the
input formula A is inconsistent, it may be proved that (♦τ ∧ ♦¬τ ) ⊃ C(di)
is a Kτ -Thesis, so C(di) is also a Kτ -thesis. Passing to di-1 we prove
that the theorem C(di) implies (♦τ ∧ ♦¬τ ) ⊃ C(di-1 ), so that also C(di-1 )
turns out to be a thesis. The proof has and end when one proves that the
characteristic wff of the first diagram, namely the input formula A, follows
from ♦τ ∧ ♦¬τ and Modus Ponens. The argument then proves
Theorem 5. For every A, if A is KΔτ -valid, A is a thesis of Kτ .

§2. Let us now extend Kτ by the standard definition of non-contingency

[Def Δ] ΔA =Df A ∨ ¬A
and KΔτ with the Definition of Necessity formulated in §0, which amounts to
[Def ] A =Df ΔA ∧ Δ(τ ⊃ A)
Now we are able to prove that the two systems KΔτ + Def and Kτ
+ DefΔ are equivalent systems.
Necessity and Relative Contingency 409
401

Given any wff A of LΔτ we say that a -variant of A is a wff A* which

is like A with the difference that every occurrence of a subformula ΔC in A
is replaced by C ∨ ¬C and every occurrence of ∇C in A is replaced by
♦C ∧ ♦¬C.
Given any wff B of Lτ we say that a Δ-variant of B is the wff B**
which is like B with the difference that any occurrence of a subformula C
in B is replaced by ΔC ∧ Δ(τ ⊃ C). The notion of a rule R*(R**) being a
-variant (Δ-variant ) of a rule R should be obvious.
It is clear that A is a thesis of KΔτ + Def if and only if A* is such,
and B is a thesis of Kτ + DefΔ iff B** is such.
We will now prove in the first place

Theorem 6. For every A, if A is a thesis of KΔτ + Def, A is a thesis of

Kτ + Def Δ.

Proof. By induction on the length of the proofs.

1) It is not difficult to see that the -variants of all axioms of KΔτ are Kτ
+ Def Δ-theorems. This is proved by using the outlined tableau method.
2) The rule  A →  ΔA has a -variant which is A →  A ∨ ¬A,
which is obviously a valid Kτ -rule. Knowing that Eq is a valid rule of
Kτ we have also A ≡ B →  (A ∨ ¬A ) ≡ (B ∨ ¬B), which is the
-variant of rule ΔEq.
3) We prove that -variant of the equivalence yielded by Def is a theorem
of Kτ + Def Δ. Since we have at our disposal a decision procedure, a way
to do it is to test via Kτ -tableaux the validity of the -variant of p ≡
(Δp ∧ Δ(τ ⊃ p)), i.e.
( ≡) p ≡ ((p ∨ ¬p) ∧ ((τ ⊃p) ∨ (τ ∧ ¬p))).
However, it is useful to observe that from left to right the implication is a
trivial K-theorem, and that the converse implication may proved syntacti-
cally as follows, after recalling that ♦τ and ♦¬τ are obvious consequences
of Axiom (Bτ ):

(1) ♦τ ⊃ ( (τ ⊃ p) ⊃ ♦p) K
(2) (τ ⊃ p) ⊃ ♦p (1), ♦τ
(3) (♦p ∧ (p ∨ ¬p)) ⊃ p K
(4) (τ ⊃ p) ⊃ ((p ∨ ¬p) ⊃ p) (2), (3)
(5) (p ∨ ¬p) ⊃ ((τ ⊃ p) ⊃ p) (4), PC
(6) (♦¬τ ∧ ¬p) ⊃ ((τ ∧ ¬p) ⊃ p) K
410
402 C. Pizzi

(7) ¬p ⊃ ( (τ ∧ ¬p) ⊃ p) (6), ♦¬ τ

(8) (p ⊃ ((τ ∧ ¬p) ⊃ p) PC
(9) (p ∨ ¬p) ⊃ ((τ ∧ ¬p) ⊃ p) (7), (8), PC
(10) ((p ∨ ¬p) ∧ ((τ ⊃ p) ∨ (τ ∧ ¬p))) ⊃ p (5), (9), PC

Theorem 7. For every A, if A is a thesis of Kτ +Def Δ then A is a thesis

of KΔτ + Def .

Proof. By induction on the length of the proofs.

1) First we have to prove in KΔτ the Δ-variant of the axiom K expressing
the normality of the box. In other words we have to prove in KΔτ
(K**) (Δ(p ⊃ q) ∧ Δp ∧ Δ(τ ⊃ (p ⊃ q)) ∧ Δ(τ ⊃ p)) ⊃ (Δq ∧ Δ(τ ⊃ q))
First we observe that the antecedent of K**, i.e.
(a) (Δ(p ⊃ q) ∧ Δp ∧ Δ(τ ⊃ (p ⊃ q)) ∧ Δ(τ ⊃ p))
implies via axiom KΔ2 the wff Δ((p ⊃ q) ∧ p) ∧ Δ((τ ⊃ (p ⊃ q)) ∧ (τ ⊃ p))
which is PC-equivalent to
(b) Δ(p ∧ q) ∧ Δ(τ ⊃ (p ∧ q))
Then we may prove that the following two wffs are KΔ-theorems:
(c) ∇r ⊃ (Δ(p ∧ q) ∧ Δ(r ⊃ (p ∧ q)) ⊃ Δq)
(d) ∇r ⊃ (Δ(p ∧ q) ∧ Δ(r ⊃ (p ∧ q)) ⊃ Δ(r ⊃ q))
A quick proof is semantic, i.e. consists in proving that there is no KΔ-
countermodel to (c) and (d).
Let us consider (c). Suppose by Reductio that there is a world x of a
KΔ-model in which (c) is false. Then ∇r is true in x along with (Δ(p ∧
q), Δ(r ⊃ (p ∧ q)) and ¬Δq, i.e. ∇q. So there are four worlds y1 , y2 , y3 , y4 all
accessible from x with the following properties: V(r, y1 ) = 1 and V(r, y2 ) =
0, V(q, y3 ) = 1 and V(q, y4 ) = 0. V(r ⊃ (p ∧ q)) should be 1 in every such
world or 0 in every such world. Since r is false in y2 , V(r ⊃ (p ∧ q)) is 1
in y2 and, due to the truth of Δ(r ⊃ (p ∧ q)) in x,V(r ⊃ (p ∧ q)) is 1 in all
y1 −y4 . Since V(r, y1 ) = 1, V(p∧q, y1 ) = 1 and, given that V(Δ(p∧q, x) = 1,
V(p ∧ q) should be 1 in all y1 − y4 . But this is impossible given that q is
false in y4 .
Let us consider (d). The argument runs as before, but in y3 we have r ⊃ q
with value 1 and in y4 r ∧ ¬q with value 1. As before, V(r ⊃ (p ∧ q)) should
be 1 in every world y1 -y4 or 0 in every such world. Since r is false in y2 , r ⊃
(p ∧ q) is true in y2 , so true in y1 -y4 because of V(Δ(r ⊃ (p ∧ q)), x) = 1.
But we have V((r ∧ ¬q), y4 ) = 1, which leads to an inconsistency.
(c) and (d) are then KΔ-valid, so by the completeness result they are KΔ-
theorems. So a KΔ-theorem is also, by PC,
Necessity and Relative Contingency 411
403

(e) ∇r ⊃ ((Δ(p ∧ q) ∧ Δ(r ⊃ (p ∧ q)) ⊃ (Δ(r ⊃ q) ∧ Δq))

In language LΔτ , an instance of (e) is (via the substitution of τ to r)
(f) ∇τ ⊃ ((Δ(p ∧ q) ∧ Δ(τ ⊃ (p ∧ q)) ⊃ (Δ(τ ⊃ q) ∧ Δq))
and by axiom KΔ4, i.e. ∇τ , and Modus Ponens, also
(g) (Δ(p ∧ q) ∧ Δ(τ ⊃ (p ∧ q)) ⊃ (Δ(τ ⊃ q) ∧ Δq))
But the antecedent of (g) is exactly (b) and, as we proved, (b) is implied
by (a): so by transitivity the wff K** is a theorem of KΔτ .
2) As regards Necessitation, the rule A →  A has as Δ-variant the rule
 A →  ΔA ∧ Δ(τ ⊃ A), whose validity depends, given A ≡ T, on the
PC-thesis  τ ⊃ T via  Δ(τ ⊃ T) and Eq.
3) It remains to prove the Δ-variant of the equivalence Δp ≡ (p ∨ ¬p),
i.e. (Δ ≡) Δp ≡ ((Δp ∧ Op) ∨ (Δp ∧ O¬p)).
The implication is obvious from right to left. In the other direction it is
useful to use the following wff:
(ΔPO) (Δp ∧ Pp) ⊃ (Δp ∧ Op)
which can be shown to be a KΔτ -theorem. Suppose in fact by a contradiction
Δp ∧ Pp ∧ ¬Op, where Pp is ∇(τ ∧ p) and ¬Op is ∇(τ ∧ ¬p). Now the
truth in any arbitrary world x of ∇(τ ∧ p) implies that in some y s.t. xRy
p is true, the truth of ∇(τ ∧ ¬p) implies that in some y  such that xRy  p is
false, but the truth of Δp in x implies that the value of p in y and in y  is
the same: contradiction. Then for any p Δp ∧ Pp implies Op, so also Δp ∧
Op.
From theorem (ΔPO) we have by US
(h) (Δp ∧ P¬p) ⊃ (Δp ∧ O¬p) i.e.
(i) (Δp ∧ ¬Op) ⊃ (Δp ∧ O¬p)
Now a PC-theorem is Δp ⊃ ((Δp ∧ Op) ∨ (Δp ∧ Op))
so by (b) we reach the thesis
Δp ⊃ ((Δp ∧ Op) ∨ (Δp ∧ O¬p))

As an obvious consequence of the equivalence of the two systems KΔτ

and Kτ , we obtain the completeness and decidability of KΔτ .

Theorem 8. For every A, if A is KΔτ -valid, A is a thesis of KΔτ .

Proof. From theorems 5, 6, 7.

Theorem 9. KΔτ + Def  is tableau-decidable.

Proof. If A is a wff of LΔτ , in order to check the KΔτ -validity of A we

test the equivalent wff A* with the Kτ -tableau method.
412
404 C. Pizzi

Remark. The tableau procedure makes it easy to check that the following
are theses of both KΔτ and Kτ :
1) ∇τ ≡ ♦τ
2) τ ≡ Δτ
3) (τ ⊃ A) ≡ Δ(τ ⊃ A)
4) ♦(τ ∧ A) ≡ ∇(τ ∧ A)
The last equivalence shows that the confusion between possibility and
contingency, which is often pointed as an incongruity in Aristotle’s Organon,
is a logical truth for every A which is coinjoined with some contingent stock
of given information.

§3. We are now able to prove that the fragment of KΔτ + Def containing
no occurrences of t is exactly KD, i.e. the deontic system K + ♦T. The
proof runs in two parts:

Theorem 10. For every τ -free A, if A is a thesis of KD, A is a thesis of

Kτ + DefΔ.

Proof. We already know that all the theorems of K are theorems of Kτ
+ Def Δ.
♦τ ⊃ ♦T is a substitution instance of the K-theorem ♦p ⊃ ♦T. So by axiom
(Bτ ) ♦T is a theorem of Kτ + Def Δ.

Theorem 11. For every τ -free A, if A is not a thesis of KD, A is not a

thesis of Kτ + Def Δ.

Proof. Let A be a wff not containing τ . As is well known, KD is character-

ized by the class of serial frames, i.e. by frames in which no world is a dead
end. If A is not a KD-theorem this means that there is a serial model M =
W, R, V which is a countermodel to A. Let us now build a model M+ =
W+, Wτ , R+, V+ which is like M except for the fact that whenever in M
we have xRy, in M+ we have xRy+ and xRy–, where y+ and y− are like y
with the difference that y+ ∈ Wτ and y− ∈ / Wτ (Wτ having the properties
described at page 404). M+ is a τ -forked model and has the properties of a
KΔτ -model. Since A does not contain occurrences of τ , if A receives value
0 in M it receives value 0 also in the KΔτ -model M+. By the soundness of
KΔτ proved in Theorem 1 and by Theorem 7 then A is not a thesis of Kτ
+ DefΔ.
Necessity and Relative Contingency 413
405

By theorems 6 and 7, theorems 10 and 11 therefore entail that any wff

A of LΔτ + Def which translates into a τ -free wff of Lτ + DefΔ is a
theorem of KΔτ + Def iff it is a theorem of KD.

§4. A problem suggested by the preceding analysis concerns the possibility

of defining in terms of Δ a -operator which has the properties not of KD
but of the minimal normal system K. A proposal is as follows. Let us call
KΔτ w a system which is KΔ extended not with ∇τ but with the weaker
KΔ4*. Δτ ⊃ Δp
A possible intuitive reading of KΔ4* is that the absoluteness of τ implies
logical determinism (i.e. that every proposition p is non-contingently true
or false).
KΔ4* is equivalent to the wff ∇τ ∨ Δp and is implied by ∇τ , so the new
system KΔτ w is included in KΔτ .
KΔτ w-models are as KΔτ -models with the only difference that the relation
R is τ -forked provided that the accessibility sphere is not empty. Formally
speaking, the condition (iii) of page 404 must be replaced by
(iii’) ∀x(¬∃ y xRy ∨∃y1 ∃y2 ((xRy1 ∧ y1 ∈ Wτ ) ∧ (xRy2 ∧ y2 ∈
/ Wτ ))
Let us call such models weakly τ -forked models.
It is straightforward to see that the new system is sound with respect to
proposed semantics, i.e. that

Theorem 12. Δτ ⊃ Δp holds in all weakly τ -forked models.

Proof. Suppose there is some world x in some weakly τ -forked model which
gives value 0 to Axiom KΔτ *, which means that in x ∇τ is 0 and Δp is 0.
There are two alternatives to be considered:
1)¬∃y xRy. In this case in x Δp is vacuously true for every instance of p,
contrary to the supposition
2) ∃y1 ∃y2 ((xRy1 ∧ y1 ∈ Wτ ) ∧ (xRy2 ∧ y2 ∈ Wτ )). In this case ∇τ is true
in x, contrary to the supposition.

Remark. Theorem 12 implies that the two systems KΔτ and KΔτ w are
distinct. In fact the axiom ∇τ of KΔτ takes value 0 in weakly τ -forked
model where W = {x} and x is a dead end.

Let us consider now a system Kτ w which is K extended with the axiom
(Necτ w ) τ ≡ ¬τ .
414
406 C. Pizzi

Obviously Necτ w is equivalent to ♦τ ≡ ♦¬τ . Notice that a theorem of

Kτ w is τ ⊃ (τ ∧ ¬τ ), so also τ ⊃ ⊥ and τ ⊃ p. By a parallel
argument we reach the thesis ¬τ ⊃ p, so also the thesis
(τ ) (¬τ ∨ τ ) ⊃ p.
Since two substitution instances of p are τ and ¬τ , consequences of
(τ ) are τ ⊃ ¬τ and ¬τ ⊃ τ , so τ ≡ ¬τ .
(τ ) and (Necτ w) provide then two equivalent axiomatizations of the same
system.
If Def  and DefΔ are as in §0, we show that in respect of wffs of LΔτ w + Def
 and their -variants, KΔτ w + Def and Kτ w + DefΔ are equivalent
systems.
Remark. The equivalence of the two systems which will be proved does
not mean that the box operator defined in KΔτ w + Def has the same
meaning of the box operator axiomatized in Kτ w + DefΔ. Despite the
typographical identity of the symbols, they have different properties, and
the box operator of Kτ w + DefΔ cannot be expressed in KΔτ w + Def:
on this point see the remarks expressed in §4.

Theorem 13. For every A, if A is a thesis of KΔτ w + Def, A is a thesis

of Kτ w + DefΔ.

Proof. By induction on the length of the proofs.

1) The axiom of KΔτ w, Δτ ⊃ Δp, follows from (τ ), i.e. from (τ ∨ ¬τ )
⊃ p. By PC in fact we have (τ ∨ ¬τ ) ⊃ (p ∨ ¬p) and by DefΔ the
required wff.
The -variants of the other axioms are K-theorems, and the proof is as
in §3. The same holds for the Necessitation Rule
2) We have to prove in Kτ w from the axiom (τ ∨ ¬τ ) ⊃ p the equiv-
alence ( ≡) p ≡ ((p ∨ ¬p) ∧ ((τ ⊃ p) ∨ (τ ∧ ¬p)).
The problem obviously is the implication from right to left.
Now the axiom (τ ∨ ¬τ ) ⊃ p yields the theorem
(1) ♦¬p ⊃ (♦τ ∧ ♦¬τ ).
In §3 (p. 409) we proved that ♦τ ∧ ♦¬τ implies (p ∨ ¬p) ∧ ((τ ⊃ p)
∨ (τ ∧ ¬p) ⊃ p, so by transitivity and (1), ♦τ ∧ ♦¬τ implies the wff
♦¬p ⊃ ((p ∨ ¬p) ∧ ((τ ⊃ p) ∨(τ ∧ ¬p) ⊃ p).
But ¬♦¬p, i.e. p, also implies ((p ∨ ¬p) ∧ ((τ ⊃ p) ∨ (τ ∧ ¬p)
⊃ p) simply because its form is C ⊃ p, and this concludes the proof.
Necessity and Relative Contingency 415
407

On the other hand, we prove

Theorem 14. For every A, if A is a thesis of Kτ w + DefΔ, A is a thesis
of KΔτ w + Def.
Proof. 1) Axiom Necτ w is a thesis of KΔτ w + Def. Given Def, a
theorem of KΔτ w + Def is in fact τ ≡ (Δτ ∧ Δ(τ ⊃ τ )), so τ ≡
Δτ . Thanks to the equivalence ¬τ ≡ (Δ¬τ ∧ Δ(τ ⊃ ¬τ )), we have also
the theorem ¬τ ≡ Δ¬τ . But Δτ ≡ Δ¬τ holds by axiom KΔ1, so by
replacement of proved equivalents τ ≡ ¬τ is a thesis of KΔτ w +Def .
2) Looking at the proof of Theorem 7 we see that we may easily use a K-
tableau to prove that the Δ-variant of axiom K is implied by ∇τ . We have
then simply to prove that the same wff follows from some instance of Δp, so
it will be proved that it follows from some instance of the axiom ∇τ ∨ Δp.
This is however obvious since the Δ-variant of K is
(KΔ) (Δ(p ⊃ q) ∧ Δ(τ ⊃ (p ⊃ q) ∧ Δp ∧ Δ(τ ⊃ p)) ⊃ (Δq ∧ Δ(τ ⊃ q))
where the two consequents Δq, Δ(τ ⊃ q) are instances of Δp, so that KΔ is
derived by the PC-theorem p ⊃ (q ⊃ p).
For the Δ-variant of Necessitation rule see p. 413.
3) The derivation of the equivalence
(Δ ≡) Δp ≡ ((Δp ∧ Δ(τ ⊃ p)) ∨ (Δp ∧ Δ(τ ⊃ ¬p)))
which is the Δ-variant of Def Δ, has some interest only in the direction from
left to right.
As instances of the axiom ∇τ ∨ Δp we have
a) ∇τ ∨ Δ(τ ⊃ p)
b) ∇τ ∨ Δ(τ ⊃ ¬p), so also
c) ∇τ ∨ (Δp ∧ Δ(τ ⊃ p)) ∨ (Δp ∧ Δ(τ ⊃ ¬p))
So by PC we have
(d) Δτ ⊃ ((Δp ⊃ (Δp ∧ Δ(τ ⊃ p)) ∨ (Δp ∧ Δ(τ ⊃ ¬p)))
On the other hand, it was proved at page 411 that (Δ≡) follows by the
axiom ∇τ , so that we have
(e) ∇τ ⊃ ((Δp ⊃ (Δp ∧ Δ(τ ⊃ p)) ∨ (Δp ∧ Δ(τ ⊃ ¬p)))
Thus the result follows by (d), (e) and PC.
It is not difficult to devise a tableau procedure for Kτ w derived from
the one for KΔτ . We have simply to take into account the fact that every
diagram of a tableau may be either a dead end (i.e. it sees no other dia-
gram) or τ -forked as before. This introduces a complexity in testing valid
416
408 C. Pizzi

formulas because of the high number of alternatives to be considered, while

the refutation of a wff is not necessarily more complex than in Kτ (since
if A is not a K-theorem it is not a Kτ w-theorem).
We omit here the proof of the following theorems, which may be proved
by a suitable extension of the proofs of Theorems 8 and 9 (page 411):

Theorem 15. KΔτ w is complete with respect to the class of KΔτ w-models.

Theorem 16. KΔτ w + Def  is tableau-decidable.

By a suitable modification of the proof of Theorem 11 we may also prove

that the fragment of KΔτ w containing simply  and truth functional oper-
ators is exactly K.
The proof amounts to proving two theorems, the first of which is trivial.

Theorem 17. For every τ -free A, if A is a thesis of K, A is a thesis of

Kτ w + Def Δ.

Theorem 18. For every τ -free A, if A is not a thesis of K, A is not a thesis

of Kτ w + Def Δ.

Proof. If A is not a K-theorem, there is an open K-tableau T having A

as input. Then there is a K-model M falsifying A which has, in the more
complex cases, both dead ends and non-dead ends. Moving from M we build
now a new K-model M* which is coincident with M as far dead ends are
concerned, but is like the forked model M+ of Theorem 11 as far as non
dead ends are concerned.
The new M* contains then in the worst case both dead ends and τ -
forked worlds, so it is a model having the properties of a KΔτ w-model. It
is clear that M* is also a countermodel for A, since the assignment to τ of
value 1 or 0 in M* cannot yield new inconsistencies, given that A has no
occurrences of τ . So A is not KΔτ w-valid, not a KΔτ w-theorem so also not
a Kτ w-theorem.
A consequence of theorems 13, 14, 17, 18 is that any wff A of LΔτ +
Def which translates into a τ -free wff of Lτ + DefΔ is a theorem of
KΔτ w + Def iff it is a theorem of K.

§5. As a final comment, we have to stress that the relation between the
modal system Kτ + DefΔ and the equivalent contingential system KΔτ
+ Def is not the same which obtains between Kτ w + Def Δ and the
equivalent contingential system KΔτ w + Def. The characteristic axiom
♦τ ∧ ♦¬τ of Kτ is the -variant of the characteristic axiom ∇τ of KΔτ ;
Necessity and Relative Contingency 417
409

and, conversely, ∇τ is equivalent to the Δ-variant of ♦τ ∧ ♦¬τ . But this

one-one translation among the characteristic axioms does not hold when we
consider the couple of systems Kτ w+ Def Δ and KΔτ w+ Def. In fact
the axiom (τ ∨ ¬τ ) ⊃ p of Kτ w is not the -variant of the axiom
Δτ ⊃ Δp of KΔτ w: the -variant of Δτ ⊃ Δp is in fact (τ ∨ ¬τ ) ⊃ (p
∨ ¬p), which is weaker than (τ ∨ ¬τ ) ⊃ p. In the other direction,
the Δ-variant of the axiom (τ ∨ ¬τ ) ⊃ p of Kτ w is equivalent to
Δτ ⊃ (Δp ∧ Δ(τ ⊃ p)), which is stronger than Δτ ⊃ Δp. So it appears
that the necessity operator axiomatized in Kτ w+ Def Δ has not exactly
the same properties of the operator introduced in KΔtw+ Def by Def .
To illustrate this point let us remark that if we consider the simple KΔ +
Def, thanks to Def we obtain the equivalence τ ≡ Δτ ; so that, given the
equivalence Δτ ≡ Δ¬τ granted by axiom KΔ1, by replacement we obtain
the equivalence τ ≡ ¬τ . But the equivalence τ ≡ ¬τ is introduced
in Kτ w+ DefΔ as an axiom, Nectw (equivalent in Ktw to (τ ∨ ¬τ )
⊃ p) which, when subjoined to K + DefΔ, yields as a theorem Δτ ⊃ Δp,
while this wff is independent from KΔ + Def.
The problem is that if (τ ∨ ¬τ ) ⊃ (p ∨ ¬p) is taken as an axiom
for the necessity operator in place of (τ ∨ ¬τ ) ⊃ p or τ ≡ ¬τ the
resulting system — let us call it Kτ w — is a system weaker than Kτ w,
and it may be seen that it is impossible to prove in it as a theorem the
equivalence (≡) proved at page 409. In particular we cannot prove in Kτ w
the implication
( ⊃) ((p ∨ ¬p) ∧ ((τ ⊃ p) ∨ (τ ∧ ¬p)) ⊃ p.
Let us consider in fact a K-frame constituted by a single reflexive world
x and and the value assignment V(p, x) = 0. This assignment implies the
falsity of p in x. But every instance of ♦p ⊃ p, i.e. of ¬p ∨p is true
in this frame, so every instance of ((p ∨ ¬p) ∧ ((τ ⊃ p) ∨ (τ ∧ ¬p))
is also such, as well as every instance of the axiom ((τ ∨ ¬τ ) ⊃ (p
∨ ¬p). So ( ⊃) and ( ≡) cannot be theorems of the system Kτ w.
The above argument of course does not work if we choose as an axiom
(τ ∨ ¬τ ) ⊃ p. In fact p turns out to be always true in a dead end
for every assignment to p, while it may be true or false in a reflexive world.
The wff Δp (and its -variant p ∨ ¬p) on the contrary is unable to make
this distinction since it is true at every world x with one or the other of the
mentioned properties4 .

4
A theorem concerning this semantic aspect of the language LΔ may be found in J.
Marcos [2005], p.203.
418
410 C. Pizzi

It appears then that there are expressive limits of the contingential lan-
guage which cannot be easily overcome by the introduction of a propositional
constant. It may be conjectured that a different definition of necessity in
terms of contingency or the introduction of stronger constrains on the propo-
sitional constant τ might allow to identify a modal system with the required
properties. The problem opened by this difficulty however lies beyond the
scope of the present paper.

Acknowledgements. The author is grateful to the anonymous referee

for the suggested improvements and especially for putting in evidence the
difficulty exposed in §4.

References
[1] Cresswell, M. J., ‘Necessity and Contingency’, Studia Logica 47 (1988), 145–149.
[2] Hughes, G. E., and M. J. Cresswell, An Introduction to to Modal Logic, Methuen,
London, 1968.
[3] Humberstone, I. L., ‘The Logic of Non-Contingency’, Notre Dame Journal of Formal
Logic 36 (1992), 214–234.
[4] Humberstone, I. L., ‘The Modal Logic of Agreement and Non Contingency’, Notre
Dame Journal of Formal Logic 43 (2002), 95–127.
[5] Kuhn, T. S., ‘Minimal Non-contingency Logics’, Notre Dame Journal of Formal Logic
36 (1995), 230–234.
[6] Marcos, J., Logics of Formal Inconsistencies, Ph.D. Thesis, UNICAMP, Campinas
University, 2005.
[7] Pizzi, C., ‘Contingency Logics and Propositional Quantification’, Manuscrito, UNI-
CAMP, Campinas University, 22, (1999), 283–303.
[8] Pizzi, C., ‘A Logic of Contingency with a Propositional Constant’, in E. Ballo and
M. Franchella (eds.), Logic and Philosophy in Italy, Polimetrica, Milano, 2006, pp.
141–151.

Claudio Pizzi
Dipartimento di Filosofia e Scienze Sociali
Università di Siena
via Roma 47 -53100 Siena
Siena, Italy
[email protected]