Download a theorem on well-finite standard consequence operations

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
Bulletin of the Section of Logic
Volume 2/3 (1973), pp. 159–163
reedition 2014 [original edition, pp. 159–165]
Stephen L. Bloom
A THEOREM ON WELL-FINITE STANDARD
CONSEQUENCE OPERATIONS
In this note we obtain a theorem characterizing those finite consequence
operations on a sentential language that can be defined in a proof-theoretic
way from a finite number of axiom schemes and/or structural (polynomial)
rules. This theorem can be regarded as a strengthening of a theorem of
Tarski ((T), theorem 27) on finite axiomatizability.
Let L be an absolutely free algebra of a finite similarity type having a
countably infinite number of generators. We will refer to the members of
L as polynomials. P (L) is the collection of all subsets of L.
1. Definition. A consequence operation on L is a function C : P (L) →
P (L) such that
1.1. X ⊆ C(X) = C(C(X)), all X ⊆ L.
1.2. If X ⊆ Y , then C(X) ⊆ C(Y ), all X, Y ⊆ L.
C is a finite consequence operation if
[
C(X) = (C(Y ) : Y ⊆ X, Y finite).
C is a structural consequence operation if
for every endomorphism h of L,
h(C(X)) ⊆ C(h(X)).
C is standard if C is both finite and structural.
2. Definition. An n-ary rule r is a subset of Ln ; if (π, δ1 , . . . , δn ),
n ≥ 0, is a sequence of members of L, the polynomial rule r(π1 δ1 ...δn ) is the
(n + 1)-ary rule defined as follows:
160
Stephen L. Bloom
(a, b1 , . . . , bn ) ∈ r(π1 δ1 ...δn ) iff for some endomorphism h of L, h(π) = a
and h(δi ) = bi , i = 1, . . . , n.
Let Q be any set of rules.
3. Definition. CQ : P (L) → P (L) is the function defined as follows:
a ∈ CQ (X) iff there is a finite sequence x1 . . . xn such that xn = a and
for each i, 1 ≤ i ≤ n, xi ∈ X or there is some (k + 1)-ary rule r ∈ Q,
and j1 , . . . , jk < i such that (xi , xj1 , . . . , xjk ) ∈ r. (In this case, we call the
sequence x1 , . . . , xn a CQ -proof of a from X.)
The following theorem is due to Roman Suszko and has been quoted
in (RW).
4. Theorem. C is a standard consequence operation on L iff there is a
finite or countable set Q of polynomial rules such that C = CQ .
5. Definition. A consequence operation C is well-finite if C = CQ for
some finite set of polynomial rules.
6. Example. The classical two-valued propositional calculus can be given
by 3 unary polynomial rules (axiom schemes) and one 3-ary polynomial
rule (modus ponens, considered as the rule r(p,q,(q→p)) (where p, q are free
generators of L.)
7. Definition. Let C1 , C2 be consequence operations on L. C1 ≤ C2 iff
every X ⊆ L, C1 (X) ⊆ C2 (X).
It is well-known that with this ordering the set of all consequence operations on L forms a complete lattice.
8. Lemma. Suppose C1 ≤ C2 ≤ C3 ≤ . . . is an infinite chain of finite
consequence operations. Let C = sup{Cn : n = 1, 2, . . .}. Then, for every
X ⊆ L,
[
C(X) =
Cn (X).
n
S
Proof. Let D(X) = Cn (X). It is easy to check that D satisfies 1.1
and 1.2. In order to see that D(D(X)) ⊆ D(X), assume a ∈ D(D(X)).
Then for some n, a ∈ Cn (D(X)). Since Cn is a finite consequence operation, there are b1 , . . . , bk ∈ D(X) such that a ∈ Cn ({b1 , . . . , bk }). Since the
Cn ’s are nested, there is an m such that {b1 , . . . , bk } ⊆ Cm (X). Thus
A Theorem on Well-Finite Standard Consequence Operations
161
a ∈ Cn (Cm (X)). Now Cm (X) ⊆ Cm+n (X), since Cm ≤ Cm+n ; also
Cn (Cm+n (X)) ⊆ Cn+m (Cn+m (X)). Hence a ∈ Cn+m (Cm+n (X)) =
Cm+n (X). Thus a ∈ D(X), and we have shown D is a consequence operation.
S
Clearly Cn ≤ D, so C ≤ D. Since Cn ≤ C for all n, we see Cn (X) ⊆
C(X), so that D ≤ C. Thus D = C, completing the proof.
9. Lemma. Suppose
Ci = CQi , i ∈ I, where for each i, Qi is a set of
S
rules. Let Q = (Qi : i ∈ I). Then CQ = sup{Ci : i ∈ I}.
Proof. Clearly Ci ≤ CQ , so that sup{Ci : i ∈ I} < CQ . Let D be any
consequence operation such that Ci ≤ D, all i ∈ I. Then, by induction
on the length of CQ -proofs, it is easy to see that CQ ≤ D, so that CQ =
sup{Ci : i ∈ I}.
10. Lemma. Let C be a standard consequence operation. If C is not
well-finite, there is an infinite sequence C1 ≤ C2 ≤ . . . of standard (and
well-finite) consequence operations such that C = sup{Cn : n = 1, 2, . . .}
and C 6= Cn , any n.
Proof. Let Q be an infinite set of polynomial rules such that C = CQ .
Say Q = {r0 , r1 , r2 , . . .}. Let Qn = {r0 , . . . , rn−1 }, n = 1, 2, . . . and set
Cn = CQn .
From 9 it follows that C = sup{Cn }, and since C is not well-finite,
C 6= Cn , any n.
We may now prove the main theorem.
11. Theorem. A finite structural consequence C on L is not well-finite iff
there is a sequence C1 ≤ C2 ≤ C3 ≤ . . . of standard consequence operations
such that C = sup{Cn }, C 6= Cn , any n.
Proof. Lemma 10 showed that the condition is necessary. Now suppose
C is the sup of the Cn as above. By 4 there are sets S and Qn of polynomial
rules such that C = CS and Cn = CQn , n = 1, 2, . . .. For each (1 + m)ary rule r(π,δ1 ,...,δm ) , m ≥ 0, in S, we have π ∈ CS (δ1 , . . . , δm ) (if m = 0,
π ∈ CS (∅).). Then π ∈ CQn (δ1 , . . . , δm ) for some n. Let A(π,δ1 ,...,δm ) be the
finite subset of Qn used in the CQn -proof of π from δ1 , . . . , δn . Let A be the
union of all sets A(π,δ1 ,...,δn ) . In order to obtain a contradiction, suppose S
is a finite set of rules. Then A is finite, and we can find an integer n0 such
162
Stephen L. Bloom
that if r(π,δ1 ,...,δm ) is a rule in S and A(π,δ1 ,...,δm ) ∈ Qn then n ≤ n0 .
We will show C = CS ≤ CA ≤ Cn0 . Indeed, suppose a ∈ CS (X). Let
x1 , . . . , xt be a CS -proof of a from X. We show that for each i, 1 ≤ i ≤ t,
xi ∈ CA (X).
It i = 1, then either x1 ∈ X or x1 follows from a unary rule rπ in S.
In the first case, clearly x1 ∈ CA (X). In the second case, x1 = h(π) for
some endomorphism h of L. But by construction, π ∈ CAπ (∅), and hence
π ∈ CA (∅). By 4 then x1 = f (π) ∈ CA (∅) ⊆ CA (X).
Now assume for j < i, xj ∈ CA (X). If (xi , xj1 , . . . , xjm ) belongs to the
rule r(π,δ1 ,...,δm ) in S, then, for some endomorphism h of L, xi = h(π) and
xjk = h(δk ), k = 1, . . . , m. By construction, π ∈ CA(π,δ1 ,...,δm ) (δ1 , . . . , δm ) ⊆
CA (δ1 , . . . , δm ). Since all rules in A are polynomial rules,
h(π) ∈ CA (h(δ1 ), . . . , h(δm )), by 4; i.e.
xi ∈ CA (xj1 , . . . , xjm ).
But by the induction assumption, {xj1 , . . . , xjm } ⊆ CA (X). Thus xi ∈
CA (CA (X)) = CA (X), completing the proof that CS ≤ CA .
Since A is a finite set of rules, each one of which belongs to some Qn ,
n < n0 , it follows from 8 that CA ≤ sup{Cn : n < n0 } ≤ Cn0 .
Thus on the assumption that S is finite, we have shown C = CS ≤ Cn
for some n0 . But this contradicts the assumption that Cn < C, all n, and
completes the proof of the theorem.
12. Example. Let L be the sentential language having only one unary
connective, say F . Let q be one of the free generators of L, and let F n x
be an abbreviation for F F . n. . F x. If 2 = p0 , 3 = p1 , . . . are the primes in
increasing order, let Rn be the binary polynomial rule determined by the
sequence (F pn q, q). Finally, put Q = {R0 , R1 , . . .}. CQ is not well-finite.
Indeed, if so, CQ ≤ CQn for some n, where Qn = {R0 , . . . , Rn }. But for
example
F pn+1 q ∈ CQ (q) but
F pn+1 q 6∈ CQn (q),
as can be easily proved by induction on the length of CQn -proofs.
The author is pleased to acknowledge several helpful conversations with
Roman Suszko on the topic of this paper.
A Theorem on Well-Finite Standard Consequence Operations
163
References
[T] A. Tarski, Fundamental Concepts of the Methodology of the Deductive Sciences, Reprinted in Logic, Semantics and Metamathematics,
Oxford (1956).
[RW] R. Wójcicki, Investigations into methodology of sentential calculi
(I), Institute of Philosophy and Sociology, Polish Academy of Sciences,
Warsaw (1971).
Stevens Institute of Technology
Hoboken, N.J. 07030
Related documents