5.1. Metatheorem of Deduction in Hilbert Like Calculi
The metatheorem of deduction is associated with implication. Since we are working in a general setting we need to make precise what is a Hilbert calculus with implication.
We say that a Hilbert schema calculus
has implication if
and the following schema metatheorems of modus ponens (MTMP) and deduction (MTD)
hold.
The schema MTD can be characterized in by some schema lemmas. That is, a deductive system has the MTD, provided that certain schema lemmas hold. Consider the following pairs:
(L1) ;
(L2) ;
(LD) for every schema derivation rule where is a fresh schema variable.
The following result gives a necessary and sufficient condition for MTD to hold in a Hilbert schema calculus.
Proposition 2. Let be a Hilbert schema calculus with and MTMP. Then MTD holds in iff L1, L2, LD are schema lemmas for every .
Proof. We must prove the three properties:
(L1) Observe that and so by MTD .
(L2) Note that and so by MTD .
(LD
) Assume that
. Since
for
then using MTMP, we have
for
. Hence,
and so by MTD
.
Assume that . Let be a schema derivation of from and . We prove the result by induction on m.
(Basis) . There are four possibilities.
(a)
is a schema axiom. Then consider the schema derivation:
(b)
is an hypothesis. Then consider the schema derivation:
(c)
is schema provable, that is
. Then consider the schema derivation:
where sThm means that
is a schema theorem (hence, has a schema proof).
(d)
is
. Then consider the schema derivation
(Step)
is an instance of the conclusion of a schema derivation rule
r and the instances of the premises are
assuming that
r has
n premises. Hence,
and so, by the induction hypothesis,
, for every
. Then consider the following schema derivation:
1 | | IH |
| ⋮ |
n | | IH |
| | LD: |
shows that
. □
In order to avoid the overloading of the notation we omit the reference to schema lemmas L1, L2, LD for in the meta conclusion of the MTD. Before extending the previous result to schema derivations using schema lemmas, we establish a useful result for the schema proof rules.
Proposition 3. Let be a Hilbert schema calculus with and MTMP. Assume that L1, L2, LD for every are schema lemmas. Then, for every instance of a schema proof rule
(LP)
is a schema lemma, provided that are schema theorems and ξ is a fresh schema variable.
Proof. Assume that is an instance of a schema proof rule and that are schema theorems. Then is also a schema theorem. So, using L2, it follows that is a schema theorem. □
Let be a pair and define where is a fresh schema variable. The following result is an immediate consequence of Proposition 2.
Proposition 4. Let be a Hilbert schema calculus with implication. Then Let
be a set of schema lemmas. We define
The following result is useful for detailing how to get a schema derivation for from out of a schema derivation for from and in the presence of schema lemmas.
Proposition 5. Let be a Hilbert schema calculus with and MTMP, and Λ be a set of schema lemmas. Then if L1, L2, LD for every are schema lemmas then
hold.
Proof. We only prove that holds. Let be a schema derivation for . Consider the sequence obtained from as follows: for each ,
is either an axiom or an element in
. Then replace
by
| | or |
j | | L2:j |
is . Then replace by justified by L1;
is the conclusion of an instance of a schema derivation rule r. Then replace by with justification ;
is the conclusion of an instance of a schema lemma . Then replace by with justification ;
is the conclusion of an instance of a schema proof rule r. Then replace by with justification .
We now show by induction on k that is a schema derivation for .
(Base) . Then is either an axiom or an element of or is . Thus, by construction of , is a schema derivation for .
(Step) (1) is the conclusion of an instance of a derivation rule r. Then, by the induction hypothesis, for . Hence, by construction of , is a schema derivation for .
(2) is the conclusion of an instance of the schema lemma . Then, by the induction hypothesis, for . Hence, by construction of , is a schema derivation for observing that .
(3) is the conclusion of an instance of a schema proof rule r. Then, by the induction hypothesis, for . Hence, by construction of , is a schema derivation for observing that if is a theorem then must also be a theorem for and, in these conditions, LP is a schema lemma by Proposition 3. □
Example 9. Note that has implication. We now prove that has the MTD. According to Proposition 2, it is enough to show that L1, L2, and LDsMP are schema lemmas. We only show LDsMP. For that, it is enough to observe that the sequence 1 | | |
2 | | |
3 | | |
4 | | :2,3 |
5 | | :1,4 |
is a schema derivation for .
Example 10. Recall Example 5 where . In this case Recall also that has implication, as shown in Example 9. Then the sequence defined is as follows | | |
1 | | L2: |
| | |
2 | | L2: |
3 | | :1,2 |
4 | | :3 |
| | K |
5 | | L2: |
6 | | :4,5 |
| | K |
7 | | L2: |
8 | | :6,7 |
9 | | L1 |
10 | | 9,8 |
is a schema derivation for Observe that this schema derivation was obtained from the schema derivation in Example 5 using the proofs of Proposition 5 and Proposition 3.
Proposition 6. Let be a Hilbert schema calculus with implication, Λ
a set of schema lemmas and ω a schema derivation for Then there is a constant κ, such that Proof. The first two terms of the bound of the schema complexity are a direct consequence of the proof of Proposition 5. The following expression
is a bound for constant
. □
5.2. Cut in Gentzen Like Calculi
Let
be a regular Gentzen schema calculus with the following cut rule
A possible instantiation of the Cut rule is
where
is said to be the
cut formula. The
depth of a schema formula
is defined as follows:
;
.
The depth of a cut instantiation is the depth of its cut formula. For more details on cut elimination, the reader should consult [
17].
We use the following notations:
means that there is a schema derivation of
from
S in
and cut in which all the cuts have a depth of at most
d and
means that there is a schema derivation of
from
S in
and cut in which the final step is a cut of depth
d, and all the other cuts have a depth of at most
. In order to investigate the impact of the elimination of cuts of depth
d, we need to work with more fine grained complexity measures. We denote by
the complexity of the smallest schema derivation for
using cuts with at most depth
d. Moreover, we denote by
the complexity of the smallest schema derivation for
ending with an application of a cut rule of depth
d and all the other cuts have depth at most
.
Finally, we use
for denoting a cut-free schema derivation. Observe that for cut elimination we will use schema lemmas that depend on the main connective of the cut formula. Hence, we associate schema lemmas with each connective depending on its type (recall Definition 2). Let
c be a connective of type (i),
and
. Then we define
as the schema lemma
with schema derivation
1 | | Hyp |
2 | | :1 |
3 | | Hyp |
4 | | Hyp |
5 | | :2,3 |
6 | | :4,5 |
in
. We denote by
the complexity of the schema lemmas for
C.
In the next results we assume without loss of generality that in a schema derivation the cut formulas are immediately expanded after the application of the cut. The general case would follow by an additional induction on the level of the cut.
Proposition 7. Let be a regular Gentzen schema calculus (recall Definition 2) and c a connective of type (i), and . Then, the following metatheorem holdsassuming that the last step of the schema derivation for on the numerator is justified by with premises and . Moreover, Proof. Let
be a schema derivation
| ⋮ | |
k | | |
| | |
| | |
| | Lc:, |
| | Rc:k |
| | :, |
From
we have schema derivations for
, for
and for
. Hence, there is a schema derivation of
using the schema lemma:
| ⋮ | |
k | | |
| | |
| | |
| | : |
Observe that in this schema derivation, all the cuts have depth less than
d. Moreover,
The inequality holds since . □
Observe that we can establish a similar result for every connective of type (i) (with
), (ii), (iii), and (iv) in a regular Gentzen calculus with signature
C. We also provide an illustration for a connective
c of type (iii) and arity 2 (the rules for disjunction in intuitionistic logic are of this kind). Let
be a regular Gentzen schema calculus with the following cut rule
and where every schema rule has premises and conclusion sequents with a unique schema formula on the right hand side. A possible instantiation of the cut rule is
Let
be the schema lemma
with schema derivation
1 | | Hyp |
2 | | Hyp |
3 | | :1,2 |
for
in
.
Proposition 8. Let be a regular Gentzen schema calculus where every schema rule has as premises and conclusion sequents with a unique schema formula on the right hand side, and c a connective of type (iii) and arity 2. Then, the following metatheorem holdsassuming that the last step of the schema derivation for on the numerator is justified by with premises and . Moreover, Proof. Let
be a schema derivation
| ⋮ | |
k | | |
| | |
| | |
| | Lc: |
| | Rc:k |
| | : |
From
we have schema derivations for
and
for
. Hence, there is a schema derivation of
using the schema lemma
. Observe that in this schema derivation, all the cuts have depth less than
d. Furthermore,
The inequality holds since . □
We now investigate the impact on the complexity of reducing the depths of the cut in schema derivations.
Proposition 9. Let be a regular Gentzen schema calculus. Then, the following metatheorem holds Proof. Let of size be a smallest schema derivation of in using cuts of at most depth d. The proof follows by induction on the number n of cuts of depth d in .
(Base) . Then is also a schema derivation for with cuts less than or equal to . Then and .
(Step)
. Assume that
is justified by a cut of depth
d with cut formula
in positions
and
. Let
be the other sequents in
justified by cuts of depth
d. Then
for every
. Then, by the induction hypothesis,
Consider the schema derivation
obtained from
by replacing the schema derivation of
in
by a schema derivation of
where the cut of depth
d is substituted by cuts of depth less than
d for
, as given by the induction hypothesis. Then,
Observe that
is a schema derivation of
with a unique cut of depth
d. That is,
Then, looking at the proof of Proposition 7, we can say that there is a schema derivation
of
where all the cut rules are of depth less than
d. Furthermore,
Finally,
where
comes from the definition of
. □
Now we can state the cut elimination theorem.
Proposition 10. Let be a regular Gentzen schema calculus. Then, the following metatheorem holds, for every , Proof. The proof follows by induction on d.
(Base) . Immediate by definition.
(Step) Assume that
. Hence, by the induction hypothesis,
The thesis follows by Proposition 9. □
Example 12. Recall Example 2 and Example 8. Then, by Proposition 10, where is the complexity of the schema lemmas for ∧, ∨ and ⊃.
5.3. Paraconsistent Nelson’s Logic
Recall Nelson’s paraconsistent logic
([
18,
19]). We discuss how the result of the complexity of the cut elimination can be used to find an upper bound on cut elimination in N4 (using the results in [
20], namely the theoremhood reduction from
to
).
The signature
for Nelson’s paraconsistent logic is as follows:
,
and
for
. Let
be the set of formulas inductively generated from the set of schema variables
. Recall Example 2. The Gentzen schema calculus
for
over
is a Gentzen schema calculus
such that
is composed by and
includes the rules in , plus the rules
for
for and is a multiset. Observe that is not a regular Gentzen schema calculus. For instance even looking at as a new connective, it is not of any type in the definition of regular calculus. For the sake of simplicity, we assume that the depth of a schema formula with negation is defined as follows:
;
.
The reason for this is that the rules for negation only consider cases where the negation appears paired with another connective. Thus, we can regard these pairs as new connectives. We can obtain a result similar to Proposition 10 by using a reduction technique.
Proposition 11. Let be a Gentzen schema calculus. Then, the following metatheorem holds, for every , Proof. Let
be the set of formulas over
. Consider the following translation
defined as follows
;
;
for ;
;
;
;
The proof follows by induction on the length of a derivation for .
(Basis) Immediate.
(Step) We only consider the specific rules for .
Assume that
was obtained by rule
from
and
with
. Let
Then
and
. Hence,
and
are schema derivations for
and
, respectively. Thus, by the induction hypothesis,
Moreover, using
, we have
Therefore, the result follows because .
Assume that
was obtained by rule
from
with
. Let
Hence,
Thus, by the induction hypothesis,
Moreover, using
, we have
Therefore, the result follows because . The case where was applied is similar. Furthermore, the other specific rules are also similar.
The proof follows by induction on the length of a derivation for .
(Basis) immediate.
(Step) we only consider ∨.
Assume that
was obtained by rule
from
and
with
. There are two possibilities. Either
or
. The first case is straightforward using
on
. For the second case, we have
where
. Thus, by the induction hypothesis,
Moreover, applying
we get
The result follows since . The other rules are similar.
So,
where the first and the third steps are justified by
and the second step is justified by Proposition 10. Furthermore, by looking into the translation of the schema derivations in both directions, we observe that no complexity is added either to the length of the derivations or the depth of the formulas in the cuts. Thus,
since the structure of the derivation is essentially the same as the derivation in
and, thus, we can replicate the proof of Proposition 9. □