You are on page 1of 6

Exhaustive abduction over formal systems

Alexandræ Bali
April 26, 2024

Abstract
Deduction is the process of finding conclusions from premises, whereas abduc-
tion is the process of figuring out premises from conclusions. Formal systems
are comprised of a formal language and a deductive system. From a deductive
system, it is quite easy to form a dual abductive operation, although it may be
notationally complicated. However, abduction by duality from deduction tends
to be non-exhaustive. It is the goal of this paper to explicate foundations to
exhaustive abduction by duality.

To notate proofs that use both abductive and deductive reasonings, we will introduce a
family of diagrams, in the form of a directed graph, such that each node is a sequent, and
each arrow is labeled with a rule of inference with which we infer the first member of the
ordered pair to the second one. However, what makes it a diagram is that arrows are always
directed south. When they’re directed to the left (↙) indicate abductive reasoning, whereas
arrows directed to the right (↘) indicate deductive reasoning. This means we can read the
diagram left-to-right to understand the actual order of reasoning. As an example, here are
two ways of writing left-hand side conjunction elimination :

Γ⊢A∧B Γ⊢A∧B

∧EL ∧EL

Γ⊢A Γ⊢A

On the diagram on the left, we start from Γ ⊢ A ∧ B, and then we deduce Γ ⊢ A. On the
other side, we start from Γ ⊢ A, and then we abduce Γ ⊢ A ∧ B.

Formulae in natural deduction can have one of many forms. We will use a nomenclature for
such formulae in natural deduction, based on the highest-priority operation :
− Literal formulae : p, q, r, etc.
− Purely conjunctive formulae : p ∧ q, (p ∨ q) ∧ (r ⇒ s), etc.
− Purely disjunctive formulae : p ∨ q, (p ⇒ q) ∨ (¬r ∧ s), etc.
− Purely negative formulae : ¬p, ¬(p ∧ q), ¬(p ⇒ q ∧ r), etc.
− Purely conditional formulae : p ⇒ q, (p ∧ q) ⇒ (r ∨ s), etc.

1
A few remarks will help us make exhaustive abduction, by truncating as many abductive
avenues as possible :
− We can only directly deduce a purely ◦ formula from a rule of inference that either
concludes a purely ◦ formula or a literal formula ; all others generate purely • formulae,
with ◦ ≠ •. To achieve our goal of reducing possibilities, we could disallow subsequent
theorems and only use basic rules of inference.
− A proof being constituted of a finite number of inferences, this means the set of number
of inferences to prove a given sequent, being a subset of the natural numbers, necessarily
has a minimal value. This means there are proofs with as few inferences as possible.
One can then assume we’re in one such proof.
We will use the following inference rules for the remainder of the article :
Name of rule Symbolic notation
Γ ⊢A Γ ⊢B
Conjunction introduction ∧I

Γ ⊢A∧B
Γ ⊢A∧B Γ ⊢A∧B
Conjunction elimination ∧EL ∧ER

Γ ⊢A Γ ⊢B
Γ, A ⊢ B
Conditional introduction ∧I

Γ ⊢A⇒B
Γ ⊢A⇒B Γ ⊢A
Conditional elimination ∧I

Γ ⊢B
Γ ⊢A Γ ⊢B
Disjunction introduction ∨IL ∨IR

Γ ⊢A∨B Γ ⊢A∨B
Γ ⊢A∨B Γ, A ⊢ C Γ, B ⊢ C
Disjunction elimination ∧I

Γ ⊢C
Γ, A ⊢ ⊥
Negation introduction ∧I

Γ ⊢ ¬A
Γ ⊢ ¬A Γ ⊢ A
Negation elimination ∧I

Γ ⊢⊥

2
In classical logic, we also use the law of excluded middle :

LEM

⊢ A ∨ ¬A

A ∨ ¬A is a very specific formula : it is purely disjunctive, and as long as we assume


consistency, both of its disjuncts are non-tautological. Indeed, A being variable, we cannot
entail A, nor can we entail ¬A : in other words, ̸ ⊢A and ̸ ⊢ ¬A. However, such formulae
cannot be entailed, unless we axiomatise one such formula. As an application to the elements
of theory of exhaustive abduction we’ve presented so far, let’s prove the latter statement.
For the sake of convenience, we’ll allow the use of hypothetical syllogisms, which means
from Γ ⊢ A ⇒ B and Γ ⊢ B ⇒ C, we directly derive ⊢ A ⇒ C ; this will only affect
abductions over purely conditional formulae. Finally, we will also allow to directly conclude
any conjunct from a chain of conjunctions, which will be implied whenever we mention
conjunction elimination.

Let A and B be non-tautological formulae, which means ̸ ⊢A and ̸ ⊢B. If we want to prove
A ∨ B, this would mean we prove a purely disjunctive formula. Let’s assume we have a proof
of A ∨ B with as few inferences as possible. As said earlier, the only rules that allow a direct
proof to a purely disjunctive formula are one of two kinds : those that directly conclude
a purely disjunctive formula, or those that conclude a literal formula. Out of all rules of
inference, only both rules of disjunction introduction yield a purely disjunctive conclusion.
We however find these :

⊢A ⊢B

∨IL ∨IR

⊢A∨B ⊢A∨B

Either way, these contradict the non-tautology of A and B respectively, meaning these are
not possible final elements of its proof. We conclude that it could be proved via a rule of
inference with a literal formula as its conclusion. This only includes the rules of conjunction
elimination, conditional elimination, and disjunction elimination.

Disjunction elimination
If we entailed A∨B through disjunction elimination, we’d have ⊢ C ∨D such that C ⊢ A∨B
and D ⊢ A ∨ B. However, if ⊢ C or ⊢ D, then ⊢ A ∨ B would have a shorter proof from
simply using the theorem ⊢ C and juxtaposing the proof for C ⊢ A∨B : as ⊢ C, this suffices
to prove ⊢ A ∨ B. This would contradict the hypothesis of inferential minimality. Therefore,
C ∨ D would be another disjunction with non-tautological disjuncts. As we’ve abduced ⊢ C
or ⊢ D, we can rewrite our statements with the following diagram :

3
2 [⊢ C ]

··· ∨I
Proof for C ⊢A∨B

⊢C ∨D
1 C ⊢A∨B ⊢A∨B
D ⊢A∨B

∨E

0 ⊢A∨B
The numbering on the left keeps track of how far back in the proof we are, compared to
⊢ A ∨ B. This scheme of reasoning actually applies to any formula E such that E ⊢ A ∨ B
in the long run : if ⊢ E were derived through ∨E, we would always be able to find a proof
for ⊢ A ∨ B in at least one less step than the proof at hand, which contradicts the hypothesis
of inferential minimality. As a diagram, this reasoning resembles the following :

3 [⊢ E ]

··· ∨I
Proof for E ⊢C

⊢E∨F
2 E ⊢C ⊢C
F ⊢C
Proof for C ⊢A∨B
∨E

1 ⊢C ⊢A∨B

Proof for C ⊢A∨B

0 ⊢A∨B
(1)
That way, we can cross out disjunction elimination from most subsequent reasonings.

Conditional elimination
There are many ways to entail A∨B through conditional elimination, which we’ll go through.
Although, the first step will always be the following, for some formula C :

4
⊢ C ⇒ (A ∨ B)
1
⊢C
(2)
∨E

0 ⊢A∨B
What’s tricky in (2) is that we have to both abduce ⊢ C ⇒ (A ∨ B) and ⊢ C at once. What
we know in the end is that ⊢ C and ⊢ C ⇒ (A ∨ B), so we necessarily have both proved at
some point in the proof, which means we can look at one and know we can use the other.
Let’s first look at ⊢ C.

First branch : ⊢ C
Without further information about C as a formula, we can only treat it as a literal formula.

If we could derive ⊢ C through disjunction elimination, then since ⊢ C ⇒ (A ∨ B), we would


end up proving ⊢ A ∨ B. Because of (1), this means that whatever formulae D and E such
that ⊢ D ∨ E, ⊢ D ⇒ C and ⊢ E ⇒ C, would have to be non-tautological.

If we could derive ⊢ C through conditional elimination, then there’d exist D a formula such
that :

⊢C⇒(A∨B)

⊢ C ⇒ (A ∨ B)

⊢D⇒C
2
⊢D

⇒E HS

1 ⊢C ⊢A∨B

This contradicts the hypothesis of inferential minimality.

If we could get ⊢ C ∧ D from conjunction elimination, we could construct the following


diagram :

3 ⊢ (C ∧ D) ∧ E

∧E ∧∧EL

2 ⊢C ∧D ⊢C

∧E

1 ⊢C

This contradicts the hypothesis of inferential minimality.

5
Second branch : ⊢ C ⇒ (A ∨ B)
Most reasonings that work with ⊢ C can be converted into reasoning that work with ⊢ C ⇒
(A ∨ B). However, since C ⇒ (A ∨ B) is a purely implicative formula, this means we could
have proven it from conditional introduction. However, if it were so, this would mean there
exists a formula D such that D ⊢ C ⇒ (A ∨ B) and ⊢ D, meaning we can juxtapose the
proof for D ⊢ C ⇒ (A ∨ B) right after ⊢ D, creating a shorter proof, which contradicts the
hypothesis of inferential minimality.

Conjunction elimination
If we could get ⊢ A ∨ B from conjunction elimination, this would mean that there exists a
formula C such that ⊢ (A∨B)∧C. Since (A∨B)∧C is a purely conjunctive formula, it could
be derived through conjunction introduction : this would, however, require that ⊢ A ∨ B
were already proven, which contradicts the hypothesis of inferential minimality. Similarly,
if we’d derived ⊢ (A ∨ B) ∧ C from disjunction elimination, there’d exist formulae D and E
such that ⊢ D ∨ E, D ⊢ (A ∨ B) ∧ C and E ⊢ (A ∨ B) ∧ C. Because of (1), both D and E
would have to be non-tautological formulae.

As for conjunction elimination, this would mean there exists a formula D such that ⊢ (A∨B)∧
C ∧D, and because we allow a generalised conjunction elimination for chains of conjunctions,
we can directly derive ⊢ A ∨ B in one less step than in the former proof, which contradicts
the hypothesis of inferential minimality.

This only leaves conditional elimination, requiring a formula D such that ⊢ D ⇒ (A ∨ B)


and ⊢ D. This couldn’t be done without having proved some sort of disjunction with non-
tautological disjuncts, due to an analogous reasoning as when we studied the possibility of
⊢ A ∨ B being finally derived from conditional elimination.

Conclusion
All in all, this shows that the proof of a disjunction with non-tautological disjuncts requires
a proof of another such formula prior to the former, indicating that it is impossible to deduce
disjunctions with non-tautological disjuncts without assuming one. Thus, since p ∨ ¬p is one
such formula, we cannot derive p ∨ ¬p, unless we already assume one such formula, within
the context of the rules of inference we’ve established.

You might also like