Sei sulla pagina 1di 17

Module I

Lecture 11
Axiomatization of PL

1
We had remarked in Lecture 1 that all mathematical laws must be de-
ducible from some ‘primitive’ or ‘unquestioned’ laws. These are the axioms.
As mentioned in that lecture, any formal theory has such a distinguished set
of wffs, and also rule(s) of inference to define the deduction procedure. Eu-
clid’s theory of points and lines was given as an example of a mathematical
theory in the lecture. We now turn to the axioms and rule of inference of PL
that make it a formal theory – in fact, a Hilbert system, and also underlie
many mathematical theories. Let us note, however, that there are math-
ematical theories based on logics different from PL, such as constructive
mathematics based on Intuitionistic logic.
For this part, we assume that ¬ and → are the primitive logical connec-
tives in the alphabet. The rest are defined in terms of these two connectives,
i.e. we introduce the following abbreviations.
Abbreviations:
(a) α ∧ β := ¬(α → ¬β).
(b) α ∨ β := ¬α → β.
(c) α ↔ β := (α → β) ∧ (β → α), where ∧ is defined as in (a).
Note that this is in consonance with what we have learnt about adequacy
of connectives in PL (cf. Lecture 8): {¬, →} do form an adequate set of
connectives.

1 Axiom schemata
Let α, β, γ be wffs of PL.
A1 α → (β → α) (Law of affirmation of consequent)
A2 (α → (β → γ)) → ((α → β) → (α → γ)) (Self-distributive law of
implication)
A3 (¬β → ¬α) → (α → β) (Law of contraposition)
Note that α, β, γ are any wffs of PL. So A1-A3 are referred to as axiom
schemata. For example, if p is a propositional variable, the wff

p → ((p → p) → p)

is an instance of axiom A1.

2 Rule of Inference
Modus Ponens (M P )

2
α
α→β
β
Again, α, β are any wffs of PL.
So, in the terminology used in Lecture 1, the wff β is a direct consequence
of the subset {α, α → β} of wffs, by the rule of inference M P .

3 The deduction procedure of PL


Following the definition of a syntactic consequence relation in a formal theory
(cf. Lecture 1), we have the relation ` of PL, in particular. Let Γ be any
set of wffs and α any wff in PL.

Definition 3.1. α is a syntactic consequence of Γ (denoted Γ ` α) if and


only if, there is a sequence α1 , ..., αn (:= α) such that each αi (i = 1, ..., n) is
either (i) an axiom of PL, or (ii) a member of Γ, or (iii) derived from some
of α1 , ..., αi−1 by M P .

Remark. If Γ is empty in the above, we simply write ` α, and say that α is


a theorem, the sequence α1 , ..., αn (:= α) constituting a proof of α. Note that
the αi ’s here need to be either axioms, or derived from previous members
of the sequence by M P . In particular, then for any axiom α, we have ` α.
‘`’ satisfies all the properties mentioned in Proposition 0.1 of Lecture 1,
i.e. for any set Γ ∪ {α} of PL-wffs, we have

Proposition 3.1.
(a) If ` α, then Γ ` α,
(b) Overlap: if α ∈ Γ, then Γ ` α,
(c) Dilution: if Γ ⊆ ∆ and Γ ` α, then ∆ ` α,
(d) Cut: if ∆ ` γ for each γ ∈ Γ and Γ ` α, then ∆ ` α.
(e) Compactness: If Γ ` α, then there is a finite subset Γ0 of Γ such
that Γ0 ` α.

Proof. Exercise!

The semantic consequence relation |= defined in Lecture 5 also satisfies these


properties. However, observe that the compactness property (property (v)
above) of ` comes directly from its definition, whereas it was established for
|= through the theorem presented in Lecture 6.

Example 3.1. Let us prove the PL-theorem α → α, for any wff α. By


the definition above, we need to provide a sequence α1 , ..., αn for some n,
with αn := α → α, satisfying the conditions mentioned in the Remark made

3
at the end of Lecture 15, viz. each αi should be either an axiom instance,
or derived by MP from previous members of the sequence. Consider the
following sequence, giving such a proof.
α1 := (α → ((α → α) → α) (A1)
α2 := (α → ((α → α) → α) → ((α → (α → α)) → (α → α)) (A2)
α3 := (α → (α → α)) → (α → α) (MP on α1 , α2 )
α4 := α → (α → α) (A1)
α5 := α → α (MP on α3 , α4 )

The following theorem gives a fundamental property of `: it, in fact,


relates three levels of ‘implication’ – those at the ‘object’ level (→), ‘meta’-
level (`) and ‘meta-meta’-level (‘if..then’) of discourse.

Theorem 3.2. (Deduction Theorem or D.T.) For any Γ, α, β,


if Γ ∪ {α} ` β then Γ ` α → β.

Proof. The proof is by induction on the number n of steps of derivation of


β from Γ ∪ {α}.
Basis: n = 0. β is an axiom, or β ∈ Γ ∪ {α}. If β is an axiom or β ∈ Γ, we
have a proof of α → β from Γ as follows:
β (Proposition 3.1 (a) or (b))
β → (α → β) (A1)
α → β (MP).
What if β is α? (Exercise!)
In the induction step, we consider the possibility that β is derived by MP.
So there would be two earlier steps of the form Γ ` γ and Γ ` γ → β
in the proof, for some wff γ. By induction hypothesis, Γ ` α → γ and
Γ ` α → (γ → β). Then we have the following proof of α → β from Γ:
α1 := α → γ
α2 := α → (γ → β)
α3 := (α → (γ → β)) → ((α → γ) → (α → β)) (A2)
α4 := (α → γ) → (α → β) (MP on α2 , α3 )
α5 := α → β (MP on α1 , α4 ).

We shall give several examples of PL-theorems, and look at some more


properties of ` in the next lecture, observing in the process, the usefulness
of the Deduction theorem.

4
Module I
Lecture 12
Theorems and metatheorems of PL

1
Notation. Let Γ ` 2 denote that there is some wff β such that Γ ` β and
Γ ` ¬β. We can read this as saying that Γ yields a contradiction.
The next two propositions include some important theorems and metathe-
orems of PL. Most are given as exercises. The deduction theorem comes in
handy in proving these results.

Proposition 0.1.
(a) (Hypothetical Syllogism, HS) {α → β, β → γ} ` α → γ.
(b) ¬α → (α → β).
(c) If Γ ` 2, then Γ ` α, for every wff α.
(d) (Reductio ad absurdum, RAA) If Γ ∪ {¬α} ` 2, then Γ ` α.
(e) ` (¬α → α) → α.
(f) ` ¬¬α → α.
(g) ` (α → ¬α) → ¬α.
(h) If Γ ∪ {α} ` 2, then Γ ` ¬α.
(i) ` α → ¬¬α.
(j) ` (α → β) → (¬β → ¬α).
(k) ` (β → α) → ((¬β → α) → α).
(l) If Γ ∪ {β} ` α and Γ ∪ {¬β} ` α then Γ ` α.
(m) (i) {¬(α → ¬β)} ` α.
(ii) {¬(α → ¬β)} ` β.
(n) (i) {α} ` ¬α → β.
(ii) {β} ` ¬α → β.

Proof. We prove a few of the results, and leave the rest as exercises.
(a) Use D.T.: It is clear that {α → β, β → γ} ∪ {α} ` γ.
(b) Use HS on ¬α → (¬β → ¬α) (A1) and (¬β → ¬α) → (α → β) (A3).
(c) Let Γ ` 2. So there is some wff β such that Γ ` β and Γ ` ¬β. We have
the following proof of α from Γ:
¬β → (¬α → ¬β) (A1)
¬β (assumption)
¬α → ¬β (MP)
(¬α → ¬β) → (β → α) (A3)
β → α (MP)
β (assumption)
α (MP).
(d) Let Γ ∪ {¬α} ` 2. Using (c),
Γ ∪ {¬α} ` α, as well as
Γ ∪ {¬α} ` ¬(¬α → α). Then we have the following proof of α from Γ:
¬α → α (D.T.)

2
¬α → ¬(¬α → α) (D.T.)
(¬α → α) → α (A3 on previous step and MP)
α (MP).

(c) tells us that if Γ yields a contradiction, it yields “anything” – indicating


a kind of collapse. (d) may be read as proof by contradiction: if you assume
“not α” and get a contradiction, you must have α. (h) gives this in another
way: if you assume α and get a contradiction, you must have “not α”. The
last two allow us to conclude that {α ∧ β} yields both α and β, while both
{α}, {β} yield α ∨ β.
Remark. We must point out that the Deduction theorem or Hypothetical
syllogism just serve as efficient tools for proving theorems of PL. They are
not, per se, constructions of proofs of theorems. There may be logical sys-
tems where these are not available, and then one may have to resort to other
techniques.

Exercise 0.1.

1. Prove the following theorems of PL.


(i) ` α → (α ∨ β).
(ii) ` ¬(α ∧ ¬α).
(iii) ` α → ¬(¬α ∧ β).
(iv) ` (α ∨ β) ∨ (β → α).

2. Prove the following:


(i) {α, ¬β} ` ¬(¬α ∨ β).
(ii) {¬(α ∨ β)} ` ¬α.
(iii) {α → β, γ ∨ α} ` γ ∨ β.
(iv) {α → (β ∧ γ)} ` α → β.
(v) {(α ∨ β) → γ} ` α → γ.
(vi) {α → γ} ` (α ∧ β) → γ.

Proposition 0.2. Consider PL with axiom A3 replaced by the two axioms


A30 ¬α → (α → β),
A300 (¬α → α) → α.
A3 can be derived as a theorem in this new system.

Proof. Exercise!

Notice that we have already proved A30 and A300 as theorems in Proposition
0.1. The above thus gives an alternate and equivalent axiomatization of PL
(with MP as the rule of inference).

3
1 Another equivalence relation on F
We have already met the relation (≡) of logical equivalence between wffs,
an equivalence relation on the set F of all wffs (cf. Lecture 3). Using the
relation `, we can define another binary relation on F, as follows:

α ∼ β if and only if ` α ↔ β,

for any wffs α, β in F. It can be shown easily, using the properties of `,


that ∼ is also an equivalence relation on F, thus giving the quotient set
F/ ∼. What is the relation between F/ ∼ and F/ ≡? We answer this in
the sequel.

4
Module I
Lecture 16
Other Proof Techniques: Sequent Calculus for PL

1
The sequent calculus (SC) gives us an alternative to the proof technique
we studied for the axiomatic PL. Here too, we derive theorems, or con-
clusions from a set of premisses, using effectively the same language as we
defined for axiomatic PL. The final question would of course be, do we get
anything different, i.e. is the set of theorems different here, or are we able to
derive conclusions that are different from those we would derive in axiomatic
PL? The answer is no. The conclusions are identical, but as we shall see,
the proof technique is much more amenable to implementation. From the
computational point of view, it may be more convenient to consider ‘rules’
rather than axioms, and that is embodied in SC. This is termed a Gentzen-
style proof system, as opposed to the Hilbert-style axiomatic system that
we studied so far.
So let us begin from the language of SC. The alphabet is the same as we
have taken earlier for axiomatic PL.

1 The alphabet
(a) A countable set P V of Propositional Variables (or letters) p1 , p2 , ...
(b) Logical Connectives ¬, →, ∧, ∨
(c) Parantheses (,)
So for SC, we take all the connectives as primitives.

2 Well-formed formulae (wffs)


Definition 2.1. The set F of wffs is the smallest set of strings (finite se-
quences of symbols) over the alphabet given above, such that
(a) any propositional variable is a wff (i.e. a member of F), called an
atomic wff,
(b) if α is a wff, so is ¬α,
(c) if α, β are wffs, so are α → β, α ∧ β, α ∨ β.

Abbreviation
(a) α ↔ β := (α → β) ∧ (β → α).

The fundamental concept in SC is that of a sequent.

Definition 2.2. A sequent is of the form Γ ⇒ ∆, where Γ, ∆ are sequences


of wffs, possibly empty. Γ is called the antecedent, ∆ the succedent.

The interpretation of Γ ⇒ ∆ is that the truth of all the wffs of Γ implies


that of at least one wff of ∆. We see this more formally in the following
definition of the semantic consequence relation |= for SC.

2
3 The semantic consequence |= of SC
Let Γ ⇒ ∆ be a sequent, and v a valuation.

Definition 3.1. v satisfies Γ ⇒ ∆ (written v |= Γ ⇒ ∆), if and only if


either there is a wff α in Γ such that v(α) = F , or there is a wff α in ∆ such
that v(α) = T. Equivalently, v |= Γ ⇒ ∆ if and only if whenever all the wffs
of Γ are satisfied by v, at least one wff of ∆ is satisfied by v.
|= Γ ⇒ ∆, when for all v, v |= Γ ⇒ ∆.
We simply write |= ⇒ ∆, or |= Γ ⇒, when Γ or ∆ is empty respectively.

Exercise 3.1. |= ⇒ α, whenever α is a tautology, while |= α ⇒ whenever


α is a contradiction.

4 Rules of SC
Let Γ, ∆ be sets of wffs, and α, β be wffs. There are two sets of rules: logical
and structural. The former gives a pair of rules for each logical connective.

4.1 Logical rules

(Ax)
Γ, α ⇒ ∆, α

(∧L) Γ, α, β ⇒ ∆ (∧R) Γ ⇒ ∆, α Γ ⇒ ∆, β
Γ, α ∧ β ⇒ ∆ Γ ⇒ ∆, α ∧ β

(∨L) Γ, α ⇒ ∆ Γ, β ⇒ ∆ (∨R) Γ ⇒ ∆, α, β
Γ, α ∨ β ⇒ ∆ Γ ⇒ ∆, α ∨ β

(→ L) Γ ⇒ α, ∆ Γ, β ⇒ ∆ (→ R) Γ, α ⇒ ∆, β
Γ, α → β ⇒ ∆ Γ ⇒ ∆, α → β

(¬L) Γ ⇒ ∆, α (¬R) Γ, α ⇒ ∆
Γ, ¬α ⇒ ∆ Γ ⇒ ∆, ¬α

(Cut) Γ ⇒ ∆, α Γ0 , α ⇒ ∆0
Γ, Γ0 ⇒ ∆, ∆0

3
4.2 Structural rules
(Lef t W eakening) Γ⇒∆ (Right W eakening) Γ⇒∆
Γ, α ⇒ ∆ Γ ⇒ ∆, α

(Lef t Contraction) Γ, α, α ⇒ ∆ (Right Contraction) Γ ⇒ ∆, α, α


Γ, α ⇒ ∆ Γ ⇒ ∆, α

(Lef t Exchange) Γ, α, β, Γ0 ⇒ ∆ (Right Exchange) Γ ⇒ ∆, α, β, ∆0


Γ, β, α, Γ0 ⇒ ∆ Γ ⇒ ∆, β, α, ∆0

5 The syntactic consequence `


Definition 5.1. Γ ⇒ ∆ is provable (written as ` Γ ⇒ ∆), provided there
is a derivation using the SC rules, beginning with instances of the rule Ax,
and ending with the sequent Γ ⇒ ∆.
A wff α is a theorem, if the sequent ⇒ α is provable.
Example 5.1. We prove ` ¬(α ∧ β) ⇒ (¬α ∨ ¬β).

(Ax)
(∧R) α, β ⇒ α α, β ⇒ β
(¬L) α, β ⇒ α ∧ β
(¬R) ¬(α ∧ β), α, β ⇒
(¬R) ¬(α ∧ β), α ⇒ ¬β
(∨R) ¬(α ∧ β) ⇒ ¬α, ¬β
¬(α ∧ β) ⇒ ¬α ∨ ¬β
Exercise 5.1. Give SC proofs of the following.
1. The axioms of PL.

2. (¬β → ¬α) → ((¬β → α) → β).

3. ((α → β) → α) → α.

4. (α ∧ (β ∨ γ)) → ((α ∧ β) ∨ (α ∨ γ)).


We state without proof the following.
Proposition 5.1. (Subformula property) In a cut-free proof of any se-
quent Γ ⇒ ∆, all sequents are composed of subformulae of wffs in Γ, ∆
only.
Proposition 5.2. (Cut-elimination) There is a terminating algorithm
that takes any proof of a sequent in SC, and turns it into a cut-free proof of
the same sequent.

4
Theorem 5.3. ` Γ ⇒ ∆, if and only if |= Γ ⇒ ∆.

As a corollary, we obtain that SC derives as theorems, exactly all the tau-


tologies of PL.

5
Module I
Lecture 17
Semantic Tableaux

1
Semantic tableaux (ST) and sequent calculus (discussed in the previous
lecture) are two sides of the same coin. Tableaux give an encoding of ter-
minating cut-free proof search algorithms of SC. In a tableau, every wff is
prefixed with a sign, +, or −, according to the intention of finding assign-
ments that make the wff true, or false (respectively). The definitions and
results in this lecture are taken from Goubalt-Larrecq and Mackie.
Let us consider Example 5.1 of Lecture 16 again in a slightly modified
form, with an added step.
Example 0.1. We prove ` ⇒ ¬(α ∧ β) → (¬α ∨ ¬β).

(Ax)
(∧R) α, β ⇒ α α, β ⇒ β
(¬L) α, β ⇒ α ∧ β
(¬R) ¬(α ∧ β), α, β ⇒
(¬R) ¬(α ∧ β), α ⇒ ¬β
(∨R) ¬(α ∧ β) ⇒ ¬α, ¬β
(→ R) ¬(α ∧ β) ⇒ ¬α ∨ ¬β
⇒ ¬(α ∧ β) → (¬α ∨ ¬β)
We look at the ST version of this proof. We proceed bottom-up, prefixing
a wff occurring on the left of the sequent with a ‘−’ sign, and that on the
right, with a ‘+’ sign.
Example 0.2.
+ ¬(α ∧ β) → (¬α ∨ ¬β)
− ¬(α ∧ β)
+ ¬α ∨ ¬β
+ ¬α
+ ¬β
− α
− β
+ α∧β
+ α | + β
Steps are not repeated: for instance, ¬(α ∧ β) occurs only once in the second
step with a prefixed − sign. We observe that arriving at axiom instances
in the SC proof, translates into obtaining both +φ and −φ in a ‘path’ of
the ST tree, for some wff φ, thus ‘closing’ that path. So, if a wff φ has a
SC proof, every path in its ST tree that is completely expanded, should be
closed.

1 Rules for semantic tableaux


Formally, we have two sets of rules in ST, based on which we obtain the
tableau expansion of any wff.

2
1.1 α-rules (the rules for branching of paths)

α α1 α2
+ φ∧ψ +φ +ψ
− φ∨ψ −φ −ψ
−φ→ψ +φ −ψ
+¬φ −φ
−¬φ +φ

1.2 β-rules (the rules for prolongation of paths)

β β1 β2
− φ∧ψ −φ −ψ
+ φ∨ψ +φ +ψ
+φ→ψ −φ +ψ

These rules correspond to the eight logical rules of SC, leaving out Ax and
Cut. Let us now define a tableau, step by step.

Definition 1.1.

• (Paths) The set of paths in a wff φ is the smallest set of sets of signed
wffs such that
(a) {+φ} is a path,
(b) if C is a path and α is a signed wff in C of type α, then
(C \ {α}) ∪ {α1 } and (C \ {α}) ∪ {α2 } are paths, where α1 , α2 are
defined as above, and in case of ¬, the second path is omitted,
(c) if C is a path and β is a signed wff in C of type β, then
(C \ {β}) ∪ {β1 , β2 } is a path, where β1 , β2 are defined as above.

• An expansion strategy f is a function from paths C with at least one


non-atomic signed wff to such a wff.

• A path C is closed, if there is a wff +φ0 in C such that −φ0 is in C as


well.

A tableau is a set of paths. A tableau is closed, if and only if all its fully
expanded paths are closed.

We state without proof, the following fundamental results of ST.

Theorem 1.1. (Soundness, Completeness) Tableaux are sound and


complete for any strategy, i.e. for any strategy, the tableaux expansion
starting from +φ terminates on a closed tableaux, if and only if φ is valid.

3
Theorem 1.2. (Termination) Let φ be any propositional wff. The ex-
pansion of a tableau for +φ terminates, whatever expansion strategy we
choose.

Exercise 1.1. Expand all the wffs given in Exercise 5.1 of Lecture 16 into
tableaux.

Tableaux are easy to implement, and the basic computational steps can
be made quite fast. The space requirement is also not much. On the other
hand, as in all proof methods, the size of a fully expanded tableau is ex-
ponential in the size of the wff being proved. Moreover, a tableau develops
every path independently of others, resulting in duplication of effort. The
subject in which modifications and other implementable proof techniques
are studied, is called Automated Theorem Proving.

Potrebbero piacerti anche