Propositional Logic, continued: Soundness, Completeness, & Special Forms

Soundness

Completeness

Conjuctive Normal Form

Horn Clauses

Slide 1 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Soundness (1)


Suppose Φ1

Φn Ψ holds.

Hence, there is a proof of Ψ having Φ1

Φn as premises.

We proceed by induction on the length of these proofs. We need to reformu-

late the soundness statement such that it is amenable to induction.

M  k  : For all sequents Φ1

Φn Ψ that have a proof of
length k, it is the case that Φ1

Φn
Ψ

We intend to use course-of-values induction on k.

Technical problem:

– Chopping a proof may nod lead to correct sub-proofs, since some boxes
may still be open.
– However, a chopped proof (a prefix of the sequence of formulas rep-
resenting a proof) may form a correct proof if the assumptions of the
open boxes are added to the premises.
Slide 2 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04
Soundness (2)

To solve our technical problem, we change the structure of the proof as

in the following example. Consider the following sequent:

p q
r p
 q
r

1 p q
r premise 1 0/ p q
r premise
2 p assumption 2  2  p assumption
3 q assumption 3  2
3  q assumption
4 p q i 2,3 4  2
3  p q i 2,3
5 r
e 1,4 5  2
3  r
e 1,4
6 q
r
i 3–5 6  2  q
r
i 3–5
7 p
 q
r
i 2–6 7 0/ p
 q
r
i 2–6

Note: The set at the right of a formula in a proof line grows and shrinks
as a stack, reflecting the way boxes are opened and closed.

Slide 3 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Soundness (3)

Just for the purpose of proving soundness, we formally change the defi-
nition of the proof as follows.

Definition: A proof of the sequent Γ Ψ is a sequence of pairs
 d1
χ1 

 dk
χk 
where:
(1) d1
0; /
(2) each di is a subset of 1

i ;
 

(3) for each i, χi is either

– a premise (i.e., χi  Γ), or
– an assumption (i.e. χi  di ), or
– χi follows from previous lines by applying deduction rules;
(4) for each i, di is equal to
– di  1 if no box was closed/opened at line i;
– di  1  i if a box is opened on line i;
– di  1  ρ if a box with assumption at line rho was closed
 

 

The length of such a proof is k.

Slide 4 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Soundness (4)

Our inductive statement now becomes:

For any proof of length k  d1
χ1 

 dk
χk 
,

and any assignment of truth values that makes the
premises in Γ and the assumptions in dk true, it is
the case that χk evaluates to T .

When there is a chopping with no open boxes, this hypothesis precisely

covers the semantic entailment.

Slide 5 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Soundness (5)

We now proceed with the proof.

Base case k
1: the proof has length 1, hence it is of the form

1 0/ χ premise

The statement is obviously true, any assignment of truth values

that makes all the premises true, shall make this premise true as
well.

Induction case k 1: Suppose we have a proof

1 0/ χ1 premise
..
.
k d k χk justi f ication k

Slide 6 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Soundness (6)

justification-k is a natural deduction rule, hence we proceed by case

analysis.
Φ Ψ Φ Ψ Φ Ψ 
i e1 e2  e
Φ Ψ Φ Ψ Φ

Φ Ψ  Φ
Φ Ψ
. .  e
. .
. .
Φ Ψ χ χ Φ

i1
i2
e
Φ
Ψ Φ
Ψ χ

Φ
.
.
.
Ψ Φ Φ Ψ
 i  e
Φ Ψ Ψ

Φ
.
.

.
 Φ
 i  e
 Φ Φ

Slide 7 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Soundness (7)

i: It must be the case that χ

χ1 χ2 , with χi appearing at line ki
k, with i  1
2 . The formulas χ1 , χ2 have shorter proofs, and
 

therefore, using the induction hypothesis, they have the truth value
T . Using the truth table for , we conclude that the truth value of
χ is T .

e: It must be the case that some formula χ1 χ2 appears in in the

proof, and that we have two boxes with assumptions χ1 and χ2 and
conclusion χ. The proof of χ1 χ2 is shorter hence, according to

the induction hypothesis, it has a truth value of T . According to the

truth table of , either χ1 or χ2 have the truth value T . Assume it

is χ2 (the case when χ1 has a truth value of T is similar). Then, the

assumption χ2 of the second box is true, and using the induction
hypothesis, its conclusion χ has the truth value T .

The other cases are similar.

Slide 8 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Completeness (1)

Theorem: Whenever Φ1

Φn

Ψ holds, there exists a natural de-

duction proof for the sequent Φ1

Φn Ψ.

The proof consists of three steps:

Step 1:
Φ1
 Φ2

  Φn
Ψ  

Step 2: Φ1
 Φ2

  Φn
Ψ  

Step 3: Φ1

Φn Ψ

Definition: A formula of propositional logic is a tautology if it is true

for all assignments of truth values to its propositional atoms, i.e. if
Φ.

Slide 9 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Completeness (2)

Step 1:
Φ1
 Φ2
   Φn
Ψ  

This step is easy. Suppose Φ1

Φn

Ψ holds. The implication truth table shows
that the only possiblity for
Φ1  Φ2 
  Φn Ψ    to fail is to have an

assignment of truth values to its atoms that results in all Φ1

Φn having the truth
value T and Ψ having the truth value F; — but this is impossible, as it contradicts
the hypothesis.

Step 3: Φ1

Φn Ψ

This step is also easy. Suppose Φ1

Φ1
 Φ2

  Φn Ψ   holds, i.e.

..
.
has a natural deduction proof Π. Then
Φn
we augment this proof by adding the ..
premises Φ1

Φn to the front, and then Π


.


Φ1 Φ2 Φn Ψ  
using the rule e at the end to produce

Ψ


e
Ψ. In other words, we produce the proof:

Slide 10 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Completeness (3)

The core of the completeness proof is Step 2, which requires to show the
following:

If
Φ holds, then Φ holds. In other words, if
Phi is a tautology, then Φ is a theorem.

The idea of the proof is the following:

Suppose
Φ holds.

If formula Ψ has n atoms p1

pn , then Φ has truth value T for

all the 2n lines in its truth table.

Then, we ”encode” each line in the truth table of Φ as a sequent

and assemble them into a proof of Φ using the disjunction rules.

Slide 11 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Completeness (4)

The first part of the proof is based on the following lemma.

Lemma: Let Φ be a formula containing the propositional atoms

p1

pn , and l a line of Φ’s truth table. Let p̂i be pi if the entry in
line l of pi is T , otherwise p̂i is pi . Then,

p̂1

p̂n Φ is provable if the entry for Φ in line l is T ;

p̂1

p̂n Φ is provable if the entry for Φ in line l is F.

Slide 12 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Completeness (5)

The proof of the lemma is by course of values induction on the height of

the syntax tree of Ψ.

Base case: If Φ is an atom (i.e. a formula of height 1), then we have to

 
show that p p and p p hold. This is immediate.

Slide 13 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Completeness (6)

Induction case: The height of Φ is greater than 1. Then, we have the following cases.

Φ is of the form Φ1 .

– If Φ evaluates to T , then Φ1 evaluates to F; Φ1 has the same atoms as Φ,


but a lower height, hence by induction hypothesis p̂1

p̂n Φ1 ; finally
Φ1 is Φ, hence we are done.
– If Φ evaluates to F, then Φ1 evaluates to T ; by induction hypothesis we
 
get p̂1

p̂n Φ1 , which can be extended to p̂1

p̂n Φ1 using the
i rule; but Φ1 is Φ1 , hence we are done.

Φ is of the form Φ1 Φ2 , where  . Let q1

ql be the atoms of Φ1




and r1

rk the atoms of Φ2 , where q1

ql  r1

rk

   p1

pn .
  

We are left with proving that

 
q̂1

q̂l Ψ1 and r̂1

r̂k Ψ2

implies p̂1

p̂n Ψ1 Ψ2

for appropriate formulas Ψ1 and Ψ2 .

Slide 14 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04
Completeness (7)

We show the proof for

, that is, we consider the case when Φ
Φ1 Φ2 .

If both Φ1 and Φ2 evaluate to T , then by the induction hypothesis

  
q̂1

q̂l Φ1 and r̂1

r̂k Φ2 , hence p̂1

p̂n Φ1 Φ2 , and we are
done.

If Φ1 evaluates to F and Φ2 evaluates to T , then we have q̂1

q̂l Φ1
 
and r̂1

r̂k Φ2 , hence p̂1

p̂n Φ1 Φ2 We are left with proving
 
p̂1

p̂n Φ1 Φ2 implies p̂1

p̂n  Φ1 Φ2 

(left as an exercise)

The other two cases are similar, requiring the following proofs:

Φ1 Φ2  Φ1 Φ2 

Φ1 Φ2  Φ1 Φ2 

Slide 15 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Completeness (8)

If Φ is of the form Φ1 Φ2 , we can reduce the proof to the search for the

following proofs.

Φ1 Φ2 Φ1 Φ2


Φ1 Φ2 Φ1 Φ2


Φ1 Φ2 Φ1 Φ2


Φ1 Φ2  Φ1 Φ2 

If Φ is of the form Φ1
Φ2 , we can reduce the proof to the search for

the following proofs.

Φ1 Φ2 Φ1 Φ2


Φ1 Φ2  Φ1 Φ2 


Φ1 Φ2 Φ1 Φ2


Φ1 Φ2 Φ1 Φ2

Slide 16 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Completeness (9)

The last piece of the puzzle is to assemble these proofs of the form

p̂1

p̂n Φ

each representing a line in the truth table, into a proof of Φ, without
premises.

We use the disjunction rules to generate the lines of the truth table, then
we appropriately insert the above proofs.

We exemplify this procedure for the case of two atoms, for the tautology

p q
p.

Slide 17 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Completeness (10)

Assembling the proof for the tautology p q
p.
1 p p LEM

2 p ass p ass

3 q q LEM q q LEM

4 q ass q ass q ass q ass

.... .... .... ....
5 .. .. .. ..

7 p q
p p q
p p q
p p q
p

8 p q p e p q p e

9 p q p e

Slide 18 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Semantic Equivalence & Validity

Definitions:

Let Φ and Ψ be propositional logic formulas. They are semantically

equivalent iff Φ
Ψ and Ψ
Φ. We denote this by Φ Ψ.

Φ is valid iff
Φ.

Remarks:

Two formulas Φ and Ψ are semantically equivalent iff

 Φ
Ψ
Ψ
Φ .

Because of soundness and completeness of propositional logic, se-


mantic equivalence is identical with provable equivalence Φ

Ψ   Ψ Φ  . (This is a fortunate case, most logics are not complete).

Our aim is to transform formulas into equivalent ones for which check-
ing validity is easier.

Slide 19 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Conjunctive Normal Form

Definitions:

A literal is either an atom p, or the negation of an atom p.

A formula Φ is in conjunctive normal form (CNF) if it is of the

form Ψ1 Ψ2  Ψn , for some n 1, where each Ψi is a dis-
junction of literals, for all i  1

n .  

Note: Sometimes we include the case n

0, in which case, by conven-
tion, the term is .

Examples of CNFs:
 q p r  p r q

 p r  p r  p r

Not in CNF:
  q p r  p r q

Slide 20 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Validity of a Disjunction of Literals

Lemma: A disjunction of literals L1 L2  Ln is valid iff there exist i, j, with

1 i
j n, such that Li is L j .

Proof:

If there exist i, j such that Li is L j , then clearly L1 L2  Ln evaluates

to T for all assignments.

For the converse, if no literal has a matching negation, then:

– For each positive literal we assign F to the corresponding atom.

– For each negative literal we assign T to the corresponding atom.
– This assignment falsifies the disjunction, which is impossiible. (Ex-
ample: for q p r, take p and r to be true, and q to be false and q

to be true.)
– Hence, there exist i, j such that Li is L j .

Slide 21 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Satisfiability

Definition: A formula Ψ is satisfiable if there exists an assignment of

truth values to its propositional atoms such that Φ is true.

Proposition: A propositional logic formula Φ is satisfiable iff Φ is not

valid.

Proof:

If Φ is satisfiable, then there exists a valuation (assignment of truth

values to its atoms) which makes Φ true. For this valuation Φ has
the truth value F, hence Φ cannot be valid.

Conversely, if Φ is not valid, then there exists a valuation for

which Φ has the truth value F. This valuation makes Φ have the
truth value T , hence Φ is satisfiable.

This is a simple, but very useful result.

Slide 22 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Useful Identities (Boolean Algebra)

and are idempotent and are distributive

Φ Φ Φ Φ  Ψ η  Φ Ψ  Φ η

Φ Φ Φ Φ  Ψ η  Φ Ψ  Φ η

and are commutative Rules for T and F

Φ Ψ Ψ Φ F Φ F Φ Φ F
Φ Ψ Ψ Φ T Φ T Φ Φ T

and are associative The de Morgan rules

Φ  Ψ η  Φ Ψ η  Φ Ψ Φ Ψ

Φ  Ψ η  Φ Ψ η  Φ Ψ Φ Ψ

and are absorbative Double negation rules

Φ  Φ η Φ Φ Φ

Φ  Φ η Φ

Slide 23 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

A Procedure to Compute CNFs

We present an algorithm to compute a CNF formula equivalent to a given

arbitrary formula Φ. The algorithm is deterministic and computes a
unique CNF for any formula.
The algorithm is described as:

CNF(NNF(IMPL FREE(Φ)))

for a given formula Φ. The CNF, NNF, and IMPL FREE functions shall
be discussed shortly.

Slide 24 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

IMPL FREE

function IMPL FREE(Φ) :

/* precondition: Φ is an arbitrary formula */
/* postcondition: returns an implication free formula equivalent to Φ */
begin function
case
Φ is a literal: return Φ
Φ is Φ1 : return  IMPL FREE  Φ1 
Φ is Φ1 Φ2 : return IMPL FREE  Φ1  IMPL FREE  Φ2 
Φ is Φ1 Φ2 : return IMPL FREE  Φ1  IMPL FREE  Φ2 

Φ is Φ1 Φ2 : return IMPL FREE  Φ1  IMPL FREE  Φ2 

end case
end function

Slide 25 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

IMPL FREE Example

Let Φ
p q
p  r
q .

IMPL FREE  Φ 
IMPL FREE  p q  IMPL FREE  p  r q 

 IMPL FREE  p  IMPL FREE  q  IMPL FREE  p  r q 

 p  IMPL FREE  q  IMPL FREE  p  r q 

 p  q  IMPL FREE  p  r q 

 p  q   p  IMPL FREE  r  IMPL FREE  q 

 p  q   p  r IMPL FREE  q 

 p  q   p  r q 

Slide 26 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

NNF

function NNF(Φ) :
/* precondition: Φ is implication free */
/* postcondition: returns an NNF formula equivalent to Φ */
begin function
case
Φ is a literal: return Φ
Φ is Φ1 : return NNF  Φ1 
Φ is Φ1 Φ2 : return NNF  Φ1  NNF  Φ2 
Φ is Φ1 Φ2 : return NNF  Φ1  NNF  Φ2 

Φ is  Φ1 Φ2  : return NNF  Φ1 Φ2 

Φ is  Φ1 Φ2  : return NNF  Φ1 Φ2 

end case
end function

Slide 27 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

NNF Example

Let Φ
 p q  p  r q  .

NNF  Φ 
NNF   p  q  NNF  p  r q 

NNF   p  q  NNF  p  r q 

 NNF  p   NNF  q  NNF  p  r q 

p  NNF  q  NNF  p  r q 

p q NNF  p  r q 

p q  NNF  p  NNF  r q 

p q  p NNF  r q 

p q  p  NNF  r  NNF  q 

p q  p  r NNF  q 

p q  p  r q 

Slide 28 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

CNF

function CNF(Φ) :
/* precondition: Φ is implication and in NNF */
/* postcondition: returns an CNF formula equivalent to Φ */
begin function
case
Φ is a literal: return Φ
Φ is Φ1 Φ2 : return CNF  Φ1  CNF  Φ2 
Φ is Φ1 Φ2 : return DISTR  CNF  Φ1 
CNF  Φ2 

end case
end function

Slide 29 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

DISTR

function DISTR(Φ1 ,Φ2 ) :

/* precondition: Φ1 , Φ2 are in CNF */
/* postcondition: returns an CNF formula equivalent to Φ1 Φ2 */

begin function
case
Φ1 is Φ11 Φ12 : return DISTR  Φ11
Φ2  DISTR  Φ12
Φ2 
Φ2 is Φ21 Φ22 : return DISTR  Φ1
Φ21  DISTR  Φ1
Φ22 
otherwise: return Φ1 Φ2

end case
end function

Slide 30 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

CNF Example

Let Φ
p q  p  r q  .

CNF  Φ 
CNF  p q  p  r q 

DISTR  CNF  p q 
CNF  p  r q 

DISTR  p q
CNF  p  r q 

DISTR  p q
p  r q 

DISTR  p q
p  DISTR  p q
r q

 p q p  p q r q

Slide 31 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Horn Clauses

Definitions:

A Horn clause is a formula of the form p1 p2  pk
q,

where k 1, and p1
p2

pk
q are atoms, , or
.

A Horn formula is a conjunction of Horn clauses, i.e. a formula

Φ of the form Ψ1 Ψ2  Ψn , (n 1), such that each Ψi is a
Horn clause, i  1

n .
 

Horn clauses have an efficient procedure to decide their satisfiability, and

are the basis for logic programming.

Slide 32 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

Horn Clauses Examples

Examples (yes)

 p q s p  q r


p  p s
s
 p q s
 q r
p 

s
 p1 p3 p5 p13  



p5   p5 p11


Examples (no)

 p q s p  q r


p  p s
s
 p q s   q r


p 

s
 p1 p3 p5 p13 p27 


p5   p5 p11


Slide 33 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04

HORN
function HORN(Φ) :
/* precondition: Φ is a Horn formula*/
/* postcondition: decides the satisfiability of Φ*/
begin function
if Φ contains a clause

then return unsatisfiable
else mark all atoms p where

p is a clause of Φ
while there is a Horn clause p1  pki
qi of Φ
such that all p j are marked, but qi isn’t do
if qi then return ’unsatisfiable’
else mark qi for all Horn clauses of Φ
end while
return ’satisfiable’
end function

Theorem: The HORN algorithm is correct: it always terminates and its

answer is ’satisfiable’ iff the given Horn formula is satisfiable.

Slide 34 CS4231 — Logic and Formal Systems — Lecture 02 — 19/08/04