CS 228 (M) - Logic in CS

Tutorial III - Solutions

Ashwin Abraham

IIT Bombay

24th August, 2023

Ashwin Abraham 2023 1 / 12

Table of Contents

1 Question 1

2 Question 2

3 Question 3

4 Question 4

5 Question 5

Ashwin Abraham 2023 2 / 12

Question 1

This statement is False. An easy counterexample to this would be

F = {p, ¬p} and G = {q, ¬q}.

Ashwin Abraham 2023 3 / 12

Question 2
Theorem
A set of formulae Σ is satisfiable iff every finite subset of it is satisfiable.
This theorem is known as the Compactness Theorem.
Proof.
Proving the backward direction is trivial, as clearly if Σ is satisfiable then
every finite subset of Σ is satisfiable (indeed, every subset is satisfiable).
Let us show that if Σ is not satisfiable, then there exists a finite subset of
it that is unsatisfiable (this suffices to show the forward direction). By the
Completenessa of our Formal Proof System, if Σ is unsatisfiable, then it is
inconsistent, ie Σ ` ⊥. The proof of this statement can use only a finite
number of formulae in Σ (since all proofs are finite). Call this finite subset
Σ0 . Our proof of Σ ` ⊥ will also show that Σ0 ` ⊥, and so this Σ0 is a
finite subset of Σ that is unsatisfiable.
a
For this proof to be airtight, our proof of completeness should not depend
on the Compactness Theorem, even in the infinite case. Such proofs do exist.

Ashwin Abraham 2023 4 / 12

Question 2

Since F is inconsistent (and therefore also unsatisfiable), by the

Compactness Theorem there exists a finite subset of F (say F 0 ) that is
unsatisfiable (and therefore
 inconsistent). Since F is closed under
f ∈ F. Call this F . Clearly {F } ≡ F 0 , and therefore
V
conjunction
f ∈F 0
{F } ` ⊥. By ⊥ elimination, for any formula G , we have {F } ` ¬G .
Therefore, we have shown that there exists F ∈ F such that for any
G ∈ F, {F } ` ¬G . This is in fact a stronger statement than what we set
out to prove!

Ashwin Abraham 2023 5 / 12

Question 3

We have to show that if F is not a contradiction and G is not a tautology,

and  (F =⇒ G ), then there exists a formula H such that  (F =⇒ H),
 (H =⇒ G ) and Vars(H) ⊆ Vars(F ) ∩ Vars(G ).
Firstly, note that we do not need the statement that F is not a
contradiction and G is not a tautology. If F is a contradiction, then we
can take H = ⊥ and if G is a tautology we can take H = >.
Removing this clause from the question statement, we shall prove the rest
via induction on |Vars(F ) − Vars(G )|. Our inductive hypothesis will be if
|Vars(F ) − Vars(G )| = k and  (F =⇒ G ), then there exists H such that
 (F =⇒ H),  (H =⇒ G ) and Vars(H) ⊆ Vars(F ) ∩ Vars(G ).
Base Case:
When k = 0, we have Vars(F ) ⊆ Vars(G ), and therefore we can choose
H = F , which satisfies all the conditions.

Ashwin Abraham 2023 6 / 12

Question 3

Before we proceed to the inductive step,

Lemma:
Say q ∈ Vars(F ) − Vars(G ) and  (F =⇒ G ).
Let H = F [q/⊥] ∨ F [q/>]. Then we have  (F =⇒ H) and
 (H =⇒ G ).
Note that for any formula F , F [p/G ] denotes the formula obtained by
replacing all instances of p in F by G .
Proof:
Say an assignment α has α  F . If α(q) = 0, then we have α  F [q/⊥]
and therefore α  H. On the other hand, if α(q) = 1, then α  F [q/>]
and we still have α  H. Therefore, we have α  F =⇒ α  H for all α,
ie F =⇒ H is valid, ie  (F =⇒ H).
Now, let us show the other part. Some notation first: For an assignment
α, α[q → b] is an assignment identical to α except at q, where it is b. We
have α[q → 0]  F ⇐⇒ α  F [q/⊥], α[q → 1]  F ⇐⇒ α  F [q/>].

Ashwin Abraham 2023 7 / 12

Question 3

Assume α  H. We have:
1 α  F [q/⊥] ∨ F [q/>]
2 α[q → 0]  F ∨ α[q → 1]  F
3 α[q → 0]  G ∨ α[q → 1]  G (Since ∀α, α  F =⇒ α  G )
Now, since q ∈
/ Vars(G ), α[q → b]  G ⇐⇒ α  G , b ∈ {0, 1}.
Therefore,
4 αG ∨αG
5 αG
Therefore, ∀α, α  H =⇒ α  G , ie  (H =⇒ G )

Ashwin Abraham 2023 8 / 12

Question 3

Now, back to the main proof.

Inductive Step:
Our inductive hypothesis is that for any formulae F and G if
|Vars(F ) − Vars(G )| = k and  (F =⇒ G ), then there exists H such that
 (F =⇒ H),  (H =⇒ G ), and Vars(H) ⊆ Vars(F ) ∩ Vars(G ).
Assuming this, we have to prove the hypothesis for the case where
|Vars(F ) − Vars(G )| = k + 1. Let q ∈ Vars(F ) − Vars(G ), and let
H = F [q/>] ∨ F [q/⊥]. By the previous lemma, we have  (F =⇒ H)
and  (H =⇒ G ). Note that |Vars(H) − Vars(G )| = k. Applying the
inductive hypothesis, there exists H 0 such that  (H =⇒ H 0 ),
 (H 0 =⇒ G ) and Vars(H 0 ) ⊆ Vars(H) ∩ Vars(G ). Using  (F =⇒ H)
and the fact that Vars(H) ⊆ Vars(F ), we get  (F =⇒ H 0 ),
 (H 0 =⇒ G ), and Vars(H 0 ) ⊆ Vars(F ) ∩ Vars(G ). Therefore, the
inductive hypothesis is proven for k + 1, and thus the statement in the
question is also proven.

Ashwin Abraham 2023 9 / 12

Question 4

Firstly, note that the empty set ∅ is satisfiable (in fact, it is valid)1 .
Now, it can be easily shown that the set
n
_
Σn = {p1 , . . . pn , ¬pi }
i=1

is an example of a minimal unsatisfiable set for n ≥ 1.

1
This is because all universally quantified propositions over the empty set are true -
these are known as vacuous truths.
Ashwin Abraham 2023 10 / 12
Question 5
(a) Mechanically keep calculating Res n (ψ) by resolution, until you find
that ∅ ∈ Res ∗ (ψ) = Res 3 (ψ). This correctly tells us that ψ is unsatisfiable
due to the soundness of the resolution proof system.
(b) Let us do resolution in a slightly different way.
Our algorithm is as follows:
1 If Vars(ψ) is empty, then we can immediately conclude the

satisfiability of ψ by checking if ∅ ∈ ψ.
2 If not, pick a variable p ∈ Vars(ψ) such that resolution2 can be done

with pairs of clauses in ψ with p as pivot.

3 If no such variable exists, then we are done with resolution, and we

can check satisfiability by checking if ∅ ∈ ψ.

4 If such a variable exists, replace ψ with R (ψ), where R (ψ) is formed
p p
by removing all clauses that were involved in resolution from ψ and
replacing them with the newly generated resolved clauses.
5 Go to step 1
2
We do not consider resolutions that lead to tautologies
Ashwin Abraham 2023 11 / 12
Question 5
To show that this algorithm works, we show that ψ and Rp (ψ) are
equisatisfiable, ie ψ ` ⊥ ⇐⇒ Rp (ψ) ` ⊥.
The reverse direction is easy to prove here, the clauses of R(ψ) are either
members of ψ or are formed from ψ by resolution, ie any proof that
Rp (ψ) ` ⊥ can easily be converted into a proof that ψ ` ⊥ by replacing
the steps assuming the resolved clauses with their resolutions.
For the forward direction, let us prove the contrapositive, ie Rp (ψ) is
satisfiable =⇒ ψ is satisfiable.
Let ψ = {{p} ∪ Ai : i ∈ {1 . . . m}} ∪ {{¬p} ∪ Bj : j ∈ {1 . . . n}} ∪ C
where Ai , Bj and C do not contain p.
We have Rp (ψ) = {Ai ∪ Bj : (i, j) ∈ [m] × [n], Ai ∪ Bj not a tautology } ∪ C
Let’s say some assignment α has α  Rp (ψ). Firstly, clearly α  C . If
α  Ai for all i ∈ [m], then α[p → 0]  ψ. If there is some k ∈ [m] such
that α 2 Ak , then for all j ∈ [n], we have α  Ak ∪ Bj (this follows from
the membership of the clause in Rp (ψ) for non-tautological clauses and by
definition for the tautologies). Since α 2 Ak , we must have α  Bj , for all
j ∈ [n]. Therefore, α[p → 1]  ψ. Therefore, ψ is satisfiable.
Ashwin Abraham 2023 12 / 12