Satisfiability Checking

Propositional Logic

Prof. Dr. Erika Ábrahám

RWTH Aachen University

Informatik 2
LuFG Theory of Hybrid Systems

WS 14/15

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 1/1
Propositional logic

The slides are partly taken from:

www.decision-procedures.org/slides/

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 2/1
Propositional logic - Outline

Syntax of propositional logic

Semantics of propositional logic
Satisfiability and validity
Modeling with propositional logic
Normal forms
Enumeration and deduction

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 3/1
Propositional logic - Outline

Syntax of propositional logic

Semantics of propositional logic
Satisfiability and validity
Modeling with propositional logic
Normal forms
Enumeration and deduction

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 4/1
Syntax of propositional logic
Abstract syntax of well-formed propositional formulae:

ϕ := a | (¬ϕ) | (ϕ ∧ ϕ)

where Prop is a set of (atomic) propositions (Boolean variables) and

a ∈ Prop.
We write PropFormulae for the set of all propositional logic formulae.

Syntactic sugar:

⊥ := (a ∧ ¬a)
> := (a ∨ ¬a)
( ϕ1 ∨ ϕ2 ) := ¬((¬ϕ1 ) ∧ (¬ϕ2 ))
( ϕ1 → ϕ2 ) := ((¬ϕ1 ) ∨ ϕ2 )
( ϕ1 ↔ ϕ2
L ) := ((ϕ1 → ϕ2 ) ∧ (ϕ2 → ϕ1 ))
( ϕ1 ϕ2 ) := (ϕ1 ↔ (¬ϕ2 ))

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 5/1
Formulae

Examples of well-formed formulae:

(¬a)
(¬(¬a))
(a ∧ (b ∧ c))
(a → (b → c))
We omit parentheses whenever we may restore them through operator
precedence:

binds stronger

¬ ∧ ∨ → ↔

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 6/1
Propositional logic - Outline

Syntax of propositional logic

Semantics of propositional logic
Satisfiability and validity
Modeling with propositional logic
Normal forms
Enumeration and deduction

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 7/1
Semantics: Assignments

Structures for predicate logic:

The domain is B = {0, 1}.
The interpretation assigns Boolean values to the variables:
α : Prop → {0, 1}
We call these special interpretations assignments and use V to denote the
set of all assignments.
Example: Prop = {a, b}, α(a) = 0, α(b) = 1

Equivalently, we can see an assignment α as a set of variables (α ∈ 2Prop ),

defining the variables from the set to be true and the others false.
Example: Prop = {a, b}, α = {b}

An assignment can also be seen as being of type α ∈ {0, 1}Prop , if we have

an order on the propositions.
Example: Prop = {a, b}, α = 01

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 8/1
Only the projected assignment matters...

Let α1 , α2 ∈ V and ϕ ∈ PropFormulae.

Let AP(ϕ) be the atomic propositions in ϕ.
Clearly AP(ϕ) ⊆ Prop.
Lemma: if α1 |AP(ϕ) = α2 |AP(ϕ) , then

Projection

(α1 satisfies ϕ) iff (α2 satisfies ϕ)

We will assume, for simplicity, that Prop = AP(ϕ).

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 9/1
Semantics I: Truth tables

Truth tables define the semantics (=meaning) of the operators.

They can be used to define the semantics of formulae inductively over
their structure.
Convention: 0= false, 1= true
L
p q ¬p p ∧ q p ∨ q p → q p ↔ q p q
0 0 1 0 0 1 1 0
0 1 1 0 1 1 0 1
1 0 0 0 1 0 0 1
1 1 0 1 1 1 1 0
Each possible assignment is covered by a line of the truth table.
α satisfies ϕ iff in the line for α and the column for ϕ the entry is 1.

Q: How many binary operators can we define that have different semantics?
A: 16

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 10 / 1
Semantics I: Example

Let ϕ be defined as (a ∨ (b → c)).

Let α : {a, b, c} → {0, 1} be an assignment with
α(a) = 0, α(b) = 0, and α(c) = 1.

Q: Does α satisfy ϕ?
A1: Compute with truth table:
a b c b → c a ∨ (b → c)
0 0 0 1 1
0 0 1 1 1
0 1 0 0 0
0 1 1 1 1
1 0 0 1 1
1 0 1 1 1
1 1 0 0 1
1 1 1 1 1
Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 11 / 1
Semantics II: Satisfaction relation

Satisfaction relation: |= ⊆ V × PropFormulae

Instead of (α, ϕ) ∈ |= we write α |= ϕ and say that
α satisfies ϕ or
ϕ holds for α or
α is a model of ϕ.
|= is defined recursively:

α |= p iff α(p) = true

α |= ¬ϕ iff α 6 |= ϕ
α |= ϕ1 ∧ ϕ2 iff α |= ϕ1 and α |= ϕ2
α |= ϕ1 ∨ ϕ2 iff α |= ϕ1 or α |= ϕ2
α |= ϕ1 → ϕ2 iff α |= ϕ1 implies α |= ϕ2
α |= ϕ1 ↔ ϕ2 iff α |= ϕ1 iff α |= ϕ2

Note: More elegant but semantically equivalent to truth tables.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 12 / 1
Semantics II: Example

Let ϕ be defined as (a ∨ (b → c)).

Let α : {a, b, c} → {0, 1} be an assignment with
α(a) = 0, α(b) = 0, and α(c) = 1.

Q: Does α satisfy ϕ?

A2: Compute with the satisfaction relation:

α |= (a ∨ (b → c))
iff α |= a or α |= (b → c)
iff α |= a or (α |= b implies α |= c)
iff 0 or (0 implies 1)
iff 0 or 1
iff 1

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 13 / 1
Semantics III: The algorithmic view

Using the satisfaction relation we can define an algorithm for the

problem to decide whether an assignment α : AP → {0, 1} is a
model of a propositional logic formula ϕ ∈ PropFormulae:

Eval (α, ϕ) {
i f ϕ ≡ a r e t u r n α(a) ;
i f ϕ ≡ (¬ϕ1 ) r e t u r n not Eval (α, ϕ1 ) ;
i f ϕ ≡ (ϕ1 op ϕ2 )
r e t u r n Eval (α, ϕ1 ) [[op]] Eval (α, ϕ2 ) ;
}

Equivalent to the |= relation, but from the algorithmic view.

Q: Complexity? A: Polynomial (time and space).

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 14 / 1
Semantics III: Example

Recall our example

ϕ = (a ∨ (b → c))
α : {a, b, c} → {0, 1} with α(a) = 0, α(b) = 0, and α(c) = 1.

Eval (α, ϕ) = Eval (α, a) or Eval (α, b → c) =

0 or (Eval(α, b) implies Eval(α, c)) =
0 or (0 implies 1) =
0 or 1 =
1

Hence, α |= ϕ.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 15 / 1
Satisfying assignments
Intuition: each formula specifies a set of assignments satisfying it.
Remember: V denotes the set of all assignments.
Function sat : PropFormulae → 2V
(a formula → set of its satisfying assignments)
Recursive definition:
sat(a) = {α | α(a) = 1}, a ∈ Prop
sat(¬ϕ1 ) = V \ sat(ϕ1 )
sat(ϕ1 ∧ ϕ2 ) = sat(ϕ1 ) ∩ sat(ϕ2 )
sat(ϕ1 ∨ ϕ2 ) = sat(ϕ1 ) ∪ sat(ϕ2 )
sat(ϕ1 → ϕ2 ) = (V \ sat(ϕ1 )) ∪ sat(ϕ2 )

For ϕ ∈ PropFormulae and α ∈ V it holds that

α |= ϕ iff α ∈ sat(ϕ)

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 16 / 1
Satisfying assignments: Example

sat(a ∨ (b → c)) =
sat(a) ∪ sat(b → c) =
sat(a) ∪ ((V \ sat(b)) ∪ sat(c)) =
{α ∈ V | α(a) = 1}∪
{α ∈ V | α(b) = 0}∪
{α ∈ V | α(c) = 1} =
{α ∈ V | α(a) = 1 or α(b) = 0 or α(c) = 1}

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 17 / 1
Extensions of |=

We define |= ⊆ 2V × PropFormulae by

T |= ϕ iff T ⊆ sat(ϕ)

for formulae ϕ ∈ PropFormulae and assignment sets T ⊆ 2V .

Examples: {α ∈ V | α(a) = α(c) = 1} |= a ∨ (b → c)
{α ∈ V | α(x1 ) = 1} |= x1 ∨ x2

We define |= ⊆ 2PropFormulae × 2PropFormulae by

ϕ1 |= ϕ2 iff sat(ϕ1 ) ⊆ sat(ϕ2 )

for formulae ϕ1 , ϕ2 ∈ PropFormulae.

Examples: a ∧ c |= a ∨ (b → c)
x1 |= x1 ∨ x2

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 18 / 1
Short summary for propositional logic
Syntax of propositional formulae ϕ ∈ PropFormulae:
ϕ := prop | (¬ϕ) | (ϕ ∧ ϕ)

Semantics:
Assignments α ∈ V :
α : Prop → {0, 1}
α ∈ 2Prop
α ∈ {0, 1}Prop

Satisfaction relation:
|= ⊆ V × PropFormulae , (e.g., α |=ϕ )
|= ⊆ 2V × PropFormulae , (e.g., {α1 , . . . , αn }|=ϕ )
|= ⊆ PropFormulae × PropFormulae, (e.g., ϕ1 |=ϕ2 )
sat : PropFormulae → 2V , (e.g., sat(ϕ) )

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 19 / 1
Propositional logic - Outline

Syntax of propositional logic

Semantics of propositional logic
Satisfiability and validity
Modeling with propositional logic
Normal forms
Enumeration and deduction

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 20 / 1
Semantic classification of formulae

A formula ϕ is called valid if sat(ϕ) = V .

(Also called a tautology).

A formula ϕ is called satisfiable if sat(ϕ) 6= ∅.

A formula ϕ is called unsatisfiable if sat(ϕ) = ∅.

(Also called a contradiction).

satisfiable unsatisfiable

valid

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 21 / 1
Some notations

We can write:

|= ϕ when ϕ is valid

6|= ϕ when ϕ is not valid

6|= ¬ϕ when ϕ is satisfiable

|= ¬ϕ when ϕ is unsatisfiable

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 22 / 1
Examples

(x1 ∧ x2 ) → (x1 ∨ x2 ) is valid

(x1 ∨ x2 ) → x1 is satisfiable
(x1 ∧ x2 ) ∧ ¬x1 is unsatisfiable

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 23 / 1
Examples

Here are some valid formulae:

|= a ∧ 1 ↔ a
|= a ∧ 0 ↔ 0
|= ¬¬a ↔ a (double-negation rule)
|= a ∧ (b ∨ c) ↔ (a ∧ b) ∨ (a ∧ c)

Some more (De Morgan rules):

|= ¬(a ∧ b) ↔ (¬a ∨ ¬b)
|= ¬(a ∨ b) ↔ (¬a ∧ ¬b)

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 24 / 1
The satisfiability problem for propositional logic

The satisfiability problem for propositional logic is as follows:

Given an input propositional formula ϕ, decide whether ϕ is satisfiable.
This problem is decidable but NP-complete.
An algorithm that always terminates for each propositional logic
formula with the correct answer is called a decision procedure for
propositional logic.
Goal: Design and implement such a decision procedure:

no
Satisfiability
Is ϕ satisfiable? checker
yes

Note: A formula ϕ is valid iff ¬ϕ is unsatisfiable.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 25 / 1
Propositional logic - Outline

Syntax of propositional logic

Semantics of propositional logic
Satisfiability and validity
Modeling with propositional logic
Normal forms
Enumeration and deduction

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 26 / 1
Before we solve this problem...

Suppose we can solve the satisfiability problem... how can this help us?

There are numerous problems in the industry that are solved via the
satisfiability problem of propositional logic
Logistics
Planning
Electronic Design Automation industry
Cryptography
...

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 27 / 1
Example 1: Placement of wedding guests

Three chairs in a row: 1, 2, 3

We need to place Aunt, Sister and Father.
Constraints:
Aunt doesn’t want to sit near Father
Aunt doesn’t want to sit in the left chair
Sister doesn’t want to sit to the right of Father

Q: Can we satisfy these constraints?

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 28 / 1
Example 1 (continued)
Notation: Aunt = 1, Sister = 2, Father = 3
Left chair = 1, Middle chair = 2, Right chair = 3
Introduce a propositional variable for each pair (person, chair):
xp,c = “person p is sited in chair c” for 1 ≤ p, c ≤ 3
Constraints:
Aunt doesn’t want to sit near Father:

((x1,1 ∨ x1,3 ) → ¬x3,2 ) ∧ (x1,2 → (¬x3,1 ∧ ¬x3,3 ))

Aunt doesn’t want to sit in the left chair:

¬x1,1

Sister doesn’t want to sit to the right of Father:

(x3,1 → ¬x2,2 ) ∧ (x3,2 → ¬x2,3 )

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 29 / 1
Example 1 (continued)
Each person is placed:
(x1,1 ∨ x1,2 ∨ x1,3 ) ∧ (x2,1 ∨ x2,2 ∨ x2,3 ) ∧ (x3,1 ∨ x3,2 ∨ x3,3 )

3 _
^ 3
xp,c
p=1 c=1

No person is placed in more than one chair:

3 ^
^ 3 3
^
(¬xp,c1 ∨ ¬xp,c2 )
p=1 c1=1 c2=c1+1

At most one person per chair:

3
^ 3
^ 3
^
(¬xp1,c ∨ ¬xp2,c )
p1=1 p2=p1+1 c=1

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 30 / 1
Example 2: Assignment of frequencies

n radio stations
For each station assign one of k transmission frequencies, k < n.
E – set of pairs of stations, that are too close to have the same
frequency.

Q: Can we assign to each station a frequency, such that no station

pairs from E have the same frequency?

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 31 / 1
Example 2 (continued)
Notation:
xs,f = “station s is assigned frequency f ” for 1 ≤ s ≤ n, 1 ≤ f ≤ k
Constraints:
Every station is assigned at least one frequency:
n k
!
^ _
xs,f
s=1 f =1
Every station is assigned at most one frequency:
n k−1
^ ^ k
^

¬xs,f 1 ∨ ¬xs,f 2
s=1 f 1=1 f 2=f 1+1

Close stations are not assigned the same frequency:

For each (s1, s2) ∈ E ,
k
^

¬xs1,f ∨ ¬xs2,f
f =1
Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 32 / 1
Example 3: Seminar topic assignment

n participants
n topics
Set of preferences E ⊆ {1, . . . , n} × {1, . . . , n}
(p, t) ∈ E means: participant p would take topic t

Q: Can we assign to each participant a topic which he/she is willing to

take?

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 33 / 1
Example 3 (continued)
Notation: xp,t = “participant p is assigned topic t”
Constraints:
Each participant is assigned at least one topic:
n n
!
^ _
xp,t
p=1 t=1

Each participant is assigned at most one topic:

n n−1
^ ^ n
^
(¬xp,t1 ∨ ¬xp,t2 )
p=1 t1=1 t2=t1+1

Each participant is willing to take his/her assigned topic:

n
^ ^
¬xp,t
p=1 (p,t)∈E
/

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 34 / 1
Example 3 (continued)

Each topic is assigned to at most one participant:

n ^
^ n n
^
(¬xp1,t ∨ ¬xp2,t )
t=1 p1=1 p2=p1+1

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 35 / 1
Propositional logic - Outline

Syntax of propositional logic

Semantics of propositional logic
Satisfiability and validity
Modeling with propositional logic
Normal forms
Enumeration and deduction

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 36 / 1
Definitions

Definition: A literal is either a variable or a negation of a variable.

Example: ϕ = ¬(a ∨ ¬b)
Variables: AP(ϕ) = {a, b}
Literals: lit(ϕ) = {a, ¬b}
Note: Equivalent formulae can have different literals.
Example: ϕ0 = ¬a ∧ b
Literals: lit(ϕ0 ) = {¬a, b}

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 37 / 1
Definitions

Definition: a term is a conjunction of literals

Example: (a ∧ ¬b ∧ c)

Definition: a clause is a disjunction of literals

Example: (a ∨ ¬b ∨ c)

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 38 / 1
Negation Normal Form (NNF)

Definition: A formula is in Negation Normal Form (NNF) iff

(1) it contains only ¬, ∧ and ∨ as connectives and
(2) only variables are negated.

Examples:
ϕ1 = ¬(a ∨ ¬b) is not in NNF
ϕ2 = ¬a ∧ b is in NNF

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 39 / 1
Converting to NNF

Every formula can be converted to NNF in linear time:

Eliminate all connectives other than ∧, ∨, ¬
Use De Morgan and double-negation rules to push negations to
operands

Example: ϕ = ¬(a → ¬b)

Eliminate ’→ 0 : ϕ = ¬(¬a ∨ ¬b)
Push negation using De Morgan: ϕ = (¬¬a ∧ ¬¬b)
Use double-negation rule: ϕ = (a ∧ b)

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 40 / 1
Disjunctive Normal Form (DNF)

Definition: A formula is said to be in Disjunctive Normal Form (DNF)

iff it is a disjunction of terms.
In other words, it is a formula of the form
 
_ ^
 li,j 
i j

where li,j is the j-th literal in the i-th term.

Example:

ϕ = (a ∧ ¬b ∧ c) ∨ (¬a ∧ d ) ∨ (b) is in DNF

DNF is a special case of NNF.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 41 / 1
Converting to DNF

Every formula can be converted to DNF in exponential time and

space:
1 Convert to NNF
2 Distribute disjunctions following the rule:
|= ϕ1 ∧ (ϕ2 ∨ ϕ3 ) ↔ (ϕ1 ∧ ϕ2 ) ∨ (ϕ1 ∧ ϕ3 )
Example:
ϕ = (a ∨ b) ∧ (¬c ∨ d )
= ((a ∨ b) ∧ (¬c)) ∨ ((a ∨ b) ∧ d )
= (a ∧ ¬c) ∨ (b ∧ ¬c) ∨ (a ∧ d ) ∨ (b ∧ d )

Now consider ϕn = (a1 ∨ b1 ) ∧ (a2 ∨ b2 ) ∧ . . . ∧ (an ∨ bn ).

Q: How many clauses will the DNF have?
A: 2n

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 42 / 1
Satisfiability of DNF

Q: Is the following DNF formula satisfiable?

(a1 ∧ a2 ∧ ¬a1 ) ∨ (a2 ∧ a1 ) ∨ (a2 ∧ ¬a3 ∧ a3 )

A: Yes, because the term a2 ∧ a1 is satisfiable.

Q: What is the complexity of the satisfiability check of DNF formulae?
A: Linear (time and space).
Q: Can there be any polynomial transformation into DNF?
A: No, it would violate the NP-completeness of the problem.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 43 / 1
Conjunctive Normal Form (CNF)

Definition: A formula is said to be in Conjunctive Normal Form (CNF)

iff it is a conjunction of clauses.
In other words, it is a formula of the form
 
^ _
 li,j 
i j

where li,j is the j-th literal in the i-th clause.

Example:

ϕ = (a ∨ ¬b ∨ c) ∧ (¬a ∨ d ) ∧ (b) is in CNF

Also CNF is a special case of NNF.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 44 / 1
Converting to CNF

Every formula can be converted to CNF in exponential time and

space:
1 Convert to NNF
2 Distribute disjunctions following the rule:
|= ϕ1 ∨ (ϕ2 ∧ ϕ3 ) ↔ (ϕ1 ∨ ϕ2 ) ∧ (ϕ1 ∨ ϕ3 )
Consider the formula ϕ = (a1 ∧ b1 ) ∨ (a2 ∧ b2 ).
Transformation: (a1 ∨ a2 ) ∧ (a1 ∨ b2 ) ∧ (b1 ∨ a2 ) ∧ (b1 ∨ b2 )
Now consider ϕn = (a1 ∧ b1 ) ∨ (a2 ∧ b2 ) ∨ . . . ∨ (an ∧ bn ).
Q: How many clauses does the resulting CNF have?
A: 2n

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 45 / 1
Converting to CNF: Tseitin’s encoding

Every formula can be converted to CNF in linear time and space if

new variables are added.
The original and the converted formulae are not equivalent but
equi-satisfiable.

Consider the formula Parse tree:

ϕ = (a → (b ∧ c))
→ h1
Associate a new auxiliary variable with each
gate.
Add constraints that define these new variables. a ∧ h2
Finally, enforce the root node.

b c

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 46 / 1
Converting to CNF: Tseitin’s encoding

Need to satisfy: → h1
(h1 ↔ (a → h2 ))∧
(h2 ↔ (b ∧ c))∧
a ∧ h2
(h1 )

b c

Each gate encoding has a CNF representation with 3 or 4 clauses.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 47 / 1
Converting to CNF: Tseitin’s encoding

Need to satisfy:
(h1 ↔ (a → h2 )) ∧ (h2 ↔ (b ∧ c)) ∧ (h1 )

First: (h1 ∨ a) ∧ (h1 ∨ ¬h2 ) ∧ (¬h1 ∨ ¬a ∨ h2 )

Second: (¬h2 ∨ b) ∧ (¬h2 ∨ c) ∧ (h2 ∨ ¬b ∨ ¬c)

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 48 / 1
Converting to CNF: Tseitin’s encoding

Let’s go back to
ϕn = (x1 ∧ y1 ) ∨ (x2 ∧ y2 ) ∨ · · · ∨ (xn ∧ yn )

With Tseitin’s encoding we need:

n auxiliary variables h1 , . . . , hn .
Each adds 3 constraints.
Top clause: (h1 ∨ · · · ∨ hn )

Hence, we have
3n + 1 clauses, instead of 2n .
3n variables rather than 2n.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 49 / 1
Propositional logic - Outline

Syntax of propositional logic

Semantics of propositional logic
Satisfiability and validity
Modeling with propositional logic
Normal forms
Enumeration and deduction

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 50 / 1
Two classes of algorithms for validity

Q: Is ϕ satisfiable? (Is ¬ϕ valid?)

Complexity: NP-Complete (Cook’s theorem)
Two classes of algorithms for finding out:
Enumeration of possible solutions (Truth tables etc.)
Deduction

More generally (beyond propositional logic):

Enumeration is possible only in some logics.
Deduction cannot necessarily be fully automated.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 51 / 1
The satisfiability problem

Given a formula ϕ, is ϕ satisfiable?

Enumeration the first:
B o o l e a n SAT( ϕ ) {
result := f a l s e ;
for a l l α ∈ V
result = result ∨ Eval(α, ϕ) ;
r e t u r n result ;
}
Enumeration the second:
Use substitution to eliminate all variables one by one:

ϕ iff ϕ[0/a] ∨ ϕ[1/a]

Q: What is the difference?

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 52 / 1
Deduction requires axioms and inference rules

Inference rules:
Antecedents
(rule name)
Consequents

Meaning: If all antecedents hold then at least one of the consequents

can be derived.
Examples:

a→b b→c
(Trans)
a→c

a→b a
(M.P.)
b

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 53 / 1
Axioms

Axioms are inference rules with no antecedents, e.g.,

(H1 )
a → (b → a)

A proof system consists of a set of axioms and inference rules.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 54 / 1
Proofs

Let H be a proof system.

Γ `H ϕ means: There is a proof of ϕ in system H whose premises are
included in Γ
`H is called the provability relation.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 55 / 1
Example

Let H be the proof system comprised of the rules Trans and M.P. that
we saw earlier:
a→b b→c
(Trans)
a→c

a→b a
(M.P.)
b

Does the following relation hold?

a → b, b → c, c → d , d → e, a `H e

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 56 / 1
Deductive proof: Example

a→b b→c a→b a

(Trans) (M.P.)
a→c b

a → b, b → c, c → d , d → e, a `H e

1. a→b premise
2. b→c premise
3. a→c 1, 2, Trans
4. c →d premise
5. d →e premise
6. c →e 4, 5, Trans
7. a→e 3, 6, Trans
8. a premise
9. e 7, 8, M.P.
Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 57 / 1
Proof graph

a→b b→c c →d d →e
trans trans
a→c c →e
trans
a→e a
M.P.
e

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 58 / 1
Soundness and completeness

For a given proof system H,

Soundness: Does ` conclude “correct” conclusions from premises?
Completeness: Can we conclude all true statements with H?
Correct with respect to what?
With respect to the semantic definition of the logic. In the case of
propositional logic truth tables give us this.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 59 / 1
Soundness and completeness

Let H be a proof system

Soundness of H : if `H ϕ then |= ϕ
Completeness of H : if |= ϕ then `H ϕ

How to prove soundness and completeness?

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 60 / 1
Example: Hilbert axiom system (H)

Let H be (M.P.) together with the following axiom schemes:

(H1)
a → (b → a)

(H2)
((a → (b → c)) → ((a → b) → (a → c)))

(H3)
(¬b → ¬a) → (a → b)

H is sound and complete for propositional logic.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 61 / 1
Soundness and completeness

To prove soundness of H, prove the soundness of its axioms and

inference rules (easy with truth-tables).
For example:
a b a → (b → a)
0 0 1
0 1 1
1 0 1
1 1 1
Completeness: harder, but possible.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 62 / 1
The resolution inference system

The resolution inference rule for CNF:

(l ∨ l1 ∨ l2 ∨ ... ∨ ln ) (¬l ∨ l10 ∨ ... ∨ lm
0 )
Resolution
(l1 ∨ ... ∨ ln ∨ l10 ∨ ... ∨ lm
0 )

Example:

(a ∨ b) (¬a ∨ c)
(b ∨ c)

We first see some example proofs, before proving soundness and

completeness.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 63 / 1
Proof by resolution

Let ϕ = (a1 ∨ a3 ) ∧ (¬a1 ∨ a2 ∨ a5 ) ∧ (¬a1 ∨ a4 ) ∧ (¬a1 ∨ ¬a4 )

We want to prove ϕ → (a3 )

(¬a1 ∨ a4 ) (¬a1 ∨ ¬a4 )

(a1 ∨ a3 ) (¬a1 )

(a3 )

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 64 / 1
Resolution

Resolution is a sound and complete inference system for CNF.

If the input formula is unsatisfiable, there exists a proof of the empty
clause.

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 65 / 1
Example

Let ϕ = (a1 ∨ a3 ) ∧ (¬a1 ∨ a2 ) ∧ (¬a1 ∨ a4 ) ∧ (¬a1 ∨ ¬a4 ) ∧ (¬a3 ).

(¬a1 ∨ a4 ) (¬a1 ∨ ¬a4 )

(¬a1 ) (a1 ∨ a3 )

a3 ¬a3

()

Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 66 / 1
Soundness and completeness of resolution
Soundness is straightforward. Just prove by truth table that
|= ((ϕ1 ∨ a) ∧ (ϕ2 ∨ ¬a)) → (ϕ1 ∨ ϕ2 ).

Completeness is a bit more involved.

Basic idea: Use resolution for variable elimination.
(a ∨ ϕ1 ) ∧ . . . ∧ (a ∨ ϕn )∧
(¬a ∨ ψ1 ) ∧ . . . (¬a ∨ ψm )∧
R
⇔
(ϕ1 ∨ ψ1 ) ∧ . . . ∧ (ϕ1 ∨ ψm )∧
...
(ϕn ∨ ψ1 ) ∧ . . . (ϕn ∨ ψm )∧
R
where ϕi (i = 1, . . . , n), ψj (j = 1, . . . , m), and R contains neither a
nor ¬a.
Satisfiability Checking — Prof. Dr. Erika Ábrahám (RWTH Aachen University) WS 14/15 67 / 1