Resolution Proof System for First Order Logic

Download Report

Transcript Resolution Proof System for First Order Logic

Resolution Proof System
for First Order Logic
Michael Kucera
0426131
Resolution for propositional logic

Ground (propositional) resolution is a single sound
inference rule that is applied to clauses.
A clause is a disjunction of literals.

Example, (p  ¬q  r) is a clause.

Resolution:

Example,

(α1 p), (α2 ¬p)
(α1 α2)
(¬p  q), (¬q r)
(¬p r)
Clausal form

In order to use resolution for first order logic the
formulas must be converted into clausal form.
A clause is a disjunction of literals with no literal
appearing twice, no existential quantifiers and all
universal quantifiers are at the left.

Ex,




xyz( P(x,y)  ¬Q(y) R(z,y) )
The empty clause, denoted by , is unsatisfiable.
A unit clause contains just one literal.
The universal quantifiers may be omitted.
Skolemization


Process for removing existential quantifiers.
Delete each existential quantifier, then
replace the resulting free variables by terms
referred to as Skolem functions.

xy P(x,y) becomes y P(c, y)
xy P(x,y) becomes x P(x, f(x))

Skolemization preserves satisfiability.

Conversion to clausal form






Eliminate and
Move ¬ inwards
Rename variables if two quantifiers have the
same bound variable name.
Eliminate existential quantifiers using Skolem
functions.
Move all universal quantifiers to the left (prenex
normal form).
Transform the matrix into conjunctive normal
form.
Conversion to clausal form

This produces the form
x1x2 ...xm ( C1  ...  Cn )

Where C1 ... Cn are clauses.

Each step preserves logical equivalence
except Skolemization which only preserves
satisfiability.
Unification

A substitution, denoted by θ, is an assignment from
variables to terms.

Example:
if θ is {x → g(y)} and E is P(x,w,f(x))
then Eθ is P(g(y),w,f(g(y)))

Unification is the process of replacing the variables in
expressions by terms to make the modified expressions
identical to each other.

Example:
Let E1=P(x,a) , E2=P(y,a) , θ = {y→x}
Then E1θ = E2θ = P(x,a)

I the above example θ is the most general unifier (mgu)
Resolution for FOL
(α1 L)θ, (α2 ¬L)θ
(α1  2)θ




θ is the most general unifier for L.
The result of resolution is called the resolvent.
There may be more than one possible
reslolvent.
The resolvent of two complementary unit
clauses is the empty clause.
Resolution for first order logic

Resolution is refutationally complete (with factoring).

Factoring is an additional rule that allows you to remove
duplicate literals within a clause.
Proving  ‫ =׀‬A is equivalent to proving that
  {¬A} is unsatisfiable.


A is called the goal.

In a resolution proof you negate the goal and add it to
the set of hypotheses, then apply resolution until the
empty clause is produced.
Example
Hypothesies
Clausal Form
x (dog(x) animal(x))
dog(fido)
y (animal(y)  die(y))
¬dog(x)  animal(x)
dog(fido)
¬animal(y)  die(y)
Conclusion
die(fido)
Negate the goal
¬die(fido)
Example
¬dog(x)  animal(x)
{x
dog(fido)
¬animal(y)  die(y)
{x → y}
¬dog(y)  die(y)
{y → fido}
¬die(fido)
die(fido)

Automated theorem proving





Resolution is used in automated theorem
provers like Otter.
Don't have to choose which rule of inference
to use because there is only one.
No axiom schemata to instantiate.
Many refinements such as Hyper-resolution
and Lock resolution.
Search strategies remove redundancy of
preforming all posible deduction sequences.