Resolution Theorem Proving
Last Updated : 06 Feb, 2024
Prerequisite:
AI | Proofs, and Inferences in Proving Propositional Theorem
Wumpus World in Artificial Intelligence
In this article, we will discuss the inference algorithms that use inference rules. Iterative deepening
search is a full search algorithm in the sense that it will locate any achievable goal. Nevertheless, if the
available inference rules are insufficient, the goal is not reachable — no proof exists that employs just
those inference rules. The proof in the preceding section, for example, would fail if the biconditional
elimination rule was eliminated. The present part introduces resolution, a single inference rule that, when
combined with any full search algorithm, gives a complete inference method.
In the wumpus universe, we start with a simplified version of the resolution rule. Take a look at the steps
that led up to the figure above – the agent travels from [2,1] to [1,1], then to [1,2], where it smells a stink
but notices no breeze. The following information has been added to the knowledge base:
R11 : ¬B1,2 .
R12 : B1,2 ⇔ (P1,1 ∨ P2,2 ∨ P1,3 )
We can now infer the lack of pits in [2,2] and [1,3] (remember that [1,1] is already known to be pitless)
using the same approach that leads to R10 earlier:
R13 : ¬P2,2
R14 : ¬P1,3
To acquire the fact that there is a pit in [1,1], [2,2], or [3,1], we may use biconditional elimination on R3,
followed by Modus Ponens on R5, as follows:
R15 :
P1,1 ∨ P2,2 ∨ P3,1
The resolution rule is now applied for the first time: the literal P2,2 in R13 resolves with the literal P2,2 in
R15, yielding the resolvent.
R16 :
P1,1 ∨ P3,1
If a pit exists in one of [1,1], [2,2], or [3,1], and it is not in [2,2], it is in [1,1] or [3,1]. Similarly, the literal
P1,1 in R1 is resolved by P3,1 when compared to the literal P1,1 in R16 to R17.
R17 :
P3,1
In English, if a pit exists in either [1,1] or [3,1], and it is not in [1,1], it is in [3,1]. The unit resolution
inference rule, l1 lk, m, is used in these last two inference stages.
ℓ1 ∨⋯∨ℓk , m
ℓ1 ∨⋯∨ℓi−1 ∨ℓi+1 ∨⋯∨ℓk
where each l is a literal and l and m are complimentary literals (In other words, negation).
As a result, the unit resolution rule creates a new clause from a clause (a disjunction of literals) and a
literal. A single literal, often known as a unit clause, can be understood as a disjunction of one literal.
The full resolution rule can be generalized to
ℓ1 ∨⋯∨ℓk , m1 ∨⋯∨mn
ℓ1 ∨⋯∨ℓi−1 ∨ℓi+1 ∨⋯∨ℓk ∨m1 ∨⋯∨mj−1 ∨mj+1 ∨⋯∨mn
where li and mj are complementary literals. This means that when two clauses are resolved, a new
clause is created that contains all of the literal from the two original clauses save the two complimentary
literals.
P1,1 ∨P3,1 , ¬P1,1 ∨¬P2,2
P3,1 ∨¬P2,2
There is one additional technical feature to the resolution rule: each literal should only appear once in the
resultant clause. Factoring is the process of removing numerous copies of literal. For example, resolving
(A ∨ B) with (A ∨ ¬B) yields (A ∨ A) , which may be simplified to just A.
Consider the literal li , which is complementary to the literal mj in the other sentence, to determine how
sound the resolution rule is. If li is true, then mj is false, and so m1 ∨ ⋯ ∨ mj−1 ∨ mj+1 ∨ ⋯ ∨ mn must be true,
since m1 ∨ ⋯ ∨ mn is supplied. If li is false, then ℓ1 ∨ ⋯ ∨ ℓk must be true because l1lk is supplied. Now,
because it can be true or false, one of these conclusions must be true—exactly as the resolution rule
requires.
The resolution rule is even more startling because it is the foundation for a family of full inference
methods. A resolution-based theorem proving can determine if α ⊨ β in propositional logic for any
statement α and β . The following two subsections describe how resolution does this.
