It is the new clause produced by a resolution step.

Does the empty clause mean there are no rules?

No. It is an unsatisfiable clause, distinct from an empty collection of clauses.

Can resolution handle variables?

First-order resolution can, using unification and appropriate variable handling in addition to the clause rule.

AI Glossary / Graphs and symbolic AI / Intermediate

Resolution

Definition

Resolution is a logical inference rule that combines clauses containing complementary literals to derive a new clause. In propositional logic, resolving P or A with not-P or B yields A or B. Resolution is especially important for proof by contradiction: a system adds the negation of a claim and attempts to derive the empty clause, indicating inconsistency.

What it means

A literal is an atomic statement or its negation. A clause is a disjunction, meaning that at least one of its literals must hold. Resolution removes one chosen complementary pair from two parent clauses and combines what remains.

The rule does not mean casually deleting contradictory statements from a document. Its correctness depends on the logical form of the premises and the precise inference step. Here resolution means automated reasoning, not image pixel dimensions, resolving an entity name, or settling a customer support request. Those other uses need their own contextual definitions.

How it works

Translate the relevant statements into clauses, commonly using conjunctive normal form. Select two clauses with a complementary pair, produce their resolvent, and retain it as a consequence of the premises. Repeat according to a proof-search strategy.

To establish that a knowledge base entails a query, include the query's negation and seek a contradiction. The empty clause has no literal that could make it true, so deriving it establishes unsatisfiability of the combined clauses.

First-order resolution additionally handles variables through unification and must preserve their scopes, typically standardizing clauses apart. Appropriate factoring and fair search are relevant to completeness. Proof search may be expensive; unrestricted first-order reasoning does not guarantee termination when no proof exists.

Three practical examples

Rule consistency checking: detect that a formal rule set and asserted facts cannot all hold together, providing a derivation for review.
Automated theorem proving: establish a logical consequence by showing that its denial conflicts with the premises.
Configuration analysis: reason over encoded compatibility requirements to explain why a requested combination cannot satisfy the formal constraints.

Think of it this way

Think of combining “either the spare is available or delivery is delayed” with “the spare is not available.” The second statement leaves delay as the consequence of the first. This captures a simple resolution step, but general clauses can contain several alternatives. The method is formal inference from premises, not evidence that the premises accurately describe reality.

Worked example

Let S mean “a spare is available,” D mean “delivery is delayed,” and N mean “the customer is notified.” Consider clauses C1 = S or D, C2 = not-S, and C3 = not-D or N. These are invented workflow statements for a logic exercise.

Resolve C1 and C2 on S. Removing S and not-S leaves D. Resolve D with C3 on D, producing N. The derivation establishes N from the three clauses.

For a refutation proof of N, add C4 = not-N. Resolve the derived N with not-N to obtain the empty clause. No truth assignment can satisfy all four clauses simultaneously, so the original three entail N.

Notice that C1 alone does not entail D: setting S true and D false satisfies it. The premise not-S is essential to the first inference.

Also consider clauses P or Q and not-P or not-Q. Resolving on P produces Q or not-Q, a tautology. It is invalid to remove both complementary pairs at once and announce the empty clause. The original pair is satisfiable, for example with P true and Q false. A valid binary resolution step chooses one pivot pair and preserves the remaining literals.

Illustrative case study

Fictional case: Spruce Configuration School encodes a demonstration product's feature rules as clauses. A requested feature set produces a contradiction, and the first interface merely displays “invalid.”

Students add a proof trace showing the parent clauses and selected pivot for each derived clause. Reviewers can now see whether the conflict comes from the request, an intended compatibility rule, or an incorrectly entered rule. They also test a satisfiable pair containing two complementary pairs to ensure the engine does not cancel both at once.

This case illustrates the explanatory value of a checked derivation. It does not claim that resolution alone produces the smallest possible explanation or identifies which business rule should be changed.

Common misconceptions and mistakes

Resolution is sound, but a sound rule applied to inaccurate premises can still support an inappropriate real-world conclusion. Logical validity and factual accuracy require separate attention.

Failure to find a contradiction before a timeout is not proof that the clauses are consistent. A complete exhausted propositional procedure has stronger implications than an interrupted search.

Another mistake is ignoring inconsistency in the original knowledge base. Under classical logic, an inconsistent set entails every statement. A proof of a query is much more useful when the premises have also been checked for relevant inconsistencies.

Business use

Resolution can support formal validation and auditable reasoning where requirements can be represented precisely. Teams should retain the mapping from clauses back to the source rules and distinguish formal contradictions from data-entry errors or misunderstood requirements.

Ask whether a tool returns a proof, a satisfying assignment, or an unresolved result. These outputs mean different things. A proof trace should be independently checkable, while operational decisions about changing a rule remain a separate responsibility. Large applications often rely on specialized reasoning systems rather than manually enumerating every possible resolvent.

Check your understanding

From A or B and not-A, what follows by one resolution step?

Answer: B. The complementary pair A and not-A is removed, leaving B. The inference depends on both parent clauses.

Related terms

Learn next

Practice several propositional resolution steps before attempting first-order examples. For each, label the parent clauses and pivot, then check the result against a truth table. Next, follow a short refutation proof and distinguish deriving a useful clause from deriving the empty clause.

Sources

Russell and Norvig, Logical Agents slides: https://aima.cs.berkeley.edu/slides-pdf/chapter07.pdf University of Rochester, Propositional Logic and Resolution: https://www.cs.rochester.edu/users/faculty/nelson/courses/csc_173/proplogic/resolution.html Russell and Norvig, First-Order Inference slides: https://aima.cs.berkeley.edu/slides-pdf/chapter09.pdf Primary teaching material was read. Workflow clauses, counterexample, and fictional case are original.

Previous: Resistance Next: Resolution closure