AI Glossary / Graphs and symbolic AI / Intermediate
Resolvent
Definition
A resolvent is the clause produced by applying a resolution inference step to two parent clauses. In propositional logic, the step removes one selected pair of complementary literals and combines the remaining literals. The resolvent is a logical consequence of the two parents, so it can be used in later reasoning without introducing an unsupported assumption.
What it means
Resolution is the operation; a resolvent is its result. If one clause says P or Q and another says not-P or R, resolving on P yields Q or R. The shared subject of the cancellation is called the pivot.
The result is not necessarily a single fact, a contradiction, or a stronger version of both parents. It may still contain alternatives, and replacing the parents with their resolvent generally loses information. Here the term belongs to logic and automated theorem proving; mathematical uses of resolvent in other fields have different definitions.
How it works
Select two clauses and identify a literal in one whose complement appears in the other. Remove that selected pair, combine the remaining literals, and remove repeated copies of identical literals. Record the parents and pivot so the step can be checked.
If nothing remains, the result is the empty clause, which is false under every interpretation. If both a literal and its complement remain, the resolvent is a tautology and usually offers no useful constraint.
In first-order logic, complementary atoms may first need a unifying substitution. Apply that substitution consistently to the remaining literals, not just the selected atoms. Variables from independent parent clauses must be handled with their proper scopes.
Three practical examples
- Proof explanations: show the exact intermediate clause that follows from two earlier clauses, allowing a reader to inspect a long proof one step at a time.
- Rule checking: derive a consequence of interacting requirements and compare it with an asserted fact to expose a contradiction.
- Theorem-prover testing: use small parent clauses with known resolvents to test pivot selection, duplicate removal, and substitution handling.
Think of it this way
Think of combining two conditional clues until one unresolved alternative disappears, leaving the alternatives that still must hold. The analogy explains why a result can remain a disjunction. Formal resolution is stricter than everyday reasoning: the clauses, complementary pair, and any substitutions must satisfy exact syntactic requirements.
Worked example
Let C1 be P or Q or R, and let C2 be not-P or Q or S. Resolve on P. Removing P from C1 and not-P from C2 leaves Q or R or Q or S. Since repeating Q does not change a disjunction's meaning, the resolvent is Q or R or S.
Why must the result hold when both parents hold? If P is false, C1 requires Q or R. If P is true, C2 requires Q or S. Either way, at least one of Q, R, and S is true.
The reverse implication does not hold. Set P, Q, and R false and S true. The resolvent is true, but C1 is false. Keeping only the resolvent would therefore discard a requirement imposed by its parents.
For a first-order example, take not-Ready(x) or CanStart(x), together with Ready(Device7). Unifying the selected atoms gives x = Device7. The resulting resolvent is CanStart(Device7), not CanStart(x) with x still unrestricted.
Finally, resolving the unit clauses T and not-T leaves no literal, producing the empty clause. This is a contradiction. Resolving T or U with not-T or not-U on T instead produces U or not-U, which is always true. These two outcomes are logically opposite, despite both involving complementary literals.
Illustrative case study
Fictional case: Elm Proof Studio designs an educational proof viewer. An early version shows only the new clause and hides the substitutions used to produce it. Students cannot tell why a general-looking variable has become a particular device name.
The team adds the two parent clauses, the selected pivot, and a substitution annotation to each proof step. A reviewer can now check whether the substitution was applied to all remaining literals. Tautological results are marked as such, while the empty clause receives a distinct explanation.
The case illustrates how a small amount of provenance makes an intermediate result understandable. It does not claim that displaying a trace automatically verifies the reasoning engine or the truth of its inputs.
Common misconceptions and mistakes
Do not cancel every complementary pair between two clauses simultaneously. A binary resolution step uses one selected pivot pair; removing additional pairs can produce an invalid conclusion.
A resolvent need not be informative. It may already exist, be subsumed by another clause, or be a tautology. Such outcomes affect proof-search efficiency without making the valid inference unsound.
Also avoid treating a derived clause as a replacement for its parents unless a separate transformation justifies that replacement. Entailment from parents to resolvent does not generally establish equivalence.
Business use
Resolvents matter to organizations mainly through the reliability and explainability of formal reasoning tools. A proof-producing system can expose intermediate consequences of requirements instead of offering only an unexplained yes/no answer.
For review, preserve parent identifiers and transformations back to the original rules. A logically valid resolvent may reveal that a rule was entered incorrectly or that the desired requirement was mistranslated. Those are modeling questions rather than failures of the inference step. Keeping the distinction clear helps teams correct the right part of a system.
People also ask
Can two clauses have more than one possible resolvent?
Yes. Different eligible pivot choices can produce different resolvents, although some may be redundant or tautological.
Is every resolvent a contradiction?
No. Only the empty clause represents contradiction; many resolvents are ordinary disjunctions.
Does a resolvent contain new factual evidence?
No. It is derived from its parents. Its real-world reliability still depends on how those premises were established.
Check your understanding
Resolve A or B with not-A or B using A as the pivot. What is the resolvent?
Answer: B. The remaining disjunction is B or B, which simplifies to B. This duplicate removal does not introduce an extra assumption.
Related terms
- Resolution
- Clause
- Literal
- Unification
- pivot
- substitution
- empty clause
- tautology
- proof trace
Learn next
Write parent clauses beside each proposed resolvent and identify exactly what was removed. Use truth assignments to check that every assignment satisfying both parents also satisfies the result. Then move to a first-order example and explicitly apply the unifier to every surviving literal.
More in Graphs and symbolic AI
- Repeated state - A repeated state is a state that a search procedure encounters more than once, usually through different paths or by returning through a cycle.
- Representational frame problem - The representational frame problem is the difficulty of compactly describing what remains unchanged when an action occurs.
- Resolution closure - The resolution closure of a clause set is the collection containing its starting clauses and every clause obtainable by repeatedly applying resolution to...
- Resource constraint - A resource constraint is a restriction imposed by the limited availability of something required to perform an activity.
- Rete - Rete is an algorithm for efficiently matching many rule conditions against a changing collection of facts.
- Revelation principle - The revelation principle is a mechanism-design result showing that an outcome achievable through an appropriate equilibrium of a mechanism can also be...
Sources
University of Rochester, Propositional Logic and Resolution: https://www.cs.rochester.edu/users/faculty/nelson/courses/csc_173/proplogic/resolution.html Russell and Norvig, Inference in First-Order Logic: https://aima.cs.berkeley.edu/slides-pdf/chapter09.pdf Primary teaching material was read. Parent clauses, truth assignments, device example, and fictional case are original.
Previous: Resolution closure Next: Resource constraint