Is unit resolution the same as unit preference?

No. Unit resolution is an inference form; unit preference is a strategy favoring such inferences.

How does it relate to unit propagation?

Propagation commonly implements the same forced consequences through assignments and clause simplification.

Can the derived clause have several literals?

Yes. A unit parent removes one complementary alternative from a longer clause.

AI Glossary / Graphs and symbolic AI / Advanced

Unit resolution

Definition

Unit resolution is a resolution inference in which at least one parent clause contains exactly one literal. The unit literal is matched with its complement in the other clause, and the remaining literals form the conclusion. This can shorten a clause or, with opposite units, produce the empty clause representing contradiction.

What it means

If P is true and ¬P∨Q must also be true, Q must be true. Unit resolution expresses that reasoning as an inference from the clauses {P} and {¬P,Q} to {Q}.
The conclusion need not be a unit. Resolving P with ¬P∨Q∨R yields Q∨R, which still has two alternatives. “Unit” describes a parent of the inference, not a guarantee about the conclusion's size. The rule provides a sound step; selecting and repeating such steps is a separate proof-search procedure.

How it works

In propositional logic, the general form is: from the unit L and the clause ¬L∨C, infer C. Here L can itself be positive or negative, and C denotes the disjunction of the remaining literals. If there are no remaining literals, the result is the empty clause.
In first-order reasoning, complementary atoms may need unification. For example, Licensed(ada) and ¬Licensed(x)∨MayApply(x) yield MayApply(ada) after substituting ada for x throughout the second clause. That step does not establish that everyone may apply.
Unit resolution should be distinguished from unrestricted resolution, which permits two nonunit parents. A procedure restricted to unit resolution is not refutation-complete for arbitrary clause sets: some inconsistent inputs have no units from which it can begin. For finite propositional Horn clause sets, however, unit resolution is sufficient to derive a contradiction whenever the set is unsatisfiable. That special result does not make it a complete general-purpose inference method for every requested conclusion.

Three practical examples

A workflow verifier combines the fact Reviewed with the rule ¬Reviewed∨Eligible to infer Eligible. This is a direct unit-resolution step once the rule is represented as a clause.
A configuration checker combines a fixed disabled option with a clause listing alternatives. Removing the impossible alternative may leave another option required or a shorter unresolved choice.
A theorem prover adds the negation of a proposed conclusion to its premises. A chain of unit resolutions producing the empty clause can establish that the original premises entail the conclusion.

Think of it this way

Imagine crossing an impossible alternative off a requirement because another established fact rules it out. The remaining alternatives must carry the requirement. If the requirement has no alternative left, the current collection of statements cannot all hold together.

Worked example

Let L mean licensed, T trained, A authorized, and E eligible. Consider these five clauses for one fictional participant:
C1=L; C2=T; C3=¬L∨¬T∨A; C4=¬A∨E; C5=¬E.
The first two assert facts. C3 says that being licensed and trained is sufficient for authorization. C4 says authorization implies eligibility. C5 denies eligibility.
First resolve C1 with C3 on L. The result is ¬T∨A, a two-literal clause. Next resolve that result with C2 on T to derive A. Resolve A with C4 to derive E. Finally, resolve E with C5 to derive the empty clause.
Each of these four inferences has a unit parent. The original five clauses are therefore inconsistent. A direct check of all sixteen assignments to L, T, A, and E finds no satisfying assignment.
Remove C5 and the remaining clauses have the unique satisfying assignment L=T=A=E=true. In that setting, the same first three inferences show that eligibility follows. Adding ¬E is a way to test this conclusion by contradiction.
Instead, keep C5 but remove the fact T. The remaining clauses have the satisfying assignment L=true, T=false, A=false, E=false. Missing the training premise changes what follows. The example therefore depends on its stated inputs, not merely on the presence of an authorization rule.

Illustrative case study

A fictional training provider uses symbolic rules to check whether a participant qualifies for a workshop. One system reports that the participant is licensed and trained; another reports ineligibility. A unit-resolution trace connects the positive facts to eligibility and then exposes their conflict with the negative report.
The reviewer can inspect four short inferences rather than an unexplained error message. However, the proof does not identify which real-world record is wrong. The training fact could be outdated, the rule could be incomplete, or the ineligibility report could refer to a different workshop.
The team checks the meaning and scope of each proposition before changing records. This case illustrates how a formal proof supports review while leaving factual verification and policy interpretation to the responsible people. It does not report an actual provider's process or results.

Common misconceptions and mistakes

Unit resolution does not require both parents to be units. One is enough, and the other may contain several literals.
It does not allow removing unrelated literals or canceling two literals with the same sign. The selected pair must be complementary, with consistent substitutions when variables are involved.
A failed unit-resolution search does not generally prove consistency. The search may need nonunit inferences or another complete method.
The empty clause is a contradiction, not a statement that no more useful information exists. An empty collection of clauses has a different meaning: it places no remaining requirements.

Business use

Unit-resolution traces can make deductions and conflicts in rule-based systems easier to audit. They are useful when a decision must be explained through specific premises and rules. Keep the original clauses, substitutions where relevant, and parent references for each inference. Verify the inputs separately: a logically correct proof can still encode an outdated rule or an incorrectly recorded fact.

Check your understanding

Answer: What is the first derived clause in the worked example? ¬T∨A.

Why is the final step a contradiction?

Answer: E and ¬E cannot both be true.

Why does removing T matter?

Answer: The licensing-and-training rule no longer forces authorization from the remaining positive fact alone.

Related terms

Learn next

Practice identifying the complementary literals and recording every substitution. Then compare a unit refutation with a proof that requires an initial inference between two nonunit clauses.

Sources

Stanford, Introduction to Logic, Resolution and its strategies: https://logic.stanford.edu/intrologic/chapters/chapter_14.html Stanford, Introduction to Logic, relational resolution and unification: https://logical.stanford.edu/intrologic/sections/section_14.html?section=4 AIMA glossary, unit resolution: https://github.com/aimacode/aima-glossary

Previous: Unit propagation Next: Unit vector