Critical pair (term rewriting)

A critical pair arises in a term rewriting system when two rewrite rules overlap to yield two different terms.

In more detail, (t1, t2) is a critical pair if there is a term t for which two different applications of a rewrite rule (either the same rule applied differently, or two different rules) yield the terms t1 and t2.

Specifically, for a pair of overlapping rules

[1] When both sides of the critical pair can reduce to the same term, the critical pair is called convergent.

Where one side of the critical pair is identical to the other, the critical pair is called trivial.

For example, in the term rewriting system with rules the only critical pair is ⟨g(x,z), f(x,z)⟩.

As another example, consider the term rewriting system with the single rule By applying this rule in two different ways to the term f(f(x,x),x), we see that (f(x,x), f(x,x)) is a (trivial) critical pair.

Confluence clearly implies convergent critical pairs: if any critical pair ⟨a, b⟩ arises, then a and b have a common reduct and thus the critical pair is convergent.

If the term rewriting system is not confluent, the critical pair may not converge, so critical pairs are potential sources where confluence will fail.

The critical pair lemma states that a term rewriting system is weakly (a.k.a.

locally) confluent if and only if all critical pairs are convergent.

Thus, to find out if a term rewriting system is weakly confluent, it suffices to test all critical pairs and see if they are convergent.

This makes it possible to find out algorithmically if a term rewriting system is weakly confluent or not, given that one can algorithmically check if two terms converge.

Triangle diagram of a critical pair obtained from two rewrite rules s → t (upper row, left) and l → r (right). The substitution σ unifies the subterm s | p with l . The resulting overlay term s σ[ lσ ] p (lower row, middle) can be rewritten to the term tσ and s σ[ rσ' ] p (lower row, left and right), respectively. The latter two terms form the critical pair. They can be eventually rewritten to a common term, if the rewrite rule set is confluent . (For notation details, see Term (logic) § Operations with terms .)