[Paper Review] Implementing Anti-Unification Modulo Equational Theory

Jochen Burghardt, Birgit Heinz|arXiv (Cornell University)|Apr 1, 2014

Logic, programming, and type systems9 references3 citations

TL;DR

This paper presents an efficient implementation of E-anti-unification using regular tree grammars to compute generalizations modulo equational theories, enabling compact, finite representations of all possible generalizations. The approach supports applications in lemma generation, intelligence tests, and theory formation by efficiently enumerating minimal generalizations while eliminating redundancies.

ABSTRACT

We present an implementation of E-anti-unification as defined in Heinz (1995), where tree-grammar descriptions of equivalence classes of terms are used to compute generalizations modulo equational theories. We discuss several improvements, including an efficient implementation of variable-restricted E-anti-unification from Heinz (1995), and give some runtime figures about them. We present applications in various areas, including lemma generation in equational inductive proofs, intelligence tests, diverging Knuth-Bendix completion, strengthening of induction hypotheses, and theory formation about finite algebras.

Motivation & Objective

• To develop a compact, finite representation of all E-generalizations modulo equational theories, avoiding infinite enumeration.
• To implement variable-restricted E-anti-unification for improved efficiency and practical applicability.
• To enable automated discovery of general patterns in equational systems, such as inductive proof lemmas and number sequence laws.
• To support theory formation in finite algebras by computing closed representations of quantifier-free, variable-bounded theorems.
• To provide runtime-optimized algorithms with empirical performance measurements for real-world use.

Proposed method

• Represent equivalence classes of terms modulo an equational theory E using regular tree grammars (sorts), enabling finite representation of generalizations.
• Compute the syntactic generalization of two sorts via a tree grammar that captures all terms unifiable modulo E.
• Use anti-narrowing to enumerate generalizations, with optimizations to avoid redundant or useless candidates.
• Apply variable restriction to limit generalization scope, improving efficiency and relevance.
• Implement a PROLOG-based system with performance measurements on finite algebras, including $\mathbb{N}\text{ mod }2$, $\mathbb{N}\text{ mod }3$, and Boolean logic.
• Integrate redundancy removal and minimal term enumeration to produce concise, usable generalizations.

Experimental results

Research questions

• RQ1 Can E-anti-unification be implemented efficiently using regular tree grammars to represent infinite sets of generalizations compactly?
• RQ2 How can variable-restricted E-anti-unification improve the relevance and efficiency of generalization in equational theories?
• RQ3 What is the performance overhead of computing generalizations modulo equational theories in finite algebras, and can it be optimized?
• RQ4 Can this method generate meaningful generalizations for applications like lemma generation in inductive proofs or number sequence completion?
• RQ5 Is it feasible to compute a closed representation of all valid quantifier-free, variable-bounded theorems in finite algebras using this approach?

Key findings

• The implementation successfully computes E-generalizations using regular tree grammars, achieving compact and finite representations of infinite sets of generalizations.
• For $\mathbb{N}\text{ mod }2$ with 3 variables, the system required 3892 seconds and produced 105 non-redundant equations, demonstrating scalability to moderate problems.
• In $\mathbb{N}\text{ mod }3$ with Boolean operators, the system took 48 seconds and generated 307 non-redundant equations, showing efficiency on richer theories.
• The method effectively computes $x*x$ as a generalization of square numbers $0, 1, 4, 9$, even though syntactic anti-unification fails to capture this pattern.
• The system detected and removed redundant instances, reducing output size significantly—e.g., from potentially infinite sets to 307 non-redundant equations in one case.
• The approach enables theory formation in finite algebras by computing all valid quantifier-free, variable-bounded formulas via a closed representation.