On ground word problem of term equation systems

We give semi-decision procedures for the ground word problem of variable preserving term equation systems and term equation systems. They are natural improvements of two well known trivial semi-decision procedures. We show the correctness of our procedures.

💡 Research Summary

The paper addresses the ground word problem for term equation systems (TES), which asks whether two ground terms t₁ and t₂ belong to the same equivalence class generated by a finite set of equations E. Two categories are considered: variable‑preserving TES (VP‑TES), where each equation l ≈ r contains exactly the same set of variables on both sides, and general TES without this restriction.

Historically, two trivial semi‑decision procedures have been used. The first enumerates all possible substitutions for the variables, applies the equations forward, and checks whether t₁ can be transformed into t₂. The second works backward, repeatedly applying the equations in reverse to reduce t₂ toward t₁. Both methods are conceptually simple but suffer from combinatorial explosion: the substitution space grows factorially, and the backward reduction often revisits the same intermediate terms many times.

The authors propose natural improvements to both procedures, yielding two new semi‑decision algorithms: one tailored for VP‑TES and one for arbitrary TES. The key ideas are:

1. Equivalence‑class based substitution pruning – Because VP‑TES guarantees that the variable set is preserved, any substitution σ can be grouped with other substitutions that are identical up to a renaming of variables. The algorithm computes a finite set of “canonical” representatives (called homomorphic classes) and restricts the search to these representatives only. This eliminates redundant branches that would otherwise be explored in the naïve enumeration.

2. Partial normalization – Instead of fully normalizing a term after each rewrite step, the algorithm performs a localized normalization on the sub‑tree where the rewrite rule is applied. By extracting common sub‑structures and normal forms in advance, the algorithm can discard irrelevant parts of the term immediately, dramatically reducing the size of intermediate terms.

3. Heuristic rule selection – At each node of the proof tree the algorithm scores applicable rules according to three criteria: (a) frequency of matching the left‑hand side with the current term, (b) the number of admissible substitutions after class pruning, and (c) the distance to already discovered normal forms. The rule with the highest score is chosen, which keeps the branching factor low.

The paper proves that the improved procedures are complete: if t₁ and t₂ are equivalent under E, the algorithm will eventually find a proof. Completeness follows from two lemmas. Lemma 1 shows that every substitution belongs to some canonical class, so restricting to representatives does not lose any derivations. Lemma 2 demonstrates that partial normalization is equivalent to full normalization with respect to the equational theory, guaranteeing that no necessary rewrite step is omitted.

To avoid infinite expansion of the proof tree, a loop‑detection mechanism based on König’s Lemma is incorporated. Whenever a pair (term, substitution) repeats, the branch is cut, ensuring termination of the semi‑decision process for negative instances.

Complexity analysis reveals that the worst‑case time remains exponential (as expected for the ground word problem), but empirical evaluation on standard benchmark TES instances and on a collection of randomly generated VP‑TES problems shows a substantial reduction in practice. On average, the number of explored nodes drops by about 30 % compared with the naïve forward enumeration and by about 28 % compared with the naïve backward reduction. Memory consumption also improves because the substitution space is dramatically smaller after class pruning.

The authors conclude that their natural refinements make the ground word problem more tractable for a wide range of practical TES, especially those arising in automated theorem proving and term rewriting. Future work is suggested in three directions: (i) further optimization of the canonical‑class computation, (ii) parallelization of the proof‑tree search, and (iii) extension of the techniques to richer equational frameworks such as conditional term rewriting systems.