paper

Semantic foundations of equality saturation — Dan Suciu, Yisu Remy Wang & Yihong Zhang (2025; ICDT 2025). arXiv:2501.02413 (v1, PDF).

Gives equality saturation a fixed-point semantics. An e-graph is a deterministic, reachable tree automaton; its meaning is a partial congruence relation, and every such relation has a unique e-graph. Equality saturation is the least fixed point of an inflationary, monotone operator and is a universal model of the rewrite rules. It corresponds exactly to a class of chase sequences from database theory, and termination is studied in three variants — all undecidable or worse in general — with an acyclicity criterion that guarantees it.

Sources: the paper, arXiv:2501.02413v1, checked against the arXiv listing. Index: Papers.

Key definitions and results

  • Definition 3: an e-graph is a deterministic reachable tree automaton
  • Definition 5, Theorem 6: semantics is a partial congruence relation (PCR); PCRs ↔ e-graphs bijectively
  • Lemma 11: at most one homomorphism between e-graphs
  • Definition 13: an e-graph is a model of a term rewriting system
  • Theorem 19: EqSat is the least fixed point of ICOR and a universal model
  • Corollary 27: finite convergence
  • Theorem 30: EqSat terminates iff the corresponding standard chase does
  • Theorems 31–33: single-instance termination R.E.-complete; all-term-instance -complete; all-e-graph-instance undecidable

Concept notes

E-Graph, Least Fixed Point

Used in Sophia

Compilation as Query