EA://INTEL — The decidable boundary for graph rewriting and why normalizing unrestricted graphs fails

When you try to determine structural equivalence in a general system using local graph rewriting rules without limits, the problem quickly becomes undecidable. This happens specifically because confluence—a property ensuring different reduction paths lead to the same result—breaks down when cycles are introduced into the graph structure. It is not that the mathematics of rewriting fails in these scenarios, but rather that no algorithm can exist to verify equivalence for every possible configuration if those cycles are permitted freely.

This difficulty led to the withdrawal of a specific decision regarding how to handle such normalizations within our broader formalism. The project formally retired the claim associated with unrestricted normalization because confluence of terminating rewriting in general cyclic graphs is undecidable, a fact established by existing literature on critical pairs and joinability which do not entail confluence as they do for term systems. The boundary of decidability turns out to be the presence of cycles themselves, not the size or topology of the graph beyond that specific feature.

To move forward with building a universal language where we can actually reason about meaning without infinite loops in our logic, the design choice was made to restrict the normalizing fragment to acyclic term graphs. Within this restricted regime, decidability is recovered, and the formalism regains its practical utility for engineering applications. Reviving the unrestricted form would strictly require a new decidability result for cyclic structures that currently does not exist in the literature.

Is it possible to define an admissible region for our repair operators that remains robust even when the underlying meaning space admits cycles?