OWL-RL rule-based reasoning non-terminating due to self-referential property chains

[ThHanke]

Problem

OWL-RL (rule-based / forward-chaining) reasoning on PMDco does not terminate.
ELK and HermiT (description logic, tableau-based) work correctly via ODK.

Root Cause

6 self-referential property chains detected. All follow the pattern:
SubObjectPropertyOf(ObjectPropertyChain(P has_part) P)

This is valid OWL DL but causes unbounded rule firing in OWL-RL:
no cycle-blocking mechanism exists in forward-chaining engines.

Property	Source
BFO:RO_0000057	BFO/RO import
BFO:RO_0001025	BFO/RO import
BFO:RO_0002233	BFO/RO import
BFO:RO_0002234	BFO/RO import
pmdco:PMD_0025998 (has relational quality)	pmdco-qualities.owl
pmdco:PMD_0025999 (relational quality of)	pmdco-qualities.owl

Additionally: BFO:RO_0002082 is both Symmetric and Transitive (amplifier).

Evidence

Triple count during OWL-RL run: 10,756 → 50,529+ in 20s, no convergence.

Diagnostic Tool

scripts/owlrl_diagnose.py added — runs static analysis + timed OWL-RL attempt on Turtle release artifacts.

Notes

• BFO/RO imports are upstream — pmdco cannot fix those 4 chains
• PMD_0025998/PMD_0025999 chains are pmdco-owned and could be reviewed
• OWL-RL profile may simply be incompatible with BFO-based ontologies by design

[ThHanke]

solution attempt!
create a pre reasoned owl-rl ready artifact that materiaizes chain axioms into the artifact and then remove the chain axioms. just another release artifact. Will make reasoning much faster on big graphs and is not losing much of discription logic.

[ThHanke]

Solution

New release artifact pmdco-owlrl.ttl built by:

1. HermiT materialises all property chain entailments as explicit triples
2. owl:propertyChainAxiom entry triples stripped via SPARQL update

All chain-derived semantics are preserved — now as asserted triples.
OWL-RL sees no chains to loop on.

Build: make owlrl-ready (ODK container)

Benchmark

Reasoner	Input	Time	Result
ELK (merge+reason)	pmdco-edit.owl	3.75s	✓
HermiT (merge+reason)	pmdco-edit.owl	30.86s	✓
OWL-RL	pmdco-full.ttl (raw)	>30s	✗ non-terminating
OWL-RL	pmdco-owlrl.ttl	28s	✓

Known Coverage Gaps

OWL-RL is forward-only Datalog — certain DL inferences remain incomplete
even on the pre-materialized artifact. Axioms stay in the graph; only
specific inference directions are unavailable:

Axiom type	Count	What is lost
equivalentClass (complex)	126	Left-to-right decomposition (no witness creation, no disjunctive rule heads)
allValuesFrom	99	Closed-world filler constraints
disjointWith	29	Classification from disjointness
unionOf	40	Disjunctive conclusions
complementOf	6	Negation-based membership
oneOf (nominals)	4	Nominal + constructor combinations
hasValue	2	Complex hierarchy interactions

For use cases requiring full DL reasoning, continue using HermiT or ELK.
pmdco-owlrl.ttl targets rule engines, SPARQL entailment, and pySHACL.

Files Changed

• src/ontology/pmdco.Makefile — owlrl-ready make target
• src/sparql/strip-property-chains.ru — SPARQL update to strip chain axioms
• scripts/owlrl_diagnose.py — diagnostic + benchmark tool
• pmdco-owlrl.ttl — pre-materialized OWL-RL ready release artifact