Weisfeiler-Leman Is Incomplete on Simple Spectrum Graphs, so Canonicalize Them
About
Graphs with a simple spectrum admit cubic-time isomorphism testing, yet we prove that for every natural number $k$, the $k$-Weisfeiler-Leman ($k$-WL) test cannot distinguish all non-isomorphic graphs with a simple spectrum. As the WL hierarchy upper-bounds the distinguishing power of widely-used Graph Neural Networks (GNNs), this incompleteness applies to all such GNNs, ruling out completeness for every $k$-WL-aligned GNN family. To close this gap, we introduce PRiSM (Partition, Refine, Solve, Match), the first provably complete canonicalization of simple-spectrum eigendecompositions. PRiSM obtains the completeness guarantee that prior canonicalizations provably lack, and resolves the open problem of achieving complete expressivity on simple-spectrum graphs. When composed with DeepSets or a Transformer, PRiSM achieves universal approximation on simple-spectrum graphs, justifying the use of canonicalized Laplacian positional encodings. Empirically, PRiSM performs comparably to or outperforms existing spectral canonicalizations on graph regression, classification, and expressivity
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Molecular property prediction | MolHIV | ROC-AUC74.3 | 39 | |
| Molecular property prediction | MOLTOX21 | ROC-AUC0.783 | 33 | |
| Graph Expressivity | BREC | Basic Expressivity Score60 | 26 | |
| Molecular property prediction | ZINC-12K | MAE0.103 | 8 | |
| Molecular property prediction | OGB MOLPCBA | AP27.8 | 7 | |
| Molecular property prediction | OGB molclintox | ROC-AUC85.76 | 4 | |
| Molecular property prediction | OGB MOLBBBP | ROC-AUC0.7088 | 4 | |
| Molecular property prediction | alchemy | MAE0.1196 | 4 |