How would graph neural networks handle variable-length decay topologies compared to fixed-architecture sequence models?

This is a thoughtful question about how two architectural paradigms handle a specific data property, but the provided preprints don't directly address it.

The archive includes several papers on GNNs in quantum computing and particle physics [AG-2024.05-2472, AG-2024.02-1203, AG-2024.05-2284], and one on evolving graph structures [AG-2024.07-1359], but none of them directly compare GNN flexibility on variable-length decay topologies against fixed-architecture sequence models (like RNNs or Transformers). The closest relevant result is that EggNet uses recursively evolving graph attention rather than fixed input graphs for track reconstruction [AG-2024.07-1359], which suggests GNNs can adapt structure iteratively—but this doesn't constitute a head-to-head comparison on decay topology handling.

To get a rigorous answer, you'd want a paper that either benchmarks GNNs against sequence models on hierarchical branching structures, or provides theoretical analysis of expressivity differences on variable-multiplicity tree graphs.