Why Multi-Layer Message Passing Works: Completeness Theory for Graph Neural Network Interatomic Potentials
Mirrored from arXiv — Machine Learning for archival readability. Support the source by reading on the original site.
Computer Science > Machine Learning
Title:Why Multi-Layer Message Passing Works: Completeness Theory for Graph Neural Network Interatomic Potentials
Abstract:We prove that the Hypergraph Neural Network, an invariant architecture with 3-body message passing, is a universal approximator for potential energy surfaces. Our main contribution is a multi-layer completeness theory. We show that $L$ layers of message passing on sparse, cutoff-based graphs achieve the same representational power as having access to the full $L$-hop neighborhood, provided the configurations are generic, satisfy an overlap condition and a connectivity condition. This provides the first rigorous justification for the common practice of using multi-layer message passing with a per-layer cutoff smaller than the physical interaction range, the setting used by virtually all practical graph neural network based machine-learned interatomic potentials. As immediate consequences, we show that both DPA3 and CHGNet architectures inherit universal approximation.
| Subjects: | Machine Learning (cs.LG); Mathematical Physics (math-ph); Chemical Physics (physics.chem-ph); Computational Physics (physics.comp-ph) |
| Cite as: | arXiv:2609.00528 [cs.LG] |
| (or arXiv:2609.00528v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2609.00528
arXiv-issued DOI via DataCite (pending registration)
|
Access Paper:
- View PDF
- HTML (experimental)
- TeX Source
Current browse context:
References & Citations
Bookmark
Bibliographic and Citation Tools
Code, Data and Media Associated with this Article
Demos
Recommenders and Search Tools
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
More from arXiv — Machine Learning
-
AhaBench: Do Agents Learn from Prior Experience? A Benchmark for Long-Horizon Continual Learning
Sep 10
-
When Do Options Help? Policy Necrosis and Redundant Coverage in Option-Critic
Sep 10
-
Multi-granularity Adaptive Hypergraph Representation Learning via Granular-ball
Sep 10
-
Capsule Lens: Locating and Tracking Concept Geometry in Model Representations
Sep 10
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.