Element priors and target support shape chemical transfer in materials graph networks
Abstract
Materials graph neural networks must often transfer to chemical regions weakly represented in training data. Such transfer can rely on predefined relations among elements or supervised evidence from target-containing structures, but these pathways are usually entangled. Here, held-out-element splits and incremental target support separate their roles. Without target-containing training structures, formation-energy errors depend strongly on the element representation, particularly for H, O and F. Matched perturbations show that representation-induced sharing matters beyond input dimension or numerical form, while a label-free similarity-graph prior reduces selected zero-shot errors. Adding a few target-containing structures sharply lowers errors and contracts differences among one-hot, k-hot and continuous inputs across ALIGNN and CGCNN. Calibration explains only part of this recovery, and freezing the initial element projection preserves most gains in five of six ALIGNN splits. Target support therefore shifts chemical transfer from reliance on static element relations toward learning from target-containing environments.
Related
- Finite-element reconstruction of molten-zone dynamics and chemical segregation in optical floating-zone grown Ni--Mn--Ga
- Compiling Chemical Knowledge into Executable Descriptors for Materials Prediction
- Toward Nanoscale Chemical Clarity: THz-Pulsed Atom Probe Tomography for Hydrated Biological Materials
- Atomistic Modeling of Chemical Disorder in Materials: Bridging Conventional Methods and AI-Assisted Approaches
- Impact of Water Mist on Heat Transfer and Chemical Species from a Liquid Pool Fire
BibTeX
@misc{zhao2026element,
title = {{Element priors and target support shape chemical transfer in materials graph networks}},
author = {Ran Zhao and Kangming Li},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.00915},
eprint = {2609.00915},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.00915},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Element priors and target support shape chemical transfer in materials graph networks
AU - Ran Zhao
AU - Kangming Li
PY - 2026
DA - 2026/09/01
PB - arXiv
DO - 10.48550/arxiv.2609.00915
UR - https://arxiv.org/abs/2609.00915
AB - Materials graph neural networks must often transfer to chemical regions weakly represented in training data. Such transfer can rely on predefined relations among elements or supervised evidence from target-containing structures, but these pathways are usually entangled. Here, held-out-element splits and incremental target support separate their roles. Without target-containing training structures, formation-energy errors depend strongly on the element representation, particularly for H, O and F. Matched perturbations show that representation-induced sharing matters beyond input dimension or numerical form, while a label-free similarity-graph prior reduces selected zero-shot errors. Adding a few target-containing structures sharply lowers errors and contracts differences among one-hot, k-hot and continuous inputs across ALIGNN and CGCNN. Calibration explains only part of this recovery, and freezing the initial element projection preserves most gains in five of six ALIGNN splits. Target support therefore shifts chemical transfer from reliance on static element relations toward learning from target-containing environments.
ER -