MagCluster-GMAF: A Physics-Inspired Hybrid Graph Architecture for Long-Range Magnetic Property Prediction
- 1School of Artificial intelligence, Changchun Normal University, Changchun 130032, China
- 2iFLYTEK Co., Ltd., No. 666 Wangjiang West Road, Hefei 230088, China
Abstract
Understanding and predicting the structure–property relationships in magnetic clusters remain challenging due to the complex interplay between geometry, electronic structure and spin configurations: density functional theory suffers from prohibitive computational cost for large-scale exploration, while existing graph neural networks struggle to balance long-range magnetic correlation capture, topological fidelity and scalability. We present MagCluster-GMAF, a physics-inspired hybrid graph architecture with magnetic cluster-aware serialization to preserve topological and electronic similarity during graph-to-sequence transformation, paired with a dual-branch design combining a Mamba-2 backbone for efficient long-range spin correlation modeling and a lightweight attention module as topological compensation. On a 2D doped magnetic material dataset, it achieved a total magnetic moment prediction MAE of 0.2683 μB, reducing the prediction error by 9.5% relative to the GPS + Transformer baseline and outperforming Graph-Mamba; at the 1400-node scale, it cut the GPU memory consumption by over 70% versus standard graph transformers while exhibiting near-linear empirical scaling at the studied graph scales. SHAP-based feature attribution analysis further identifies the key physicochemical descriptors governing the predictions, providing an effective approach for nanoscale magnetic material screening and the high-throughput screening of spintronic materials.
Related
- Systematic Exploration of the Gd–Os–Ge Ternary System with Magnetic and Transport Property Studies of Highly-Disordered Quasiskutterudite Gd3.4Os4Ge12.6
- Toward a New Paradigm for Renewable Biopolymer Selection: A Structure–Processing–Architecture–Property–Scalability Framework (Alginate, Cellulose, Lignin, Chitosan, and Starch)
- Materials Property Prediction with a Site-Resolved Three-Dimensional Voxel Convolution Neural Network
- Closed-loop Auto Research for Molecular Property Prediction: Discovering and Certifying Generalizable Improvements
- Topology-Guided Multi-Level Graph Learning: A Unified Framework for Molecular Property Prediction
BibTeX
@article{zhang2026magclustergmaf,
title = {{MagCluster-GMAF: A Physics-Inspired Hybrid Graph Architecture for Long-Range Magnetic Property Prediction}},
author = {Yu Zhang and Jiale Wang and Haotian Liu and Yishuo Wang and Guangjie Liu and Wei Liu},
journal = {Nanomaterials},
year = {2026},
month = nov,
doi = {10.3390/nano16211422},
url = {https://www.mdpi.com/2079-4991/16/21/1422},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - MagCluster-GMAF: A Physics-Inspired Hybrid Graph Architecture for Long-Range Magnetic Property Prediction
AU - Yu Zhang
AU - Jiale Wang
AU - Haotian Liu
AU - Yishuo Wang
AU - Guangjie Liu
AU - Wei Liu
PY - 2026
DA - 2026/11/04
JO - Nanomaterials
DO - 10.3390/nano16211422
UR - https://www.mdpi.com/2079-4991/16/21/1422
AB - Understanding and predicting the structure–property relationships in magnetic clusters remain challenging due to the complex interplay between geometry, electronic structure and spin configurations: density functional theory suffers from prohibitive computational cost for large-scale exploration, while existing graph neural networks struggle to balance long-range magnetic correlation capture, topological fidelity and scalability. We present MagCluster-GMAF, a physics-inspired hybrid graph architecture with magnetic cluster-aware serialization to preserve topological and electronic similarity during graph-to-sequence transformation, paired with a dual-branch design combining a Mamba-2 backbone for efficient long-range spin correlation modeling and a lightweight attention module as topological compensation. On a 2D doped magnetic material dataset, it achieved a total magnetic moment prediction MAE of 0.2683 μB, reducing the prediction error by 9.5% relative to the GPS + Transformer baseline and outperforming Graph-Mamba; at the 1400-node scale, it cut the GPU memory consumption by over 70% versus standard graph transformers while exhibiting near-linear empirical scaling at the studied graph scales. SHAP-based feature attribution analysis further identifies the key physicochemical descriptors governing the predictions, providing an effective approach for nanoscale magnetic material screening and the high-throughput screening of spintronic materials.
SN - 2079-4991
ER -