Abstract:The difficult problem of relating the static structure of glassy liquids and their dynamics is a good target for Machine Learning, an approach which excels at finding complex patterns hidden in data. Indeed, this approach is currently a hot topic in the glassy liquids community, where the state of the art consists in Graph Neural Networks (GNNs), which have great expressive power but are heavy models and lack interpretability. Inspired by recent advances in the field of Machine Learning group-equivariant representations, we build a GNN that learns a robust representation of the glass' static structure by constraining it to preserve the roto-translation (SE(3)) equivariance. We show that this constraint significantly improves the predictive power at comparable or reduced number of parameters but most importantly, improves the ability to generalize to unseen temperatures. While remaining a Deep network, our model has improved interpretability compared to other GNNs, as the action of our basic convolution layer relates directly to well-known rotation-invariant expert features. Through transfer-learning experiments displaying unprecedented performance, we demonstrate that our network learns a robust representation, which allows us to push forward the idea of a learned structural order parameter for glasses.

What problem does this paper attempt to address?

The paper focuses on the problem of dynamic transformation in glassy liquids, which is an unsolved problem in condensed matter physics. When a glass forms, the viscosity increases significantly as the liquid is cooled to its characteristic temperature (Tg), but the structure (spatial arrangement of particles) appears to remain unchanged. Researchers believe that using machine learning to uncover the hidden complex patterns behind this dynamic transformation is an effective approach. The current state-of-the-art method is the Graph Neural Network (GNN), which has powerful expressive capabilities but lacks interpretability. Inspired by recent advances in the machine learning field regarding group-equivariant representations, the paper constructs a GNN that learns robust representations of the static structure of glass by constraining the network to maintain rotational and translational (SE(3)) equivariance. This approach improves prediction ability, reduces the number of parameters, and enhances generalization to unseen temperatures. Compared to traditional deep networks, this model has better interpretability because the operations of its basic convolutional layers are directly related to known rotation-invariant expert features. Through transfer learning experiments, the paper demonstrates that the learned representations by the network can drive the establishment of a learned structural order parameter for describing glass. The paper explains how to construct SE(3) equivariant graph neural network layers and discusses various preprocessing choices that affect performance in detail. Additionally, it explores the learned representations' interpretability as order parameters and tests their robustness. Future directions of work include further improving the model and exploring the potential of equivariance in other applications. In conclusion, the paper aims to address how to better understand and predict the dynamics of glassy liquids through equivariant graph neural networks, providing new tools and insights for understanding glass transitions.

Rotation-equivariant Graph Neural Networks for Learning Glassy Liquids Representations

A geometry-enhanced graph neural network for learning the smoothness of glassy dynamics from static structure

Enhancing the Prediction of Glass Dynamics by Incorporating the Direction of Deviation from Equilibrium Positions

Learning molecular dynamics: predicting the dynamics of glasses by a machine learning simulator

Predicting dynamic heterogeneity in glass-forming liquids by physics-inspired machine learning

Toward Interpreting the Thermally Activated β Dynamics in Metallic Glass Using the Structural Constraint Neural Network

Roadmap on machine learning glassy dynamics

Predicting the pathways of string-like motions in metallic glasses via path-featurizing graph neural networks

What Do Deep Neural Networks Find in Disordered Structures of Glasses?

Towards interpreting the thermally activated $β$ dynamics in metallic glass with the structural constraint neural network

Relaxing Continuous Constraints of Equivariant Graph Neural Networks for Broad Physical Dynamics Learning

Relaxing Continuous Constraints of Equivariant Graph Neural Networks for Physical Dynamics Learning

Importance of equivariant and invariant symmetries for fluid flow modeling

GLASS: GNN with Labeling Tricks for Subgraph Representation Learning

Evaluation of GlassNet for physics-informed machine learning of glass stability and glass-forming ability

A Variational Graph Partitioning Approach to Modeling Protein Liquid-liquid Phase Separation

Dynamic heterogeneity at the experimental glass transition predicted by transferable machine learning

Rotation-Equivariant Graph Convolutional Networks for Spherical Data Via Global-Local Attention.

Multi-scale rotation-equivariant graph neural networks for unsteady Eulerian fluid dynamics

Predicting fluid-structure interaction with graph neural networks

Predicting and Interpreting Energy Barriers of Metallic Glasses with Graph Neural Networks