Abstract:Modern machine learning techniques have been extensively applied to materials science, especially for property prediction tasks. A majority of these methods address scalar property predictions, while more challenging spectral properties remain less emphasized. We formulate a crystal-to-sequence learning task and propose a novel attention-based learning method, Xtal2DoS, which decodes the sequential representation of the material density of states (DoS) properties by incorporating the learned atomic embeddings through attention networks. Experiments show Xtal2DoS is faster than the existing models, and consistently outperforms other state-of-the-art methods on four metrics for two fundamental spectral properties, phonon and electronic DoS.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the Density of States (DoS) prediction problem in the crystal structure - to - sequence learning tasks in materials science. Specifically, the paper focuses on the prediction of two fundamental spectral properties - phonon Density of States (phDoS) and electronic Density of States (eDoS). Traditional machine - learning methods perform poorly when dealing with these complex spectral properties, especially when predicting the Density of States in the form of continuous functions. Therefore, this paper proposes a new attention - mechanism - based graph - to - sequence learning method, namely Xtal2DoS, aiming to improve the prediction speed and accuracy by combining the learned atomic embeddings to decode the sequential representation of the Density of States properties of materials. The main contributions of the paper are as follows: 1. **Proposing a new graph - to - sequence learning framework**: Xtal2DoS, which can effectively extract features from crystal structures and convert them into sequential representations of the Density of States. 2. **Adopting an improved graph attention network**: for encoding crystal - graph structures, which not only considers the local atomic environment but also captures the information of the entire crystal structure through a global attention mechanism. 3. **Introducing an attention - mechanism - based Transformer decoder**: for generating the sequential representation of the Density of States. This method can better capture the self - correlation of the sequence and the similarity between the source and the target, thereby improving the prediction accuracy and efficiency. 4. **Experimental verification**: By comparing with several existing advanced methods, it is proved that Xtal2DoS has superior performance on multiple evaluation indicators, especially in predicting phonon and electronic Density of States. In conclusion, this paper solves the difficult problem of predicting complex spectral properties in materials science by proposing an innovative graph - to - sequence learning framework, providing a new solution for materials property prediction.

Xtal2DoS: Attention-based Crystal to Sequence Learning for Density of States Prediction

Density of states prediction for materials discovery via contrastive learning from probabilistic embeddings

Density of States Prediction of Crystalline Materials via Prompt-guided Multi-Modal Transformer

Towards Large-Scale and Spatiotemporally Resolved Diagnosis of Electronic Density of States by Deep Learning

Self-supervised learning for crystal property prediction via denoising

AlphaCrystal-II: Distance matrix based crystal structure prediction using deep learning

Partial Density of States Representation for Accurate Deep Neural Network Predictions of X-ray Spectra

Study of Crystal Property Prediction Based on Dual Attention Mechanism and Transfer Learning

Atomic Positional Embedding-Based Transformer Model for Predicting the Density of States of Crystalline Materials.

Prediction of the electron density of states for crystalline compounds with Atomistic Line Graph Neural Networks (ALIGNN)

Predicting Density of States via Multi-modal Transformer

A Crystal-Specific Pre-Training Framework for Crystal Material Property Prediction

AlphaCrystal: Contact map based crystal structure prediction using deep learning

A Cluster‐Based Deep Learning Model Perceiving Series Correlation for Accurate Prediction of Phonon Spectrum

ADA-GNN: Atom-Distance-Angle Graph Neural Network for Crystal Material Property Prediction

Direct Prediction of Gas Adsorption Via Spatial Atom Interaction Learning

Pattern Learning Electronic Density of States

How to represent crystal structures for machine learning: Towards fast prediction of electronic properties

Towards Large-Scale and Spatio-temporally Resolved Diagnosis of Electronic Density of States by Deep Learning

Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties