Abstract:We present a general-purpose machine learning model for predicting properties of crystals. Specifically, energy of formation, Fermi level energy, band gap, partial charges, and bulk modulus as well as spectral properties, including electronic and phonon densities of states are targeted. Thus, our model can be used to screen materials for specific properties. The model is based on atomic representations which enables it to effectively capture complex information about each atom and its surrounding environment in a crystal. The accuracy achieved for band gap values exceeds results previously published. By design, our model is not restricted to electronic properties discussed here but can be extended to fit diverse chemical databases.

What problem does this paper attempt to address?

This paper proposes a solution to the problem of predicting electronic properties of crystals. In the study, the authors develop a general machine learning model that can predict the formation energy, Fermi level energy, band gap, partial charges, bulk modulus, and spectral properties of crystals, including electronic and phononic densities of states. By using atomic representation, the model can effectively capture the complex information of each atom and its surrounding environment. The accuracy of the model surpasses previous results in band gap prediction and is designed to be scalable to accommodate different chemical databases. The paper first introduces the application of machine learning in computational chemistry, particularly successful cases in materials property prediction. It then discusses two common structural representation methods - graph neural networks (GNN) and high-dimensional neural networks (HDNN), and compares their advantages and disadvantages. Electronic density of states (eDOS) is a key quantity for describing material's electronic properties, and its accurate prediction is crucial for the design of new materials, but quantum chemical methods are computationally expensive. Therefore, the paper proposes a deep learning model that combines dual distribution functions and atomic embedding representation as input parameters to compute eDOS and other electronic properties. The model is trained on the Materials Project dataset, demonstrating competitiveness compared to existing models, especially in predicting band gaps and eDOS. Through testing on different datasets, the predictive capability of the model, particularly in predicting Fermi level energy and band gap, is shown to be comparable to deep graph neural networks with lower computational cost. In conclusion, this paper aims to develop a fast and accurate model for predicting the electronic properties of crystals using machine learning techniques, providing a tool for material screening and future research.

Structure to Property: Machine Learning Methods for Predicting Electronic Properties of Crystals

Structure to Property: Chemical Element Embeddings for Predicting Electronic Properties of Crystals

Machine learning for the structure-energy-property landscapes of molecular crystals

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

Band gap prediction for large organic crystal structures with machine learning

Machine learning approaches for feature engineering of the crystal structure: Application to the prediction of the formation energy of cubic compounds

On the Robustness of Machine Learning Models in Predicting Thermodynamic Properties: a Case of Searching for New Quasicrystal Approximants

Machine Learning of Molecular Electronic Properties in Chemical Compound Space

What Information is Necessary and Sufficient to Predict Materials Properties using Machine Learning?

Machine learning modeling for the prediction of materials energy

Machine learning assisted prediction of organic salt structure properties

Topological representations of crystalline compounds for the machine-learning prediction of materials properties

Machine-Learned Fragment-Based Energies for Crystal Structure Prediction

Crystal Structure Representations for Machine Learning Models of Formation Energies

Efficient Surrogate Models for Materials Science Simulations: Machine Learning-based Prediction of Microstructure Properties

Interpretable Ensemble Learning for Materials Property Prediction with Classical Interatomic Potentials: Carbon as an Example

Prediction and Classification of Formation Energies of Binary Compounds by Machine Learning: An Approach without Crystal Structure Information

Predicting the Electronic and Structural Properties of Two-Dimensional Materials Using Machine Learning

Machine Learning-Based Prediction of Crystal Systems and Space Groups from Inorganic Materials Compositions

LLM-Prop: Predicting Physical And Electronic Properties Of Crystalline Solids From Their Text Descriptions

Crystal structure prediction with machine learning-based element substitution