Abstract:We introduce machine learning (ML) models that predict the electronic structure of materials across a wide temperature range. Our models employ neural networks and are trained on density functional theory (DFT) data. Unlike most other ML models that use DFT data, our models directly predict the local density of states (LDOS) of the electronic structure. This provides several advantages, including access to multiple observables such as the electronic density and electronic total free energy. Moreover, our models account for both the electronic and ionic temperatures independently, making them ideal for applications like laser-heating of matter. We validate the efficacy of our LDOS-based models on a metallic test system. They accurately capture energetic effects induced by variations in ionic and electronic temperatures over a broad temperature range, even when trained on a subset of these temperatures. These findings open up exciting opportunities for investigating the electronic structure of materials under both ambient and extreme conditions.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to predict the electronic structure of materials over a wide temperature range. Specifically, the researchers introduced a machine - learning (ML) model that can directly predict the local density of states (LDOS) of materials at different temperatures based on density - functional - theory (DFT) data. This method overcomes the problem of low computational efficiency in traditional DFT calculations when dealing with large systems and high - temperature conditions, making it possible to study the electronic structure of materials at different temperatures, especially in extreme conditions, such as in applications like laser - heated materials. ### Core problems of the paper 1. **Electronic structure prediction**: Traditional DFT calculations have problems of high computational complexity and long calculation time when dealing with large systems and high - temperature conditions. This paper proposes a machine - learning - based method. By training a neural - network model to predict the local density of states (LDOS) of materials, important physical quantities such as electronic density and total free energy can be indirectly obtained. 2. **Temperature dependence**: Most existing ML models only consider the ion temperature (\(\tau_i\)), while ignoring the influence of the electron temperature (\(\tau_e\)). The model proposed in this paper can independently consider the electron temperature and the ion temperature, which is especially important when dealing with the properties of materials under non - equilibrium conditions, such as the situation where the electron temperature rises rapidly while the ion temperature is relatively low during the laser - heating process. 3. **Applicability in multiple temperature ranges**: Traditional DFT calculations need to be retrained at different temperatures, while the model proposed in this paper can effectively predict the electronic structure in a wider temperature range after being trained in a limited temperature range, thus greatly reducing the computational cost. ### Main contributions - **Direct prediction of LDOS**: By directly predicting LDOS, the model can provide multiple observables, such as electronic density and total free energy, rather than just a single physical quantity. - **Temperature independence**: The model can independently handle the electron temperature and the ion temperature, and is suitable for material research under non - equilibrium conditions. - **Prediction ability across temperature ranges**: After being trained in a limited temperature range, the model can accurately predict the electronic structure in a wider temperature range, showing its application potential under extreme conditions. ### Method overview - **Finite - temperature DFT**: The paper details the theoretical framework of finite - temperature DFT, including the treatment methods of electron and ion temperatures. - **ML model**: A neural - network is used to train the model. The input is the feature vector describing the local ion structure, and the output is LDOS. The model learns the relationship between the electronic structure and temperature through the training data set. - **Verification and application**: Through the test of the aluminum - metal system, the prediction ability and accuracy of the model at different temperatures are verified. In conclusion, this paper proposes a new machine - learning method that can efficiently and accurately predict the electronic structure of materials in a wide temperature range, providing a powerful tool for research in the fields of materials science and chemistry.

Machine learning the electronic structure of matter across temperatures

A Machine-Learning Surrogate Model for ab initio Electronic Correlations at Extreme Conditions

Electronic excited states from physically-constrained machine learning

Machine Learning-based estimation and explainable artificial intelligence-supported interpretation of the critical temperature from magnetic ab initio Heusler alloys data

Machine Learning for Harnessing Thermal Energy: From Materials Discovery to System Optimization

Electronic Excited States from Physically Constrained Machine Learning

Machine learning approach for vibronically renormalized electronic band structures

A Deep Dive into Machine Learning Density Functional Theory for Materials Science and Chemistry

Deep dive into machine learning density functional theory for materials science and chemistry

Machine Learning Prediction of Thermodynamic Stability and Electronic Properties of 2D Layered Conductive Metal-Organic Frameworks

Universal Machine Learning Kohn-Sham Hamiltonian for Materials

Machine learning with systematic density-functional theory calculations: Application to melting temperatures of single and binary component solids

Accelerating finite-temperature Kohn-Sham density functional theory with deep neural networks

Machine learning for molecular simulation

Lattice Thermal Conductivity: An Accelerated Discovery Guided by Machine Learning

Machine Learning approaches to classical density functional theory

Machine-learning free-energy functionals using density profiles from simulations

Machine learning for accuracy in density functional approximations

Incorporating electronic information into Machine Learning potential energy surfaces via approaching the ground-state electronic energy as a function of atom-based electronic populations