a) Warm dense matter (WDM) exists in the interior of giant planets in astrophysics and inertial confinement fusion. It is challenging to model WDM since partially degenerate electrons strongly interact with ions. In this regard, a quantum mechanics description of the electrons and ions in WDM is needed. Due to the lack of sufficient experimental data of WDM, first-principles computational methods have attracted increasing attention. Therefore, quantum-mechanics-based first-principles methods are ideal for studying WDM.
b) Designing reactors that can generate fusion energy as a viable energy source has been a great challenge for decades. One of the challenging issues is to build plasma-facing components that survive intense particle bombardments present in the harsh environment of a fusion reactor. In this regard, solid plasma-facing materials unavoidably suffer from erosion when they are exposed to high fluxes of particles, which may also lead to performance degradation of the plasma facing components. Liquid metals own a series of attractive properties as alternative plasma-facing materials. We conduct molecular dynamics simulations to study properties of liquid metals at high temperatures.
c) Water is one of the upmost important material for life and technology. We utilize state-of-the-art first-principles molecular dynamics to study liquid water and ions (hydronium and hydroxide). We also applied recently proposed SCAN functional, which is a form of meta-GGA functional that satisfies all 17 known constraints, to study liquid water and found excellent agreement between simulation and experimental results.
2. Developments and Applications of Density Functional Theory
a) We develop orbital-free density functional theory (OFDFT), which utilizes a kinetic energy density functional (KEDF) for evaluations of electronic kinetic energies. The accuracy of OFDFT is largely limited by the accuracy of KEDFs. We aim to provide more accurate KEDFs for real applications.
c) We apply density functional theory to different applications such as mechanical properties of metals and alloys.
a) The past few years have witnessed the rapid developments of machine learning techniques for modeling the potential energy surface of molecular systems. In particular, the Deep Potential (DP) method is suitable for molecular dynamics with the accuracy comparable to that of DFT and with the efficiency similar to that of empirical potentials. Specifically, the DP model constructs a deep neural network to describe the interactions among atoms in a system. By choosing first-principles molecular dynamics trajectories as the training data, the DP method enables accurate determination of energies and forces for all of the atoms in the system. Therefore, the computational cost for electronic structure evaluation in DFT is eliminated and the DP model is a linear-scaling method. The DP method has been successfully applied to a variety of condensed matter systems.
b) We develop machine learning methods in order to solve complicated engineering problems encountered by real experiments. This is a new direction of the group and we are actively working on it.