In this talk, we will present a Jx software [1-2] for calculating the magnetic exchange interactions, and the machine learning (ML) based analytic continuation [3] for correlated systems.
First, we will describe our newly-developed open-source software, named by Jx, to perform magnetic force linear response calculations based on first-principles calculations. Jx is a user-friendly and efficient tool to calculate magnetic interaction in solids and molecules [4-6]. Without supercell calculation, it computes both short- and long-range interactions. It is also possible to calculate an orbital-resolved matric form of magnetic couplings.
Then, we will present our results of the combining ML to the condensed matter physics. As ML technology advances in industrial applications, scientists are increasingly paying attention to the potential of ML for basic research, and physics is also no exception. From a wide range of ML applications in the physics area, condensed matter physics-related works will be briefly reviewed. And then we will introduce some of our recent works, including supervised learning in analytic continuation [3] and a hybrid approach for Monte Carlo sampling [7]. Especially, here we present a new Monte Carlo method whose sampling is assisted by the modern ML technique.
[1] H. Yoon et al, Phys. Rev. B 97, 125132 (2018).
[2] H. Yoon et al., Comput. Phys. Commun. 247, 106927 (2020).
[3] H. Yoon et al., Phys. Rev. B 98, 245101 (2018).
[4] S. W. Jang et al., Phys. Rev. B 98, 125126 (2018).
[5] T.J. Kim et al., Phys. Rev. B 97, 214431 (2018).
[6] S.W. Jang et al., Phys. Rev. Materials 3, 031001 (2019).
[7] in preparation
Host: Prof. Jisang Park