Data-driven approach to condensed-matter physics

2015 Fiscal Year Final Research Report

• Project/Area Number: 25610102
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Mathematical physics/Fundamental condensed matter physics
• Research Institution: The University of Tokyo
• Principal Investigator: Hukushima Koji 東京大学, 総合文化研究科, 准教授 (80282606)
• Project Period (FY): 2013-04-01 – 2016-03-31
• Keywords: 物性理論 / 磁性モデル / 磁化曲線 / 機械学習 / モンテカルロ法 / 中性子散乱

Outline of Final Research Achievements

Our main purpose in this project is to develop a data-driven method for materials science. We first propose a systematic method based on the machine learning in which a theoretical model with magnetic interactions is established from an input data of magnetization curve. The proposed method enables us to infer a suitable model among many candidates. Consequently, one may obtain microscopic spin structure which is difficult to see in a conventional experiments and it provides useful information on a following experiment design. It is also found to be efficient to use a Bayesian optimization for quantum spin systems. Meanwhile, dealing with the possibility of handling a big data from large experimental facilities, we develop an inference method of a relaxation-time distribution from neutron-diffraction experiments and offer a new direction including real-data analyses.

Free Research Field

統計物理学