Materials design using first principles calculations and machine learning

2021 Fiscal Year Final Research Report

• Project/Area Number: 19H02419
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Review Section: Basic Section 26010:Metallic material properties-related
• Research Institution: Kyoto University
• Principal Investigator: Seko Atsuto 京都大学, 工学研究科, 准教授 (10452319)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: 機械学習 / 第一原理計算 / 結晶構造探索

Outline of Final Research Achievements

The machine-learning potential (MLP) providing an accurate description of the relationship between the energy and the crystal structure and its potential applications are of growing interest. Such an approach is a framework of polynomial MLP, in which the introduction of group-theoretical high-order rotational polynomial invariants contributes to systematically derive MLPs with high predictive power for a wide range of structures, including extreme structures. This approach successfully constructs accurate and efficient MLPs in a variety of elemental metals and alloys. The Pareto optimal polynomial MLPs with different trade-offs between accuracy and computational efficiency for various systems are distributed in Polynomial Machine Learning Potential Repository with our implementation (polymlp-package) that enables us to use the polynomial MLPs in the LAMMPS code.

Free Research Field

計算材料科学

Academic Significance and Societal Importance of the Research Achievements

基礎的な第一原理計算は，系の元素・結晶構造をもとに，エネルギーや電子状態を計算するものであり，材料研究に広く用いられている．しかし，実際の材料の物性や現象に対しては，非常に単純なモデルを導入し第一原理計算を行う以外なく，その精度を確かめる手段すらない．そのような状況において，本研究は，第一原理計算の精度で，実際の材料物性や現象を取り扱う方法を構築することで，材料研究を大幅に進展させるものである．