First-principles calculations for strongly-correlated materials with an interdisciplinary approach based on machine learning and physics

2022 Fiscal Year Final Research Report

• Project/Area Number: 20K14423
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 13030:Magnetism, superconductivity and strongly correlated systems-related
• Research Institution: Keio University (2022); Institute of Physical and Chemical Research (2020-2021)
• Principal Investigator: Nomura Yusuke 慶應義塾大学, 理工学部(矢上), 准教授 (20793756)
• Project Period (FY): 2020-04-01 – 2023-03-31
• Keywords: 強相関電子系 / 量子相関 / 機械学習 / 人工ニューラルネットワーク

Outline of Final Research Achievements

This study aimed to establish a new numerical method for strongly correlated electron systems based on a novel approach combining machine learning schemes with first principles and quantum many-body calculations. Specifically, we focused on 1) development of numerical methods, 2) verification of the accuracy of the developed methods, and 3) application to strongly correlated electron systems.
As for 1 and 2, we successfully established a deep learning model incorporating quantum and thermal fluctuations at finite temperatures by combining the artificial neural network method with the physics concept of "purification." As an achievement of 3, we studied the electronic states of strongly correlated ferromagnetic metals and discovered an unusual quantum correlation in a paramagnetic phase.

Free Research Field

物性理論

Academic Significance and Societal Importance of the Research Achievements

強相関電子系は、様々な機能物性が発現するが、それらの現象は量子性と多体性の兼ね合いにより発現しているために、その数値的解析は物理の挑戦的課題として知られている。本研究では、その挑戦的課題に対し、人工ニューラルネットワーク・機械学習を用いるという新機軸を導入し、その手法をさらに発展させることで、非自明な量子相関を計算するスキームを大きく進展させた。非自明な量子相関を正確かつ定量的に計算できるようになると、強相関電子系の物性予測に繋がる。今回の研究はそのような理論主導の物質設計につながる礎を気付いたという点で意義深いものである。