2023 Fiscal Year Final Research Report
Distribution Matching Principle for Machine Learning Based Molecular Simulation
Project/Area Number |
20K20907
|
Research Category |
Grant-in-Aid for Challenging Research (Exploratory)
|
Allocation Type | Multi-year Fund |
Review Section |
Medium-sized Section 13:Condensed matter physics and related fields
|
Research Institution | National Institutes for Quantum Science and Technology |
Principal Investigator |
Sakuraba Shun 国立研究開発法人量子科学技術研究開発機構, 量子生命科学研究所, 主幹研究員 (90647380)
|
Project Period (FY) |
2020-07-30 – 2024-03-31
|
Keywords | 記号回帰 / 分子動力学シミュレーション / パラメータサーチ |
Outline of Final Research Achievements |
Classical molecular dynamics (MD) simulations enable us to analyze various physicochemical properties of molecules by reproducing and simulating molecular structures on computers. In classical MD simulations, the function of the force field that describes the interactions between atoms (or groups of atoms), as well as the parameters of the force field function, determines the behavior of the molecules in the computer. Finding appropriate functions and their parameters is thus vital in the simulation. In this research, I aimed to find force field functions with low calculation costs by computer. I implemented a symbolic regression algorithm fitted to the molecular simulation and searched the functions. In an existing dataset of quantum chemical calculation, functions that fit well with the dataset were successfully found.
|
Free Research Field |
計算科学
|
Academic Significance and Societal Importance of the Research Achievements |
これまで、分子シミュレーションの力場の提案は人の直感に基づく関数系の提案と、長い時間を掛けた人手によるパラメータ改善の試行錯誤により実現されてきた。本研究では実際のデータからnon-trivialな関数系の「発見」を行っており、分子シミュレーションの力場の提案をデータ中心に行う一助となることが期待される。これにより、現在は高コストな計算(量子化学計算、全原子シミュレーション)がより低コストな計算(古典、陰溶媒、粗視化)で近似できるシステマティックな手法が整備され、階層的なシミュレーションがより容易に、低コストで、大規模に実現されていくことが期待できる。
|