有效的原子间描述器用于精确的机器学习扩展分子的力场
Adil Kabylda1, Valentin Vassilev-Galindo1, Stefan Chmiela2,3
1Department of Physics and Materials Science, University of Luxembourg, L-1511, Luxembourg City, Luxembourg.
Nature communications
|June 15, 2023
概括
本研究介绍了一种自动化方法,通过减少描述符特征来创建更有效的机器学习力场 (MLFF). 这种方法提高了复杂系统分子动力学模拟的准确性和速度.
科学领域:
- 计算化学是一种计算化学.
- 材料科学是一种材料科学.
- 生物物理学的生物物理.
背景情况:
- 机器学习力场 (MLFFs) 旨在以降低计算成本实现精确的分子动力学 (MD) 模拟.
- 当前的MLFF面临着有效描述非局部交互和减少描述符维度以获得更广泛的适用性方面的挑战.
- 对现实的分子系统进行预测MLFF模拟需要改进描述器设计.
研究的目的:
- 开发一种自动化的方法来减少MLFF中的原子间描述符特征.
- 同时应对高效的非局部交互描述符和描述符维度减少的挑战.
- 提高分子模拟MLFF的准确性,可解释性和效率.
主要方法:
- 开发了一种自动化特征选择/减少方法,并应用于梯度域机器学习 (GDML) 框架.
- 该方法侧重于优化非局部原子间相互作用的描述符.
- 该方法在各种系统上进行了测试,包括,DNA基对,脂肪酸和超分子复合体.
主要成果:
- 拟议的方法显著减少了描述符特征的数量,同时保持了MLFF的准确性.
- 发现,延伸至15 Å的非局部特征对于在各种分子系统中保持精度至关重要.
- 在优化描述器中,基本非局部特征的数量与局部特征 (低于5 Å) 变得可比.
- 增加了MLFF的效率,计算成本可能与系统大小线性扩展.
结论:
- 开发的自动化方法有效地减少了MLFF中的描述符维度.
- 精确的分子动力学模拟需要包括关键的非局部原子间相互作用.
- 这项工作可以开发具有线性扩展计算成本的全球MLFF,从而促进更大,更复杂的模拟.
更多相关视频
12:11Computation of Atmospheric Concentrations of Molecular Clusters from ab initio Thermochemistry
Published on: April 8, 2020
8.2K
10:52Multiscale Sampling of a Heterogeneous Water/Metal Catalyst Interface using Density Functional Theory and Force-Field Molecular Dynamics
Published on: April 12, 2019
12.9K
相关概念视频
03:13Intermolecular Forces
58.8K
Atoms and molecules interact through bonds (or forces): intramolecular and intermolecular. The forces are electrostatic as they arise from interactions (attractive or repulsive) between charged species (permanent, partial, or temporary charges) and exist with varying strengths between ions, polar, nonpolar, and neutral molecules. The different types of intermolecular forces are ion–dipole, dipole–dipole, hydrogen bonds, and dispersion; among these, dipole–dipole, hydrogen...
58.8K
01:24Van der Waals Interactions
64.1K
Atoms and molecules interact with each other through intermolecular forces. These electrostatic forces arise from attractive or repulsive interactions between particles with permanent, partial, or temporary charges. The intermolecular forces between neutral atoms and molecules are ion–dipole, dipole–dipole, and dispersion forces, collectively known as van der Waals forces.
64.1K
02:56Intermolecular Forces and Physical Properties
20.9K
20.9K
02:00Molecular Models
38.7K
Physical models representing molecular architectures of chemical compounds play essential roles in understanding chemistry. The use of molecular models makes it easier to visualize the structures and shapes of atoms and molecules.
38.7K
03:00Intermolecular vs Intramolecular Forces
87.8K
Intermolecular forces (IMF) are electrostatic attractions arising from charge-charge interactions between molecules. The strength of the intermolecular force is influenced by the distance of separation between molecules. The forces significantly affect the interactions in solids and liquids, where the molecules are close together. In gases, IMFs become important only under high-pressure conditions (due to the proximity of gas molecules). Intermolecular forces dictate the physical properties of...
87.8K
02:27Predicting Molecular Geometry
34.6K
VSEPR Theory for Determination of Electron Pair Geometries
34.6K