Abstract:

Molecular dynamics simulation has become an important tool for the research and development of chemical engineering processes and technologies. However, the insufficient accuracy of classical molecular dynamics simulations and the high computational cost of ab initio molecular dynamics simulations have restricted the widespread applications of molecular simulation technology. The emergence and development of machine learning technology has led to the rapid development of molecular simulation based on machine learning potentials, which offers an efficient way to achieve a greatly improved accuracy at a lower computing loading, thereby bolstering the potential of molecular simulations in practical applications. This review started by an overview of the development of machine learning potentials with emphasis on the construction methods and principles of machine learning potential models. The techniques associated with machine learning potentials including dataset construction, model training, model transfer and application were detailed. The strengths and weaknesses of different types of machine learning models were also discussed, followed by the prospects for the development and applications machine learning potentials.

Key words: machine learning potentials, molecular simulation, computational chemistry, thermodynamics

摘要：

分子动力学模拟已经成为化工过程和技术研发的重要工具，但经典分子动力学模拟的精度不足和从头计算分子动力学模拟的高昂计算成本，制约了分子模拟技术的广泛应用。机器学习技术的出现和发展使得基于机器学习势的分子模拟快速发展起来，该方法兼具速度快与准确性高的优势，将极大地加速分子模拟技术在化工中的应用。首先回顾了机器学习势的发展历程，给出了构建机器学习势模型的原则，介绍了数据集构建、模型训练和模型迁移与应用等，分析了不同类型的机器学习势的特点和局限性，最后对机器学习势的应用前景进行了展望。

关键词: 机器学习势, 分子模拟, 计算化学, 热力学