Advanced learning strategies for potential energy surfaces applied to organic electrolytes

在这个项目中，我们将通过机器学习技术从准确的DFT数据中开发有机电解质的有效原子间势。我们的工作流程旨在优化开发过程的每个阶段，包括数据生成和选择，通过主动学习进行模型训练和改进，以及在生产模拟中准确部署的不确定性估计。该项目的成果将包括一个完全优化的工作流程，用于开发任意原子间势的分子建模，包括远程相互作用，以及能够实现该工作流程的高性能软件套件。除了方法上的改进，我们的项目将促进几乎从头到脚精确的有机电解质原子间电位的发展，因此，高度精确的分子动力学模拟将对纳米级及更高级别的复杂分子液体的结构和动力学产生重要的见解。

In this project we will develop efficient inter-atomic potentials for organic electrolytes via machine-learning techniques from accurate DFT data. Our workflow aims at optimizing each stage of the development process including data generation and selection, model training and refinement through active learning, as well as uncertainty estimation for accurate deployment in production simulations. The outcome of this project will include a fully optimized workflow for the development of arbitrary inter-atomic potentials for molecular modelling, including long range interactions, as well as high-performance software suites capable of implementing this workflow. Beyond methodological improvements, our project will facilitate the development of almost ab-initio accurate inter-atomic potentials for organic electrolytes and therefore, highly accurate MD simulation which should yield important insights into the structure and dynamics of complex molecular liquids at the nanoscale and beyond.