构建高精度的复杂多维机器学习势能面指标数据集

本数据集主要是针对考核指标1.2中构建高精度的多电子态复杂多维势能面，主要面向复杂多维高精度势能面研究，对于动力学模拟过程中对于势能面精度以及效率的需求建设；基于量子化学概念的启发，开发了基于消息传递网络的递归嵌入原子神经网络模型，获得了原子化学环境的完备描述，主要记录了在甲烷分子和液相水的拟合精度等观测值，可以发现开发的模型能够精确的描述甲烷分级和液相水的体系，数据量1.85GB。

This dataset is developed to construct high-precision multi-electronic state complex multidimensional potential energy surfaces (PES) as specified in assessment indicator 1.2. It targets research on complex multidimensional high-precision PES, and is designed to address the requirements for PES accuracy and efficiency in dynamical simulations. Inspired by quantum chemistry concepts, we developed a recurrently embedded atom neural network model based on message-passing networks, which enables a comprehensive description of atomic chemical environments. This dataset mainly records observational metrics such as the fitting accuracy for methane molecules and liquid water systems. It is demonstrated that the developed model can accurately describe relevant systems including methane and liquid water, with a total dataset size of 1.85 GB.

搜集汇总

背景与挑战

背景概述

该数据集旨在构建高精度的多电子态复杂多维势能面，以支持动力学模拟中对精度和效率的需求。它基于量子化学概念，开发了递归嵌入原子神经网络模型，记录了甲烷分子和液相水的拟合精度观测值，数据量为1.85GB。数据集由中国科学技术大学在'局域场下受限小量子体系的量子调控'项目下创建，属于国家重点研发计划。