NEP89
收藏资源简介:
NEP89是一个基于神经进化势(NEP)架构的通用模型,适用于89种化学元素的无机和有机材料的原子模拟。该数据集包含1100万个结构,通过描述符空间抽样和迭代主动学习过程从多个数据集中筛选而来,确保了数据集的多样性和可靠性。NEP89模型在预测静态和动态性质方面具有竞争力,并且比现有模型计算效率提高了3-4个数量级,使得之前无法实现的大规模原子模拟成为可能。该模型还支持在小数据集上进行微调,以快速适应特定应用。NEP89的推出标志着机器学习势能的重大进展,能够支持在各个研究领域和社区进行高性能的原子模拟。
NEP89 is a universal model based on the neuroevolutionary potential (NEP) architecture, designed for atomic simulations of inorganic and organic materials covering 89 chemical elements. This dataset contains 11 million structures, which are screened from multiple datasets via descriptor space sampling and iterative active learning processes, ensuring the diversity and reliability of the dataset. The NEP89 model delivers competitive performance in predicting both static and dynamic properties, with its computational efficiency improved by 3 to 4 orders of magnitude compared to existing models, enabling large-scale atomic simulations that were previously infeasible. Additionally, the model supports fine-tuning on small datasets to rapidly adapt to specific applications. The release of NEP89 marks a significant advancement in machine learning potentials, enabling high-performance atomic simulations across various research fields and communities.




