A Machine-Learning based Deep Potential (DP) model for Al clusters is developed through training with an extended database including ab initio data of both bulk and several clusters in only 6 CPU/h. This DP model has good performance in accurately predicting the low-lying candidates of Al clusters in a broad size range. Based on our developed DP model, the low-lying structures of 101 different sized Al clusters are extensively searched, among which the lowest-energy candidates of 69 sized clusters are updated. Our calculations demonstrate that machine-learning is indeed powerful in generating potentials to describe the interaction of atoms in complex materials.

中文翻译：

---

基于机器学习的深层潜力来寻找Al集群的低层候选者

通过使用扩展数据库进行培训，开发了针对Al群集的基于机器学习的深潜（DP）模型，该数据库包括批量和几个群集的从头算数据，仅需6 CPU / h。这种DP模型在准确预测较大尺寸范围内的Al团簇的低洼候选者方面具有良好的性能。基于我们开发的DP模型，广泛搜索了101个不同尺寸的Al簇的低层结构，其中更新了69个尺寸的簇的最低能量候选。我们的计算表明，机器学习在生成描述复杂材料中原子相互作用的潜力方面确实非常强大。