Unsupervised identification of local atomic environment from atomistic potential descriptors
arxiv(2024)
摘要
Analyzing local structures effectively is key to unraveling the origin of
many physical phenomena. Unsupervised algorithms offer an effective way of
handling systems in which order parameters are unknown or computationally
expensive. By combining novel unsupervised algorithm (Pairwise Controlled
Manifold Approximation Projection) with atomistic potential descriptors, we
distinguish between various chemical environments with minimal computational
overhead. In particular, we apply this method to silicon and water systems. The
algorithm effectively distinguishes between solid structures and phases of
silicon, including solid and liquid phases, and accurately identifies
interstitial, monovacancy, and surface atoms in diamond structures. In the case
of water, it is capable of identifying an ice nucleus in the liquid phase,
demonstrating its applicability in nucleation studies.
更多查看译文
AI 理解论文
溯源树
样例
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要