arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2408.06156physics.comp-phcond-mat.dis-nn
收藏

用于凝聚态物质的基于图的描述符

Graph-based Descriptors for Condensed Matter

An Wang, Gabriele C. Sosso

AI总结:

本文将网络科学中的节点中心性、聚类系数等图描述符引入凝聚态系统表示,并在 Lennard-Jones 系统中证明其预测动力学性质与相变优于对称函数。

摘要:

计算科学家长期以来一直在开发多种多样的方法组合,以表征凝聚态系统。由这些努力产生的大多数描述符最终都基于这些系统中粒子、原子或分子的空间构型。值得注意的例子包括对称函数和原子位置平滑重叠(SOAP)描述符,它们显著提升了针对凝聚态物质和小分子的预测性机器学习模型的性能。然而,尽管基于图的描述符经常被用于机器学习模型中以预测小分子的功能性质,但它们在凝聚态物质背景下的应用一直有限。在本文中,我们提出了若干传统上用于网络科学的基于图的描述符(例如节点中心性和聚类系数),作为凝聚态系统的替代表示。我们应用这种基于图的形式体系来研究典型的 Lennard-Jones 系统的动力学性质和相变。我们发现,在预测该系统的动力学性质和相变方面,我们的基于图的形式体系优于对称函数描述符。这些结果证明了基于图的特征在表示凝聚态系统方面的广泛适用性,并通过融合网络科学概念,为凝聚态物质领域令人振奋的进展铺平了道路。

英文摘要:

Computational scientists have long been developing a diverse portfolio of methodologies to characterise condensed matter systems. Most of the descriptors resulting from these efforts are ultimately based on the spatial configurations of particles, atoms, or molecules within these systems. Noteworthy examples include symmetry functions and the smooth overlap of atomic positions (SOAP) descriptors, which have significantly advanced the performance of predictive machine learning models for both condensed matter and small molecules. However, while graph-based descriptors are frequently employed in machine learning models to predict the functional properties of small molecules, their application in the context of condensed matter has been limited. In this paper, we put forward a number of graph-based descriptors (such as node centrality and clustering coefficients) traditionally utilised in network science, as alternative representations for condensed matter systems. We apply this graph-based formalism to investigate the dynamical properties and phase transitions of the prototypical Lennard-Jones system. We find that our graph-based formalism outperforms symmetry function descriptors in predicting the dynamical properties and phase transitions of this system. These results demonstrate the broad applicability of graph-based features in representing condensed matter systems, paving the way for exciting advancements in the field of condensed matter through the integration of network science concepts.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑