Automated analysis and validation of interatomic potentials for application in Materials Science

工程材料的结构和化学复杂性需要计算工具来描述和预测其在原子建模层次上的特性，从原子相互作用到材料微观结构再到工程规模。原子模拟中长度尺度的差异导致了原子间相互作用模型的差异，从多电子问题的复杂近似到简单的经验势。通常不清楚哪些属性是由不同模型一致预测的，以及哪些属性的预测存在显着差异。我们建议为原子间相互作用模型开发和实施透明且全面的验证框架。我们计划考虑实验参考数据，但重点是将原子间势与密度泛函理论数据进行比较。当我们计划对大量原子间势进行表征和验证时，我们将实施自动化工作流程和专用数据库来处理大量计算。特别是，为了提供参考数据，我们计划使用不同的交换相关泛函以受控的高精度自动生成密度泛函理论数据集。原子间势将首先针对基本材料特性进行验证，例如内聚能、弹性常数、声子谱和点缺陷形成能。第二步，我们专注于更抽象的措施，通过使用合适的描述符来评估原子间势的可转移性，这些描述符能够在低维空间中对势进行采样。将建立一个专门的网站atomistictools.org来显示原子间势的属性，从而使原子建模和模拟社区能够利用我们项目的结果进行自己的研究。该网站将提供交互式工具和图表，用于比较原子间势和密度泛函理论数据。评估原子间势的性质是每个原子模拟项目的第一步。我们希望我们的网站将帮助许多研究人员加快他们的项目，因为我们将消除评估原子间势预测的繁琐且耗时的任务。该提案是材料科学中原子间相互作用建模合作的一部分。我们与原子间势的开发和应用项目密切互动。

The structural and chemical complexity of engineering materials necessitates computational tools to describe and predict their properties on an atomistic modelling hierarchy spanning from the atomic interactions via the materials microstructure to the engineering scale. The disparity of length scales in atomistic simulations has led to a disparity of models of the interatomic interaction, from complex approximations of the many-electron problem to simple empirical potentials. It is often not clear which properties are consistently predicted by different models and for which properties the predictions differ significantly. We propose to develop and implement a transparent and comprehensive validation framework for models of the interatomic interaction. We plan to take into account experimental reference data but focus on comparing interatomic potentials to density functional theory data. As we plan the characterization and validation of a large number of interatomic potentials, we will implement an automated workflow and a dedicated database for handling large numbers of calculations. In particular, to provide reference data we plan the automated generation of density functional theory data sets with a controlled high accuracy using different exchange-correlation functionals.The interatomic potentials will initially be validated for basic materials properties, such as cohesive energies, elastic constants, phonon spectra and point defect formation energies. In a second step we focus on more abstract measures to assess the transferability of interatomic potentials by using suitable descriptors that enable to sample the potentials in a low-dimensional space.A dedicated website atomistictools.org will be set up to display the properties of interatomic potentials and in this way enable the atomistic modelling and simulation community to take advantage of the results of our project for their own research. The website will provide interactive tools and graphs for the comparison of interatomic potentials and density functional theory data. The assessment of the properties of an interatomic potential is the first step in every atomistic simulation project. We expect that our website will help many researchers to speed up their project, as we will take away the often cumbersome and time consuming task of evaluating the predictions of an interatomic potential.The proposal is part of a collaboration on modelling the interatomic interaction in materials science. We interact closely with projects on the development and application of interatomic potentials.