论文标题
同等分子动力学模拟和缺陷形成能的功能不确定性定量
Functional uncertainty quantification for isobaric molecular dynamics simulations and defect formation energies
论文作者
论文摘要
最近提出了功能不确定性定量（FUNUQ），以量化源自输入函数的模型和模拟中的不确定性，而不是参数。本文扩展了FUNUQ，以量化不确定性，该不确定性源自等温静脉分子动力学（MD）模拟中的原子间电位以及缺陷形成能的计算。我们根据扰动理论得出并验证计算廉价表达以计算MD中的功能衍生物。我们表明，相对于原子间潜能的这种功能量（在我们案例中平均内部能量，体积和缺陷能量）的功能衍生物可用于预测这些数量，以获得不同的原子间潜能，而无需重新运行模拟。在MD中执行FUNUQ的代码和脚本可自由下载。此外，为了促进可重复性并启用该方法的最佳实践，我们创建了Jupyter Notebooks来对MD模拟进行Funuq分析，并使其可在NanoHub进行在线模拟。该工具使用云计算资源和用户可以从标准Web浏览器中查看，编辑和运行端到端的工作流，而无需下载或安装任何软件。
Functional uncertainty quantification (FunUQ) was recently proposed to quantify uncertainties in models and simulations that originate from input functions, as opposed to parameters. This paper extends FunUQ to quantify uncertainties originating from interatomic potentials in isothermal-isobaric molecular dynamics (MD) simulations and to the calculation of defect formation energies. We derive and verify a computationally inexpensive expression to compute functional derivatives in MD based on perturbation theory. We show that this functional derivative of the quantities of interest (average internal energy, volume, and defect energies in our case) with respect to the interatomic potential can be used to predict those quantities for a different interatomic potential, without re-running the simulation. The codes and scripts to perform FunUQ in MD are freely available for download. In addition, to facilitate reproducibility and to enable use of best practices for the approach, we created Jupyter notebooks to perform FunUQ analysis on MD simulations and made them available for online simulation in nanoHUB. The tool uses cloud computing resources and users can view, edit, and run end-to-end workflows from a standard web-browser without the need to need to download or install any software.
