Collaborative Research: CDS&E: ReaxFF2: Efficient and Scalable Methods for Long-time Reactive Molecular Dynamics Simulations

该项目旨在通过高效和可扩展的技术实现反应分子系统的长时间模拟。长时间反应模拟对于催化、电池界面、涉及水的生物模拟以及表面氧化和化学气相沉积（CVD）生长等新兴领域等几个科学问题至关重要。然而，这些方面的进展是有限的，因为长时间的模拟大规模系统是非常困难的，如果不是不可能的，使用现有的方法来执行。反作用力场（ReaxFF）方法原则上非常适合此目的。然而，在目前的ReaxFF模拟和计算昂贵的力场公式所需的时间步长短限制了ReaxFF的时间能力，以缩小模拟时间范围。该项目旨在通过创建ReaxFF 2来克服这些限制，ReaxFF 2将时间尺度扩展一到两个数量级-从而使大规模，长时间的RMD模拟可供广泛社区使用。开发的代码将公开提供，该项目的结果将在一个专门的网站上突出显示，他们也将被纳入由PI的研讨会。在创建ReaxFF 2时，PI将显着增强Reax力场公式，并开发创新算法和软件实现可扩展的模拟。更具体地说，将制定替代ReaxFF相互作用，以消除能量方面的尖锐导数，并将ReaxFF时间步长增加至少四倍。为了加速RMD中所需的动态电荷分布模型，将开发用于迭代求解器的可扩展并行预处理技术。一个任务并行的方法来计算的相互作用，层次化的问题分解，向量化的关键内核，并使用混合精度的arithmetics constitutthe主要技术，将被用来充分利用大型计算机集群的性能能力。最后，将评估拟议的ReaxFF 2公式中的加速RMD概念的能力，并开发RMD的内联轨迹分析工具，以促进长期RMD模拟的研究。该项目将显著增强PI的软件开发、社区建设和RMD社区的维护工作。开发的代码、函数形式和参数集将公开提供，从而能够对本项目范围之外的各种反应系统进行快速准确的建模。对于社区外展，该项目的成果将在一个专门网站上突出显示，他们也将被纳入PI的研讨会。高级网络基础设施办公室的这一奖项由NSF数学和物理科学理事会的材料研究部和化学部共同支持。这一奖项反映了NSF的法定使命，并被认为值得通过使用基金会的知识价值和更广泛的影响审查标准进行评估。

This project aims to enable long-time simulations of reactive molecular systems through efficientand scalable techniques. Long-time reactive simulations are critical for several scientific problemssuch as catalysis, battery interfaces, biological simulations involving water, and emergingareas like surface oxidation and chemical vapor deposition (CVD) growth. However, progress onthese fronts is limited because long-time simulations of large-scale systems are very difficult, ifnot impossible, to perform using existing methods. The Reactive Force Field (ReaxFF) method is in principle ideally suited for this purpose. However, the short time steps required in current ReaxFF simulations and the computationally expensive force field formulation limit ReaxFF's temporal capabilities to narrow simulation time ranges. This project aims to overcome such limitations by creating ReaxFF2,which will extend time scales by one to two orders of magnitude - thus making large-scale, long-time RMD simulations accessible to a wide community. Codes developed will be made publicly available and results from this project will be highlighted on a dedicated website, and they will also be incorporated into workshops by the PIs.In creating ReaxFF2, the PIs will enhance the Reax force field formulation significantly, and develop innovative algorithms and software implementations for scalable simulations. More specifically, alternative ReaxFF interactions will be formulated to eliminate sharp derivatives in energy terms and enhance ReaxFF time step lengths by at least a factor of four. To accelerate the dynamic charge distribution models needed in RMD, scalable parallel preconditioning techniques for the iterative solvers will be developed. A task parallel approach to compute interactions, hierarchical problem decomposition, vectorization of the key kernels, and use of mixed precision arithmetics constitutethe main techniques that will be utilized to fully leverage the performance capabilities of largecomputer clusters. Finally, capabilities of accelerated RMD concepts in the proposed ReaxFF2 formulation will be evaluated and inlined trajectory analysis tools for RMD will be developed to facilitate the study of long-time RMD simulations. This project will significantly enhance the PIs' software development, community building, and sustenance efforts for the RMD community. Codes, functional forms, and parameter sets developed will be made publicly available, enabling fast and accurate modeling of diverse reactive systems beyond the scope of this project. For community outreach, results from this project will be highlighted on a dedicated website, and they will also be incorporated into workshops by the PIs.This award by the Office of Advanced Cyberinfrastructure is jointly supported by the Division of Materials Research and the Division of Chemistry within the NSF Directorate for Mathematical and Physical Sciences.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.