SI2-SSE: Infrastructure Enabling Broad Adoption of New Methods That Yield Orders-of-Magnitude Speedup of Molecular Simulation Averaging

今天，人们强烈期望，未来的材料将首先在计算机上进行大量研究，并通过计算确定实验室合成的最佳候选者。通过这种方式，工程师可以有效地配制出更轻、更坚固或其他功能更有效的新材料。从能源到医药，从交通到制造业，所有技术领域都需要这样的进步。分子模型界在量化原子相互作用方面的最新进展正在迅速消除实现这一愿景的一个关键障碍。然而，一个重要的障碍仍然存在：在实际感兴趣的条件下，预测晶体结构和性能需要材料的热性能--除了最低温度之外，材料的热性能对所有材料都很重要。这些性质太昂贵了，无法一次计算许多材料，因为这是基于计算的筛选工作所需的。项目团队已经开发了一种算法，在不损失任何精度的情况下显著加快了这些计算，因此在消除这一障碍方面有很大的帮助。这个项目的目的是让正在使用分子模拟来理解和开发新材料的研究人员能够获得这一突破。为此，该项目将改进和扩展这些方法，然后将计算机代码添加到广泛使用的分子模拟程序包中，以便它们可以使用这些新技术进行计算。此外，该小组还在努力提高人们的认识，确保这些方法及其实施的易用性。“映射平均”是最近公布的一项重新制定总体平均的方案。该框架使用统计力学理论的近似结果来推导出准确代表该理论中的误差的新的集合平均(映射平均)。构思良好的地图平均值可以通过分子模拟来计算，具有显着的精度和效率；在有利的情况下，计算节省了许多数量级。对于晶体系统，谐和近似提供了一个合适的起点，允许模拟精确地计算非谐对性质的贡献。其结果是一种以前所未有的、变革性的效率计算晶体性质的技术。该项目的目的是在用于晶体系统模拟的成熟和广泛使用的软件包上实施这些方法，并为这些系统的用户感兴趣的新应用开发映射平均值。该项目的理论基础最近(2015年)才出现在文献中，因此拟议的工作完全是新颖的。这些技术不容易理解，而且实现起来也很繁琐，因此被更大的社区采用将需要这种有针对性的基础设施开发，以使临时用户更容易访问它们。完整的开发团队包括计算科学家和软件工程师，他们对引入这些元素的包进行编码、维护和分发。该小组协助项目调查人员与模拟包对接，同时确保新代码以最高标准编写。完整的开发团队还一起工作，以确保软件元素的正确性和可用性得到彻底验证。除实施外，该项目还旨在扩大地图平均法的范围，以涵盖以前没有应用过的性质和物质。该项目使地图平均方法能够在几个广泛使用的分子模拟程序包上使用：VIZ、LAMMPS、HOOMD、Cassandra和Vasp，这些程序包总共有数千名用户。这个项目中实现的软件元素在许多情况下对这些包的用户是完全透明的，并且可以由他们使用，而不会增加复杂性，以将其计算速度加快数量级。因此，该项目所做的努力将产生一种使能技术，使科学家和工程师有新的能力来为实际应用制定材料。在这个项目中构建了开发工具和脚本，这将有助于其他开发人员将映射平均方法扩展到更多的分子模拟包、材料属性和分子模型系统。为该项目开发的软件是开源的分布式软件。在这个项目中开发的知识被巩固，以形成在网络上可用的课程材料，并作为由PI教授的研究生分子模拟课程的一部分。在整个项目期间，对1名博士生和大量硕士和本科生进行培训。涉及论文、文档、演示文稿和研讨会的强有力的传播努力确保这些方法和工具被社区理解和采用。最后，开发了指导性的、以图形为导向的分子模拟模块，并在网络上提供，以传达与晶体行为的谐和非谐分量相关的概念，通过绘制的平均框架使独特的能力成为可能。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

There is, today, a strong expectation that future materials will be studied in huge numbers first on the computer, and the best candidates for synthesis in the laboratory will be identified computationally. In this way engineers can efficiently formulate new materials that are lighter, stronger, or otherwise more functionally effective. Such advances are needed across all fields of technology, from energy to medicine to transportation to manufacturing. Recent advances from the molecular modeling community toward quantifying atomic interactions are rapidly eliminating a key obstacle to realization of this vision. Yet, an important obstacle remains: the thermal properties of materials -- those that are important at all but the lowest temperatures -- are needed to predict crystal structures and properties at conditions of practical interest. These properties are too expensive to compute for many materials at once, as needed for a computation-based screening effort. The project team has developed an algorithm that significantly accelerates these calculations without any loss of accuracy, and therefore goes a long way toward removing this obstacle. The aim of this project is to make this breakthrough available to researchers who are using molecular simulation to understand and develop new materials. To this end, this project will refine and extend these methods, and then add computer code to widely-used molecular simulation packages so that they can perform calculations using these new techniques. The team is additionally making efforts to promote awareness and ensure ease-of-use of the methods and their implementation."Mapped averaging" is a recently published scheme for the reformulation of ensemble averages. The framework uses approximate results from statistical mechanical theory to derive new ensemble averages (mapped averages) that represent exactly the error in the theory. Well-conceived mapped averages can be computed by molecular simulation with remarkable precision and efficiency; in favorable cases the computational savings are many orders of magnitude. For crystalline systems, a harmonic approximation provides a suitable starting point, allowing simulation to compute precisely the anharmonic contribution to the properties. The result is a technique for computing crystalline properties with unprecedented, transformative efficiency. The aim of this project is to implement these methods on well-established and widely used software packages for simulation of crystalline systems, and to develop mapped averages for new applications of interest to the users of these systems. The theoretical basis for this project appeared in the literature very recently (2015), so the proposed work is completely novel. The techniques are not trivial to understand and are tedious implement, hence adoption by the larger community will require this targeted infrastructure development to make them more accessible to casual users. The full development team includes the computational scientists and software engineers who coded, maintain and distribute the packages where these elements will be introduced. This group assists the project investigators to interface with the simulation packages while ensuring that the new codes are written to the highest standards. The full development team works together also to ensure that the software elements are thoroughly validated for correctness and usability. In addition to the implementation, the project also aims to expand the scope of the mapped-averaging method to encompass properties and substances to which it was not previously applied. This project enables mapped averaging methods to be employed on several widely-used molecular simulation packages: viz, LAMMPS, HOOMD, Cassandra, and VASP, which altogether have a base encompassing thousands of users. Software elements implemented in this project are in many cases completely transparent to the users of these packages, and can be employed by them with no added complication, to speed up their calculations by orders of magnitude. Thus the efforts made in this project will produce an enabling technology, giving scientists and engineers new capabilities to formulate materials for practical applications. Development tools and scripts are constructed in this project, which will facilitate the extension of mapped-averaging methods by other developers to even more molecular simulation packages, material properties, and molecular model systems. Software developed for this project is distributed open-source. Knowledge developed in this project is consolidated to form course materials made available on the web, and used as part of a large component of a graduate molecular simulation course taught by the PI. Training of 1 PhD student and numerous MS and undergraduates occurs across the project period. A strong dissemination effort involving papers, documentation, presentations, and workshops ensure that these methods and tools are understood and adopted by the community. Finally, instructional, graphically-oriented molecular simulation modules are developed and made available on the web to convey concepts related to harmonic and anharmonic components of crystalline behavior, with unique capabilities made possible by the mapped averaging framework.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.