CDS&E: Fast, Scalable GPU-Enabled Software for Predictive Materials Design

非技术总结材料研究部和化学部为该奖项提供资金。它支持理论和计算研究、计算工具开发和教育，以支持通过计算机模拟预测新材料及其性能。材料科学家可以对从原子到分子再到纳米粒子的新材料的“构件”进行建模，并在速度较快的计算机上求解这些模型。尽管仍有许多工作要做，以提高额外的物理模型的保真度，通知模型来自实验的测量，并开发更快的算法在现代建筑上求解这些模型，但预测材料模拟的范例已经很好地掌握了。预测材料模拟可以被认为是一种“前进型”模拟，在这种模拟中，人们基本上从给定系统的初始条件运行计算机实验并观察结果。材料科学家现在正处于材料模拟范式转变的门槛，在这种情况下，为材料结构和性能进行设计变得可行，而不是简单地基于物理模型预测它们。将分子模拟用于“逆问题”方法来设计材料--即确定积木的属性，使这些积木将自组装成具有特定结构、光学、机械、热力学或其他性质的材料--将代表着一项重大飞跃。进入逆材料设计时代需要新的计算工具，能够执行模拟，翻转通常的正向模拟。这些工具必须具有科学有效性、健壮性、可访问性和易用性，并应利用最快的可用硬件。该项目将开发计算工具，与现有的和快速增长的用户群广泛共享，并将它们应用于纳米颗粒自组装和胶体结晶的重要问题。新工具将在笔记本电脑、台式机，甚至是大规模GPU(图形处理单元)超级计算机上高效运行，从而为多种用户类型提供服务。根据该奖项开发的方法和工具是可转让的，并将立即引起对逆向设计感兴趣的材料、工程和化学界的更广泛兴趣。技术摘要材料研究部和化学部为该奖项提供资金。它支持理论和计算研究、计算工具开发和教育，以支持通过计算机模拟预测新材料及其性能。今天的计算材料研究生态系统中缺少的是逆型预测软件；也就是说，能够从所需特性开始进行材料设计的软件。在设计的“反问题”模拟中，指定了目标材料的性质、行为和/或结构，并预测了实现该目标的构件和热力学条件。逆向设计方法可能仍然需要分子动力学、蒙特卡罗或其他主力算法，但整个概念方法是颠覆的。PI和她的团队最近开发了一种名为“数字炼金术”的逆向设计算法。该算法严格建立在统计热力学的基础上，基于广义热力学系综的思想。他们已经开发出一种方法来模拟“炼金术集合”中的材料，在这种集合中，形状或颗粒间相互作用等粒子属性被视为热力学变量，并在模拟过程中发生变化，在几分钟内采样数百万种可能性。在这些模拟中，定义了目标结构，并且模拟找到将自组装目标结构的最优构建块属性集。他们已经将数字炼金术应用于目标、单组分胶体晶体结构的纳米粒子和胶体粒子形状的设计，包括一些非常复杂的结构。在这个项目下，PI将充分将数字炼金术发展成一种可以处理其他实验相关粒子属性的方法，并将该方法应用到公开可用的开源粒子模拟工具包HOOMD-BLUE中。HOOMD-BLUE将能够“发现”针对目标结构、目标属性和/或目标行为进行优化的构建块属性和热力学条件。研究小组将展示利用HOOMD-BLUE的这种新的反向设计能力来设计多组分胶体晶体、片状颗粒的胶体晶体和可重新配置的胶体晶体。该项目的总体目标是向材料界提供能够进行逆向材料设计的开源软件，并通过几个与纳米颗粒自组装和胶体结晶相关的科学用例来展示基于GPU的逆向设计模拟的力量和可能性。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

NONTECHNICAL SUMMARYThe Division of Materials Research and the Chemistry Division contribute funds to this award. It supports theoretical and computational research, computational tool development, and education in support of the prediction of new materials and their properties by computer simulation. Materials scientists can model the "building blocks" of new materials, from atoms to molecules to nanoparticles, and solve those models on fast computers. Although much work is still needed to improve the fidelity of models with additional physics, inform the models with measurements from experiments, and develop faster algorithms on modern architectures to solve these models, the paradigm of predictive materials simulation is well in hand. Predictive materials simulation can be thought of as a "forward-type" simulation, where one essentially runs computer experiments from initial conditions for a given system and observes the result. Materials scientists are now at the threshold of a paradigm shift in materials simulation, where it is becoming feasible to design for materials structure and properties, rather than simply predicting them based on physical models. This use of molecular simulation for the "inverse problem" approach to design materials - that is, the determination of attributes of building blocks such that those building blocks will self-assemble into materials with specific structural, optical, mechanical, thermodynamic or other properties - would represent a major leap forward.Moving into an era of inverse materials design requires new computational tools capable of performing simulations that flip the usual forward-type simulations on their head. These tools must be scientifically valid, robust, accessible and easy to use, and should exploit the fastest available hardware. This project will develop computational tools, share them broadly with an existing and rapidly growing user base, and apply them to the important problems of nanoparticle self-assembly and colloidal crystallization. The new tools will run efficiently on laptops, desktops, and even massive GPU (graphics processing units) supercomputers, thereby serving multiple user types. The approaches and tools developed under this award are transferable and will be of immediate and even broader interest to the materials, engineering, and chemistry communities interested in inverse design.TECHNICAL SUMMARY The Division of Materials Research and the Chemistry Division contribute funds to this award. It supports theoretical and computational research, computational tool development, and education in support of the prediction of new materials and their properties by computer simulation. Missing in today's computational materials research ecosystem is inverse-type predictive software; that is, software that enables materials design starting from desired properties. In "inverse problem" simulations for design, a target materials property, behavior, and/or structure, is specified and the building blocks and thermodynamic conditions that will achieve that target are predicted. Inverse design methods may still require molecular dynamics, Monte Carlo, or other workhorse algorithms, but the entire conceptual approach is flipped on its head. The PI and her group recently developed an inverse design algorithm called "digital alchemy." The algorithm is grounded rigorously in statistical thermodynamics, and is based on the idea of generalized thermodynamic ensembles. They have developed a way to simulate materials in an "alchemical ensemble," where particle attributes like shape or interparticle interactions are treated as thermodynamic variables, and change during the course of the simulation, sampling many millions of possibilities in minutes. In these simulations, a target structure is defined, and the simulation finds the optimal set of building block attributes that will self-assemble the target structure. They have applied digital alchemy to the design of nanoparticle and colloidal particle shapes for target, one-component colloidal crystal structures, including some very complex ones.Under this project, the PI will fully develop digital alchemy into a method that can treat other, experimentally relevant particle attributes and implement the method into the publicly available, open source, particle simulation toolkit, HOOMD-blue. HOOMD-blue will be able to "discover" building block attributes and thermodynamic conditions optimized for a target structure, target properties, and/or target behavior. The research team will demonstrate the use of this new inverse design capability of HOOMD-blue to design multicomponent colloidal crystals, colloidal crystals of patchy particles, and reconfigurable colloidal crystals. The overarching goals of this project are to provide open-source software capable of inverse materials design to the materials community, and demonstrate the power and possibilities of GPU-enabled inverse design simulations via several scientific use cases relevant to nanoparticle self-assembly and colloidal crystallization.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.