CDS&E: Fast, scalable GPU-enabled software for predictive materials design & discovery

非技术总结这个计算和数据使能科学与工程奖支持计算材料研究和材料研究的计算工具的开发。设计新型材料需要新的计算工具，这些工具能够进行模拟，揭示意想不到的见解，从而为我们思考材料和材料工艺的方式提供信息。这些工具必须具有科学有效性、可靠性、可访问性和易用性，并应利用最快的可用硬件。今天，这种硬件涉及图形处理单元，称为GPU，其架构利用了大量的并行性，允许在单个芯片上每秒同时完成比当前更传统的CPU更多的计算。 为了使用GPU来研究材料系统和化学过程，研究人员社区使用的主力算法和代码必须专门针对该架构进行重新设计和重写。该项目将开发这些工具，与现有的和快速增长的用户群广泛分享，并将其作为一个示范领域应用于胶体结晶的突出和计算要求高的问题。在胶体结晶中，悬浮在溶液中的微米级颗粒自组装成有序结构，从而产生具有广泛应用的性质和行为。预测这些结构需要大量的计算，特别是对于复杂的晶体结构。 该项目还将培养学生在软件工程，算法设计和开源软件开发材料模拟。在此奖项下开发的HOOMD-Blue、DEM-HOOMD-Blue和HPMC的增强功能将通过密歇根大学CodeBlue上的HOOMD-Blue网站提供给更广泛的社区。技术总结此计算和数据支持科学与工程奖支持计算材料研究和材料研究计算工具的开发。该项目将开发材料和化学系统的模拟软件。 PI建立在被称为高度优化的面向对象的许多粒子动力学-蓝色的开源软件平台上，将扩展HOOMD-Blue的功能，以包括为图形处理器单元优化的离散元分子动力学和蒙特卡罗算法。作为颗粒物质的主力工具，DEM将在没有摩擦的情况下用于硬颗粒碰撞，从而可以高保真地研究胶体系统的动力学和热力学。蒙特卡罗-一种传统的串行算法，用于采样相空间随机-将利用HOOMD基础设施，以实现高度并行使用棋盘式策略。这两种添加物，DEM-HOOMD-blue和HPMC，将允许模拟基于粒子的系统和相当复杂的材料过程。PI将证明代码的有效性，通过将它们应用于硬粒子系统中的晶体成核和生长问题，通过熵最大化驱动到有序。结合罕见事件采样工具，DEM-HOOMD-blue和HPMC将使热力学和动力学途径的研究，硬颗粒流体组装成准晶体和开放的，手性的，或分层的晶体，其特征在于大或复杂的晶胞。鉴于目前纳米粒子和胶体的成核和生长模拟研究的最新技术水平仅限于简单的布拉维晶格，该项目将扩大设计新晶体材料所需的知识基础。 按照目前的HOOMD-blue策略，新的计算工具将在笔记本电脑、台式机和大规模GPU集群上高效运行，从而为多种用户类型提供服务。PI的发现将立即引起纳米粒子和胶体组装社区的兴趣。PI的方法和工具是可转移的，并且将对对适当的原子、分子或纳米级构建块的结晶感兴趣的材料、工程和化学社区产生直接甚至更广泛的兴趣。该项目还将培养学生在软件工程，算法设计和开源软件开发材料模拟。在此奖项下开发的HOOMD-Blue，DEM-HOOMD-Blue和HPMC的增强功能将通过密歇根大学CodeBlue上的HOOMD-Blue网站提供给更广泛的社区。

NONTECHNICAL SUMMARYThis Computational and Data-Enabled Science and Engineering award supports computational materials research and the development of computational tools for materials research. Designing novel materials requires new computational tools capable of performing simulations that reveal unexpected insights that, in turn, inform the way we think about materials and materials processes. These tools must be scientifically valid, robust, accessible and easy to use, and should exploit the fastest available hardware. Today this hardware involves graphics processing units, known as GPUS, whose architecture exploits massive parallelism, allowing many more calculations to be done simultaneously per second on a single chip than on current, more traditional CPUs. To use GPUs for the study of materials systems and chemical processes, the workhorse algorithms and codes used by that community of researchers must be redesigned and rewritten specifically for that architecture. This project will develop those tools, share them broadly with an existing and rapidly growing user base, and apply them, as an exemplar area, to the outstanding and computationally demanding problem of colloidal crystallization. In colloidal crystallization, micron-sized particles suspended in solution self-assemble into ordered structures, giving rise to properties and behavior with wide-ranging application. Predicting these structures requires considerable computation, especially for complex crystal structures. This project will also train students in software engineering, algorithm design, and open source software development for materials simulation. The enhancements to HOOMD-Blue, DEM-HOOMD-Blue and HPMC developed under this award will be made available to the broader community through the HOOMD-Blue website on University of Michigan Codeblue.TECHNICAL SUMMARYThis Computational and Data-Enabled Science and Engineering award supports computational materials research and the development of computational tools for materials research. This project will develop simulation software for materials and chemical systems. Building on the open source software platform known as highly optimized object oriented many particle dynamics-blue, the PI will expand the capabilities of HOOMD-Blue to include discrete-element molecular dynamics and Monte Carlo algorithms optimized for Graphics Processor Units. A workhorse tool for granular matter, DEM will be adopted for hard particle collisions in the absence of friction, allowing high fidelity studies of the dynamics and thermodynamics of colloidal systems. Monte Carlo - a traditionally serial algorithm for sampling phase space stochastically - will leverage the HOOMD infrastructure to achieve a high degree of parallelism using a checkerboarding strategy. Both additions, DEM-HOOMD-blue and HPMC, will allow the simulation of particle-based systems and materials processes of considerable complexity. The PI will demonstrate the efficacy of the codes by applying them to the problem of crystal nucleation and growth in hard particle systems driven to order by entropy maximization. Combined with rare event sampling tools, DEM-HOOMD-blue and HPMC will enable the study of thermodynamic and kinetic pathways by which hard-particle fluids assemble into quasicrystals and open, chiral, or hierarchical crystals characterized by large or complex unit cells. Given that the current state of the art in nucleation and growth simulation studies of nanoparticles and colloids is limited to simple Bravais lattices, this project will expand the knowledge base needed to design new crystalline materials. Following current HOOMD-blue strategy, the new computational tools will run efficiently on laptops, desktops, and massive GPU clusters, thereby serving multiple user types. The PI's findings will be of immediate interest to the nanoparticle and colloidal assembly communities. The PI's approaches and tools are transferable and will be of immediate and even broader interest to the materials, engineering, and chemistry communities interested in crystallization of appropriate atomic, molecular, or nanoscale building blocks. This project will also train students in software engineering, algorithm design, and open source software development for materials simulation. The enhancements to HOOMD-Blue, DEM-HOOMD-Blue and HPMC developed under this award will be made available to the broader community through the HOOMD-Blue website on University of Michigan Codeblue.