CDS&E: Matrix-Free Algorithms for Large-Scale Hydrodynamic Brownian Simulations

布朗动力学（BD）是一种用于大分子模拟的计算方法，在包括生物学和化学工程在内的多个领域具有无数的应用。 目前，许多BD模拟忽略了流体中颗粒之间的长程流体动力学相互作用（HI）的影响。 这种选择是为了降低这些模拟的计算成本，尽管现在认识到HI是我们在现实中观察到的大分子的大部分动力学行为的原因。 该项目将开发无矩阵算法和软件，以实现大规模BD模拟中的流体动力学相互作用建模。此外，虽然BD已经是一种钉技术，但它不能准确地模拟高体积分数系统中的HI，例如生物细胞内的拥挤环境。 一个解决方案是使用斯托克斯动力学（SD）。 SD的缺点是其计算成本高，但对于使用无矩阵方法的大型系统，也可以降低此成本。 将开发新的无基质的SD方法，其比目前可用的方法更有效，并且也适用于BD。高计算成本的HI是由于一个大的流体动力学迁移率或扩散矩阵的建设和操作。 该项目的方法是用一个基于FFT的粒子-网格Ewald求和的算子来代替这个迁移率矩阵，从而与常用方法相比，渐进地降低了计算复杂性和存储要求。 由于迁移率矩阵不再可用，这种方法需要使用无矩阵算法来计算布朗位移。这些算法将在多核节点和GPU的高性能软件库中实现。 其他研究人员可以使用该库来向其现有的BD代码添加快速HI功能和SD功能。 还将开发数据分析工具，以处理长而大的轨迹输出文件。 该软件工具将在大规模生物模拟中进行演示，以研究E.大肠杆菌类核。BD模拟是生物学中用于研究几乎所有细胞过程背后的细胞内运输和扩散机制的主要工具。 这项工作将为其他研究人员在大规模上进行更逼真的BD模拟铺平道路。 此外，这里开发的算法可以应用于机械和化学工程以及材料科学中的广泛的其他粒子模拟方法，其中布朗粒子通过长程力相互作用。 这项工作的结果将传播到应用和计算科学界。 实施新算法的软件将在开放源代码许可证下以图书馆形式发布，计算机科学和数学学生可以免费获得的粒子模拟教育材料将在网上免费提供。

Brownian dynamics (BD) is a computational method for macromolecular simulation with a myriad of applications in multiple areas including biology and chemical engineering. Currently, many BD simulations ignore the effect of the long-range hydrodynamic interactions (HI) between particles in a fluid. This choice is made to reduce the computational cost of these simulations, although it is now appreciated that HI is responsible for much of the dynamic behavior of macromolecules that we observe in reality. This project will develop matrix-free algorithms and software to enable the modeling of hydrodynamic interactions in large-scale BD simulations. Further, although BD has been a staple technique, it is not able to accurately model HI in high volume fraction systems, such as the crowded environment inside biological cells. A solution is to use Stokesian dynamics (SD). A drawback of SD is its high computational cost, but this cost can also be reduced for large systems using a matrix-free approach. New matrix-free free approaches for SD will be developed that are more efficient than those currently available, and that are also applicable to BD. The high computational cost of HI is due to the construction and manipulation of a large hydrodynamic mobility or diffusion matrix. This project's approach is to replace this mobility matrix by an operator applied with a FFT-based particle-mesh Ewald summation, thereby asymptotically reducing the computational complexity and storage requirements compared to commonly used methods. Since the mobility matrix is no longer available, this approach requires the use of matrix-free algorithms for computing Brownian displacements. The algorithms will be implemented in a high-performance software library for multicore nodes and GPUs. The library can be used by other researchers to add fast HI capabilities and also SD functionality to their existing BD codes. Data analysis tools to handle long and large trajectory output files will also be developed. The software tools will be demonstrated on large-scale biological simulations to study the mechanisms behind diffusion in the E. coli nucleoid. BD simulation is the main tool used in biology for studying intra-cellular transport and the diffusive mechanisms that lie behind almost all cellular processes. This work will pave the way for other researchers to perform much more realistic BD simulations at large scales. Furthermore, the algorithms developed here may be applied to a wide range of other particle simulation methods in mechanical and chemical engineering and materials science where Brownian particles interact through long-range forces. The results of this work will be disseminated to both the application and computational science communities. Software implementing the new algorithms will be released in library form under an open-source license, and educational materials on particle simulations, accessible to computer science and mathematics students, will be made freely available on the Web.