SSE: Development of a High-Performance Parallel Gibbs Ensemble Monte Carlo Simulation Engine

在过去的十年里，随着CPU处理能力的提高和并行分子动力学代码的开发，分子模拟在原子水平上研究复杂物理现象的使用呈指数级增长，这些代码可以有效地扩展到数千个处理器上。利用CPU和GPU上的并行计算的分子动力学程序开发得相对较好，然而，有许多问题不能用这种方法来模拟。具体地说，需要模拟开放系统的问题，如在多孔材料中的吸附，需要一种允许系统中分子数量波动的替代方法。此外，在许多系统中，大的自由能垒和缓慢的扩散阻碍了标准分子动力学的使用。值得注意的例子包括预测多组分脂双层或聚合物中的相平衡。对于这些类型的问题，蒙特卡罗或混合蒙特卡罗/分子动力学模拟具有显著提高计算效率的潜力。该项目致力于开发开源的蒙特卡罗模拟引擎GOMC，该引擎能够使用低成本的图形处理单元(GPU)和多核处理器(CPU)来显著减少计算时间。这一努力将使蒙特卡罗模拟能够以更高的保真度进行，并适用于比目前标准蒙特卡罗模拟代码可访问的更大的系统，从而使领域科学家能够加速开发新材料。此外，该项目将为研究生和本科生提供蒙特卡洛模拟、在各种硬件架构上设计用于并行计算的高效算法以及软件开发方面的培训。将制作教程和其他教育材料，以支持使用GOMC向韦恩州立大学和其他大学的本科生和研究生教授分子系统的蒙特卡罗模拟。GOMC的免费分发，以及使用该软件的教程，将使其他研究小组能够快速而准确地解决重要的研究问题。该项目由高级数字基础设施办公室(OAC)、数学和物理科学局的材料研究和化学司以及工程局的化学、生物工程、环境和运输系统司(CBET)支持，将产生能够实现新的和更好的科学的软件。通过研究生和本科生的积极参与，蒙特卡罗的并行化也服务于国家的教育使命。蒙特卡罗的并行化由于算法固有的顺序性质而变得复杂，这限制了来自分子动力学的代码的重复使用，并需要开发新的方法。该团队之前的努力表明，尽管蒙特卡罗具有顺序性质，但与传统的串行CPU蒙特卡洛代码相比，图形处理器(GPU)和多核CPU可以显著减少给定计算所需的挂钟时间。这一努力导致了开源蒙特卡罗模拟引擎GPU优化蒙特卡洛(GOMC)的创建。这项工作将大大增强GOMC的功能和计算能力。这些改进包括：(1)支持基于Drude振子和阿米巴模型的可极化力场；(2)高级构型偏向运动，如协调旋转、双桥和聚集-体积偏向；(3)多分子运动；(4)混合蒙特卡罗/分子动力学模拟；(5)针对多核和GPU架构的新优化。该项目将能够在恒定化学势下模拟大系统(100,000个原子)，提供对一系列广泛问题的洞察，如聚合物、脂类和离子液体相行为、分子自组装、用于药物输送的纳米和微颗粒分散体的稳定性，以及生理相关条件下的膜融合。

The use of molecular simulation to study complex physical phenomena at the atomic level has grown exponentially over the last decade with increasing CPU power and the development of parallel molecular dynamics codes that scale efficiently over thousands of processors. Molecular dynamics codes that utilize parallel computation on CPUs and GPUs are relatively well developed, however, there are a number of problems that cannot be simulated with this methodology. Specifically, problems that require the simulation of an open system, such as adsorption in porous materials, require an alternative methodology that allows for fluctuation in the number of molecules in the system. In addition, there are a number of systems where the presence of large free energy barriers and slow diffusion preclude the use of standard molecular dynamics. Notable examples include the prediction of phase equilibria in multi-component lipid bilayers, or polymers. For these types of problems, Monte Carlo or hybrid Monte Carlo/molecular dynamics simulations have the potential to significantly improve computational efficiency. This project is focused on the development of the open-source Monte Carlo simulation engine, GOMC, which is able to use low cost graphics processing units (GPUs) and multi-core processors (CPUs) to significantly reduce computational time. This effort will enable Monte Carlo simulations to be performed with higher fidelity and for larger systems than is currently accessible with standard Monte Carlo simulation codes, enabling the accelerated development of new materials by domain scientists. In addition, this project will provide training for graduate and undergraduate students in Monte Carlo simulation, design of efficient algorithms for parallel computation on a variety of hardware architectures, and software development. Tutorials and other educational materials will be created to support the use of GOMC for teaching Monte Carlo simulation of molecular systems to students in undergraduate and graduate courses at Wayne State as well as other universities. The free distribution of GOMC, along with the tutorials for using the software, will enable other research groups to solve important research problems quickly and accurately. This project, supported by the Office of Advanced Cyberinfrastructure (OAC), and the divisions of Material Research and Chemistry in the Directorate of Mathematical and Physical Sciences, and the Division of Chemical, Bioengineering, Environmental and Transport Systems (CBET) in the Directorate of Engneering, will result in software that enables new and better science. It also serves the educational mission of the National, through its active involvement of graduate and undergraduate students.Parallelization of Monte Carlo is complicated by the inherently sequential nature of the algorithm, which limits the reuse of code from molecular dynamics, and necessitates the development of new approaches. The team's previous efforts have shown that despite the sequential nature of Monte Carlo, graphics processors (GPU) and multi-core CPUs can be used to yield significant reductions in wall-clock time required for a given calculation compared to a traditional serial CPU Monte Carlo code. This effort led to the creation of the open-source Monte Carlo simulation engine GPU Optimized Monte Carlo (GOMC). This work will add significant functional and computational enhancements to be added to GOMC. These enhancements include: (1) support for polarizable force fields based on the Drude oscillator and AMOEBA models, (2) advanced configurational bias moves, such as concerted rotation, double bridging, and aggregation-volume bias (3) multi-molecule moves (4) hybrid Monte Carlo/molecular dynamics simulations (5) new optimizations for multi-core and GPU architectures. The project will enable the simulation of large systems (100,000 atoms) at constant chemical potential, providing insight into a broad array of problems such as polymer, lipid and ionic liquid phase behavior, molecular self-assembly, the stabilization of nano and micro particle dispersions for drug delivery, and membrane fusion under physiologically relevant conditions.