A Scalable, Parallel Algorithm for Adaptive Smoothed Particle Hydrodynamics and the Large-Scale Simulation of Galaxy Formation in Cosmology

本提案将并行化并应用一种新的创新方法，该方法是为宇宙学中三维流体力学和引力聚类的数值模拟而开发的。我们的目标是通过模拟在膨胀的宇宙中引力、气体动力学和辐射传递的耦合效应来解释宇宙中星系和大尺度结构的形成。一个正确的解必须解决引力不稳定性的多尺度、非线性增长，导致各向异性引力坍缩、高超音速运动、强冲击、辐射冷却和许多数量级的密度对比。该方法是我们对平滑粒子流体动力学（SPH）方法（称为自适应SPH或ASPH）的大幅改进版本和著名的粒子-粒子/粒子-网格（P DSG1 3M）引力n体方法的合并，该方法适用于宇宙膨胀，物质主导的弗里德曼宇宙。ASPH方法彻底改变了标准的SPH方法，引入了时空变量、各向异性平滑和冲击跟踪的新算法，以消除由于人为粘度远离冲击而产生的虚假加热。ASPH方法大大增加了数值水动力模拟可以获得的动态范围，是目前实现上述问题研究所需分辨率的最佳方法。这个项目将开发和应用我们现有算法的一个新的并行计算机版本，这将大大减少每次计算的CPU时间。DSG1

This proposal will parallelize and apply a new and innovative method that was developed for the numerical simulation of 3D hydrodynamics and gravitational clustering in cosmology. Our goal is to explain the formation of galaxies and large-scale structure in the universe by simulating the coupled effects of gravity, gas dynamics, and radiative transfer in an expanding cosmological universe. A correct solution must resolve the multi-scale, nonlinear growth of gravitational instability, which leads to anisotropic gravitational collapse, highly supersonic motions, strong shocks, radiative cooling, and many orders of magnitude of density contrast. The method is a merger of our substantially improved version of the Smoothed Particle Hydrodynamics (SPH) Method, called Adaptive SPH, or ASPH, and the well-known Particle-Particle/Particle-Mesh(P DSG1 3M) gravitational N-body method, in a cosmologically expanding, matter-dominated Friedmann universe. The ASPH method revolutionizes the standard SPH method by introducing new algorithms for time-and space-variable, anisotropic smoothing and for shock tracking to eliminate spurious heating due to artificial viscosity away from shocks. The ASPH method has significantly increased the dynamical range that can be achieved by numerical hydrodynamical simulations, and is currently the best method to achieve the resolution required to study the problems mentioned above. This project will develop and apply a new parallel computer version of our existing algorithm, that will significantly reduce the CPU time per calculation. DSG1