三维多物质弹塑性流体动力学Euler方法的并行算法研究及程序测试

 并行计算是解决爆炸与冲击问题大规模数值模拟最有效的手段之一。针对Euler方法并行程序设计的复杂性,阐述了三维多物质弹塑性流体动力学程序MMIC-3D并行设计的总体策略,基于消息传递接口(MPI)设计出相应的PMMIC-3D并行程序,并提出了一套实用的程序测试方案。结合聚能射流形成过程的数值模拟算例,在八节点的集群上测试了加速比、并行效率及可扩放性,分析了影响并行性能的因素。     

基于大规模并行计算的三维多群中子扩散方程有限差分方法

三维多群中子扩散方程的精确、高效求解是核动力堆芯设计及燃料管理的基础。应用有限差分方法求解该方程具有简便、精确、成熟的优点;然而,该方法的计算量和存储量均较大,极大地限制了它的计算规模和应用范围。本文基于大规模并行计算,研究三维多群中子扩散方程有限差分方法:采用中心有限差分格式离散中子扩散方程;基于MPI并行编程模型,采用空间区域分解的方式实现大规模并行计算;采用多群多区域耦合PGMRES算法进行并行加速。在集群服务器上开发了ParaFiDi程序,并采用IAEA3D,PHWR等多个基准题对该程序进行验证。数值结果表明,ParaFiDi程序具有较高的计算精度和计算效率。     

二维等离子体模拟粒子云网格方法的并行计算与性能分析

在共享和分布式存储两种并行计算环境下,组织了二维等离子体模拟粒子云网格法程序（２ＤＣＩＣ）的并行计算。首先,通过分析串行算法和２ＤＣＩＣ程序的单机执行特征,设计了并行算法和实现策略;然后,基于共享存储和消息传递现任中并行程序设计方式,在四种共享和颁式存储并行机上,组织了具体的并行计算,对性能结果进行了的比较分析,获得也若干重要的结论。     

三维激光烧蚀瑞利-泰勒不稳定性并行计算

给出了用三维激光烧蚀不稳定性程序求解的物理方程组,简要介绍了所用的数值方法.根据所模拟问题的物理性质和数值模拟方法探索了在多处理机上实行并行计算的技术和方法,并行效率达75%,在此基础上进行了三维多模激光烧蚀不稳定性的数值模拟,深化了对有关物理规律的认识.     

油资源运移聚集并行数值模拟和分析

多层油资源运移聚集并行数值模拟,其功能是重建油气盆地的运移聚集演化史,对于油资源的评价,确度油藏位置,估计油藏贮量均具有重要的价值.从生产实际出发,提出数学模型,构造大规模精细并行计算的耦合算子分裂隐式迭代格式,设计了并行计算程序和并行计算的信息传递和交替方向网格剖分方法.并对不同的网格步长进行了并行计算和分析.对滩海地区进行精细数值模拟,计算结果在油田位置等方面和实际情况相吻合.对模型问题(非线型性耦合问题)进行数值分析,得到最佳阶L²误差估计,解决了这一计算渗流力学和石油地质的困难问题.     

并行自适应结构网格应用支撑软件框架研制进展

JASMIN框架全称为“并行自适应结构网格应用支撑软件框架（Jparallel Adaptive Structured Mesh Applications Infrastructure）”。该框架面向科学与工程计算的（自适应）结构网格应用,支持应用领域专家在无需了解并行计算和共性数值算法细节的前提下,通过编写串行数值计算程序,     

流动和传热问题的分区并行计算

针对交错网格下的SIMPLE数值算法实施了分区并行计算方法,在小型局域网下实现了流动和传热问题的并行数值计算.对两个经典的流动和传热问题的数值模拟实验表明,所建立的并行计算环境和分区并行算法能够得到正确的和收敛的数值结果.但与串行计算结果相比,并行计算误差明显大于串行计算误差.对并行算法做出的性能分析表明,所给出的并行算法得到了明显的加速效率.随着计算规模的增大,加速比和并行效率提高更显著.     

LARED集成程序辐射输运模拟的性能优化

分析定位高置信度辐射传输数值模拟效率瓶颈,针对物理模型和数值离散格式的特点,优化求解算法和程序设计,改进激光聚变二维总体LARED集成程序,提高输运建模模拟黑腔靶实验模型的计算效率：在相同的并行计算条件下,辐射输运方程离散求解计算效率提高2倍以上,全过程整体模拟计算时间减半.     

Ising模型的并行计算

以Ising模型为例,介绍有关格点系统的Monte Carlo数值模拟并行算法的设计和编程,并给出在本组建造的PC集群式高性能并行计算系统上的测量结果.本文的结果对格点量子色动力学的大规模数值模拟研究有一定的参考价值     

基于时域多分辨率的非球形气溶胶散射并行计算模型设计及验证EI北大核心CSCD

为解决串行时域多分辨率(MRTD)散射模型运行时间长和内存消耗大的问题,基于消息传递接口(MPI)技术设计了一种非球形气溶胶散射并行计算模型。介绍了MRTD散射模型的基本框架和2种并行数据通信方案,并基于MPI重复非阻塞通信技术实现了MRTD散射模型的并行化设计;搭建了网络并行计算平台,实现了模型的并行化计算。将MRTD散射模型与Mie散射模型、T矩阵法进行了对比,验证了并行MRTD散射模型的计算准确性。结果表明,MRTD模型可较准确地模拟非球形粒子散射特性,并行计算技术可显著提高计算效率;电磁场分量同时交换的并行设计方案的计算效率略高于仅交换磁场分量的方案;通过增加中央处理器核数,程序的并行加速比随之增大,但单核运行效率却略有降低。随着粒子尺度参数的增大,单核计算效率随之增加,复折射率的改变并不会显著影响并行计算效率。     

基于网格片的氧碘化学激光器多块并行数值模拟

为实现氧碘化学激光器(COIL)喷管流场大规模数值模拟,采用VICON程序中的基本方程与数值算法,应用JASMIN框架中的基本数据结构——网格片,以及多块结构网格拼接并行算法,发展了三维多块COIL并行模拟程序。数值实验结果展示了该并行模拟程序的正确性及可扩展性,并在2048个处理器核上模拟450万网格单元算例,加速比超过420。     

二维多群辐射输运程序LARED-R-1的并行化

利用有向图描述数据依赖关系,应用已有的并行流水线通量扫描算法,实现基于非协调网格的二维辐射输运程序LARED-R-1的并行化.同时,采用消息缓冲技术提高并行程序的性能.经测试,对于典型的问题规模(100群、3800个网格单元、40个方向),在某并行机的64个和128个处理器上,并行程序分别获得80%和53%的并行效率.     

基于静态双重区域分解的两种接触并行算法

CHAP3D是北京应用物理与计算数学研究所自主研发的Lagrange通用弹塑性流体力学分析程序.文章介绍了在CHAP3D程序中使用的、针对多处理器集群的、基于静态双重区域分解的两种接触并行算法.第一种是分配单个完整接触面的接触并行算法，此算法将一对完整的接触面分配到一个处理器上，并建立计算域与接触域的通信关系.此接触并行算法的优点是简单，在具有接触面的处理器上可以直接使用串行的接触搜索算法和接触力耦合计算算法.另一种是主面剖分区域分解的接触并行算法，此算法将所有接触面的主面区域分解到所有处理器上.须建立计算域与接触域以及接触域内各处理器间的两种通信关系.该接触并行算法是一个负载平衡的并行算法，具有很好的并行效率和可扩展性.数值算例显示，这两种接触并行算法都能够很好地模拟多种不同类型的接触问题.      

CFETR低杂波电流驱动数值模拟研究

利用射线追踪和福克-普朗克方程,研究了CFETR上驱动电流和功率沉积对低杂波注入位置和耦合波谱的依赖关系,讨论了边界非线性效应对电流驱动与功率沉积的影响,给出了低杂波平行折射率和低杂波注入位置的一个优化值。初步数值计算表明：边界非线性效应会导致驱动电流降低;不考虑低杂波非线性效应时,驱动电流的差异约为6%左右,小于考虑该效应时的差异(~25%)。     

Multi-level Monte Carlo finite volume methods for nonlinear systems of conservation laws in multi-dimensions

We extend the multi-level Monte Carlo (MLMC) in order to quantify uncertainty in the solutions of multi-dimensional hyperbolic systems of conservation laws with uncertain initial data. The algorithm is presented and several issues arising in the massively parallel numerical implementation are addressed. In particular, we present a novel load balancing procedure that ensures scalability of the MLMC algorithm on massively parallel hardware. A new code is described and applied to simulate uncertain solutions of the Euler equations and ideal magnetohydrodynamics (MHD) equations. Numerical experiments showing the robustness, efficiency and scalability of the proposed algorithm are presented.     

Experimental validation of numerical simulations for an acoustic liner in grazing flow: Self-noise and added drag

A coordinated experimental and numerical simulation effort is carried out to improve our understanding of the physics of acoustic liners in a grazing flow as well our computational aeroacoustics (CAA) method prediction capability. A numerical simulation code based on advanced CAA methods is developed. In a parallel effort, experiments are performed using the Grazing Flow Impedance Tube at the NASA Langley Research Center. In the experiment, a liner is installed in the upper wall of a rectangular flow duct with a 2 in. by 2.5 in. cross section. Spatial distribution of sound pressure levels and relative phases are measured on the wall opposite the liner in the presence of a Mach 0.3 grazing flow. The computer code is validated by comparing computed results with experimental measurements. Good agreements are found. The numerical simulation code is then used to investigate the physical properties of the acoustic liner. It is shown that an acoustic liner can produce self-noise in the presence of a grazing flow and that a feedback acoustic resonance mechanism is responsible for the generation of this liner self-noise. In addition, the same mechanism also creates additional liner drag. An estimate, based on numerical simulation data, indicates that for a resonant liner with a 10 percent open area ratio, the drag increase would be about 4 percent of the turbulent boundary layer drag over a flat wall.     

MC程序并行设计及提高加速比措施

MC程序的并行设计涉及算法及模块划分,它直接关系到并行加速效率的高低,中子－γ耦合输运蒙特卡罗程序MCNP经过行改造,实现了PVM和MPI两种系统下的并行化,由于作了模块化设计,并行加速效率极佳,PVM版和MPI版并行程序在多个处理器下的加速比均呈线性增长,相比PVM,MPI的适应性列强,多数情况下其效率高于OPVM,并行MCNP程序的计算结果可靠,MPI并行程序在16、32和64个处理器上的并行效率分别达到99％、97％和89％。     

Numerical simulation of the hydrodynamic instability experiments and flow mixing

Based on the numerical methods of volume of fluid (VOF) and piecewise parabolic method (PPM) and parallel circumstance of Message Passing Interface (MPI), a parallel multi-viscosity-fluid hydrodynamic code MVPPM (Multi-Viscosity-Fluid Piecewise Parabolic Method) is developed and performed to study the hydrodynamic instability and flow mixing. Firstly, the MVPPM code is verified and validated by simulating three instability cases: The first one is a Riemann problem of viscous flow on the shock tube; the second one is the hydrodynamic instability and mixing of gaseous flows under re-shocks; the third one is a half height experiment of interfacial instability, which is conducted on the AWE’s shock tube. By comparing the numerical results with experimental data, good agreement is achieved. Then the MVPPM code is applied to simulate the two cases of the interfacial instabilities of jelly models accelerated by explosion products of a gaseous explosive mixture (GEM), which are adopted in our experiments. The first is implosive dynamic interfacial instability of cylindrical symmetry and mixing. The evolving process of inner and outer interfaces, and the late distribution of mixing mass caused by Rayleigh-Taylor (RT) instability in the center of different radius are given. The second is jelly layer experiment which is initialized with one periodic perturbation with different amplitude and wave length. It reveals the complex processes of evolution of interface, and presents the displacement of front face of jelly layer, bubble head and top of spike relative to initial equilibrium position vs. time. The numerical results are in excellent agreement with that experimental images, and show that the amplitude of initial perturbations affects the evolvement of fluid mixing zone (FMZ) growth rate extremely, especially at late times.     

Lattice-Boltzmann Simulation of Particle Suspensions in Shear Flow

Inclusion of short-range particle–particle interactions for increased numerical stability in a lattice-Boltzmann code for particle-fluid suspensions, and handling of the particle phase for an effective implementation of the code for parallel computing, are discussed and formulated. In order to better understand the origin of the shear-thickening behavior observed in real suspensions, two simplified cases are considered with the code thus developed. A chain-like cluster of suspended particles is shown to increase the momentum transfer in a shear flow between channel walls, and thereby the effective viscosity of the suspension in comparison with random configurations of particles. A single suspended particle is also shown to increase the effective viscosity under shear flow of this simple suspension for particle Reynolds numbers above unity, due to inertial effects that change the flow configuration around the particle. These mechanisms are expected to carry over to large-scale particle-fluid suspensions.