基于格子Boltzmann方法的平板射流大涡模拟

应用多GPU技术,将格子Boltzmann方法与大涡模拟相结合(LBM-LES),使用1.12×10⁸网格,对雷诺数Re=4 000,倾斜角α=30°,吹风比M=0.5工况下的平板单孔射流进行了大规模高性能数值模拟研究.合理的定性与定量结果验证了LBM-LES模拟平板射流的有效性与可行性.使用上亿的计算网格捕捉了精细的湍流拟序结构,有利于主流与射流之间的掺混机理研究.此外,使用6个K20M GPU并行计算,模拟了71 680 LBM时间步长,仅耗时15 402秒,计算性能达到521.24MLUPS,即每秒更新5.212 4×10⁸个网格点的数据.     

激波与火焰面相互作用数值模拟的GPU加速

为考察计算机图形处理器（GPU）在计算流体力学中的计算能力,采用基于CPU/GPU异构并行模式的方法对激波与火焰界面相互作用的典型可压缩反应流进行数值模拟,优化并行方案,考察不同网格精度对计算结果和计算加速性能的影响.结果表明,和传统的基于信息传递的MPI 8线程并行计算相比,GPU并行模拟结果与MPI并行模拟结果相同;两种计算方法的计算时间均随网格数量的增加呈线性增长趋势,但GPU的计算时间比MPI明显降低.当网格数量较小时（1.6×10⁴）,GPU计算得到的单个时间步长平均时间的加速比为8.6;随着网格数量的增加,GPU的加速比有所下降,但对较大规模的网格数量（4.2×10⁶）,GPU的加速比仍可达到5.9.基于GPU的异构并行加速算法为可压缩反应流的高分辨率大规模计算提供了较好的解决途径.     

结构材料辐照损伤的分子动力学程序GPU并行化及优化

基于NIVIDIA公司的CUDA架构对结构材料辐照损伤的分子动力学程序在单个GPU上进行并行化,并对影响程序运行效率的相关因素进行分析和测试.经过一系列优化,当粒子数为两百万时,对比单CPU的执行时间,优化后的GPU程序其双精度加速比可达112倍,单精度加速比达到了三百倍,为后续扩展多GPU结构材料辐照损伤的分子动力学程序奠定基础.     

采用GPU加速复杂结构角系数计算的方法研究

采用离散传递法计算目标的红外辐射特征首先要计算角系数,对于复杂目标,角系数计算会耗费大量时间。针对角系数计算中遮挡判断问题,提出了一种不需要求解微元面中心点连线与遮挡面交点的矢量法。开发了基于矢量法的GPU并行程序,并使用一个有理论解的模型验证了程序的正确性,最后针对航空发动机排气系统复杂模型,在一台配置有448个核心的GPU处理器和主频为3.4GHz CPU处理器的工作站上,评估了基于矢量法的GPU程序相对单核CPU程序的加速效果。结果表明;在微元面数量为13670时,矢量法可获得约73倍的加速比,是采用线面求交法获得的加速比的1.4倍。使用离散传递法计算航空发动机排气系统的红外辐射特征时,仅将角系数模块使用GPU并行计算,可减少45%的计算时间。     

基于GPU的液晶自适应光学波前重构计算

利用GPU进行液晶自适应光学波前重构的加速计算.介绍了液晶自适应光学的Zernike模式波前重构算法,详细论述了GPU的通用架构和GPU实现波前重构的方法,给出了GPU与CPU的实验对比结果.结果表明,GPU计算波前重构不但可以准确无误地计算出液晶波前校正器的灰度级分布,计算速度更是传统CPU波前计算的几十倍.     

节块内嵌离散纵标(SN)方法的算法研究及程序开发

介绍节块内嵌离散纵标(S_N)方法求解三维堆芯中子输运/扩散方程的算法框架.在基于扩散理论的三维粗网节块展开方法(NEM)的算法体系中,用基于输运理论的径向二维细网节块离散纵标方法(NDOM)的内迭代过程,替代节块展开方法(NEM)内迭代的径向求解过程.该算法充分考虑了核电厂反应堆堆芯的三维结构特点,另一方面,也充分利用了已经成熟的三维粗网节块展开方法(NEM)和二维离散纵标方法(S_N)的研究成果,同时有效避免了利用离散纵标方法(S_N)求解三维中子输运方程所面临的计算内存和计算时间的瓶颈问题.编制开发二维多群节块离散纵标方法(NDOM)模块程序NS_NM和三维多群节块展开方法(NEM)模块程序MGNEM,并以此为基础编制开发节块内嵌S_N方法的模块程序HANWIND;其中,NS_NM为HANWIND求解两维问题的功能模块.针对OECD/NEA-2D C5G7MOX基准问题以及两环路核电厂三维堆芯的数值验算结果表明,节块内嵌S_N方法的算法开发及程序编制有效、切实可行.     

粒子输运方程的确定论计算方法

针对实际应用中辐射和中子输运数值模拟,讨论球一维和柱二维几何粒子输运方程确定论计算方法的研究现状,包括离散纵标、球谐函数、迭代加速、并行计算等方法.重点讨论输运计算方法所取得的若干研究进展,包括离散纵标求积组、自适应时间离散格式、本征值迭代求解方法、简化球谐函数方法、修正的子网格隅角平衡方法、灰体综合加速方法、迭代初值选取方法、输运与扩散耦合方法、基于预估校正的并行格式等.简要介绍了相关输运计算程序的研制情况,并分析输运计算方法存在的难点,提出待开展研究的内容.     

基于格子Boltzmann方法的多孔介质流动模拟GPU加速

利用NVIDIA CUDA平台,在GPU上结合稀疏存贮算法实现基于格子Boltzmann方法的孔隙尺度多孔介质流动模拟加速,测试该算法相对基本算法的性能.比较该算法在不同GPU上使用LBGK和MRT两种碰撞模型及单、双精度计算时的性能差异.测试结果表明在GPU环境下采用稀疏存贮算法相对基本算法能大幅提高计算速度并节省显存,相对于串行CPU程序加速比达到两个量级.使用较新构架的GPU时,MRT和LBGK碰撞模型在单、双浮点数精度下计算速度相同.而在较上一代的GPU上,计算精度对MRT碰撞模型计算速度影响较大.     

求解二维三温辐射扩散方程组的一种代数两层迭代方法

在二维三温辐射扩散方程离散代数方程组的求解中,由于光子、电子和离子温度之间存在耦合关系,而且三个温度在同种介质中有不同的扩散性质,使得经典的代数多重网格(AMG)方法难以直接应用.基于特殊粗化策略,在粗网格层解除了这种耦合关系,得到一种代数两层网格方法,而粗网格方程由经典AMG方法求解.将这一算法具体应用于JFNK(Jacobian自由的Newton-Krylov)框架中预处理方程的求解,并基于该框架求解二维三温辐射扩散方程组.数值结果显示了算法的可扩展性和健壮性.     

NeuDATool：支持GPU硬件加速和计算机集群跨节点并行的开源中子散射数据分析软件

实验势精修是20世纪80年代英国散裂中子源无定型材料组开发的用于分析中子散射实验数据的软件. 实验势精修的目标是根据中子散射数据重建样品的三维原子结构. 在过去的几十年,实验势精修被广泛用于中子散射实验数据分析,为实验用户提供了可靠的分析结果. 但是实验势精修是基于共享内存并行计算(OpenMP)的Fortran程序,不支持计算机服务器集群跨节点并行加速和GPU加速;这限制了它的分析速度. 随着计算机服务器集群的广泛建设和GPU加速技术的普遍使用,有必要重新编写EPSR程序以提高运算速度. 本文使用面向对象的C++语言,开发了一套实现EPSR算法的开源软件包NeuDATool;软件通过MPI和CUDA C实现了计算机集群跨节点并行和GPU加速. 使用液态水和玻璃态二氧化硅的中子散射实验数据对软件进行了测试. 测试显示软件可以正确重建出样品的三维原子结构;并且模拟体系达到10万原子以上时,使用GPU加速可以比串行的CPU算法提高400倍以上的模拟速度. NeuDATool为中子实验用户尤其是对熟悉C++编程并希望定义特殊分析算法的实验科学家提供了一种新的选择.     

Application of the SCVT Orientation Grid to the Simulation of CW EPR Powder Spectra

The orientation grid obtained using the spherical centroidal Voronoi tessellation (SCVT) method is assessed for the simulation of continuous-wave electron paramagnetic resonance (CW EPR) powder spectra. The SCVT spherical code presents less distorted Voronoi cells associated to the points of the grid compared to the regular Igloo, SOPHE and EasySpin grids and similar distortions of the cells to the irregular Repulsion grid. The degree of distribution of the SCVT cells’ areas is smaller than for SOPHE, but higher than for the Igloo and Repulsion grids. All compared spherical codes have nearly the same electrostatic potential energy. Simulated CW EPR powder spectra for rhombic systems of spin S = 1/2 are similar when using the energy minimization SCVT and Repulsion spherical codes and show less intense ripples than the spectra calculated with the SOPHE and Igloo grids, but more intense than those obtained using the EasySpin grid. The classical projection method and a new modified form involving the spectral intensities at the orientations of the grid have been tested for the attenuation of the simulation noise, for grids of relatively reduced size.     

Arrays of Charged (±2π)-Twist Disclinations in Ferroelectric Liquid Crystals

The notion of an electrostatic charge of (±2)-twist disclinations is used to approximate the evaluation of the electrostatic interaction energy among disclinations forming arrays in finite samples of ferroelectric chiral smectic C liquid crystals. Screening effects of free charges in a material surrounding the disclination are taken into account by introducing a phenomenological depolarisation factor.The electrostatic interaction energy is important in chiral smectic C materials with high values of the spontaneous polarisation when screening effects of free charges are small. Then the electrostatic interaction leads to elimination of disclinations from the sample. When there is a high concentration of free charges in the sample (smaller value of depolarisation factor), the electrostatic interaction energy is of the order of the elastic interaction energy of disclinations what influences the equilibrium of disclination arrays in the sample. Two disclination configurations are considered. In the Brunet-Williams configuration the disclinations of opposite topological charge have also the opposite electrostatic charge so their attraction is augmented. This attraction can be balanced by the helical structure in the central part of the sample when the sample thickness is rather high.On the contrary, in the Glogarová-Pavel configuration the disclinations of opposite topological charge have the electrostatic charge of the same sign. The equilibrium in this configuration is either a balance of elastic attraction and electrostatic repulsion if elastic and Coulomb forces are of the same order or it is governed by the value of the anchoring energy when electrostatic interaction prevails over the elastic one.     

Effect of the size of macroparticles on their electrostatic interaction in a plasma

The electrostatic interaction of two spherical macroparticles in a plasma has been considered. Primary attention has been focused on investigating the electrostatic interaction at short distances where polarization effects of the surface charge of finite-size macroparticles begin to play a dominant role. The first part of this study is devoted to the interaction of a point charge with a charged conducting sphere in an equilibrium plasma. It has been shown that the presence of a plasma in the system leads to a decrease in the potential barrier when two like-charged macroparticles approach each other and that this decrease proves to be the most significant in the case where the macroparticle radius is comparable to the Debye screening length. The second part of this study is concerned with the interaction of two conducting spheres in the bispherical system of the coordinates under the assumption that the charges of the conducting spheres are constant and under the assumption that the surface potentials of the spheres are constant. The latter case is closer to the physics of electrostatic interaction of two macroparticles in a plasma medium where the electrostatic potential of their surface is determined by the floating potential of the plasma. It has been demonstrated that the interaction potentials in these two cases are substantially different from each other and that, at constant macroparticle charges, the energy of the electrostatic field is an interaction potential, but, in the case of macroparticles with constant surface potentials, which are independent of the interparticle distance, the energy of the electrostatic field is not an interaction potential. In the latter case, account must be taken of the work done by external sources on the macroparticle potentials to maintain them constant. The form of the interaction potential has been established in this case from the analysis of the interaction force in terms of the Maxwell tension tensor. In the third part of this study, the interaction of two macroparticles has been considered in the spherical system of coordinates and analytical expressions for the interaction potentials have been derived for both the case of constant macroparticle charges and the case of constant surface potentials of the macroparticles.     

Connecting shock velocities to electron-injection mechanisms

Electrons can be accelerated by their interaction with nonlinearly saturated electrostatic waves up to speeds with which they can undergo diffusive acceleration across supernova remnant shocks. Here, we model this wave-electron interaction by particle-in-cell and Vlasov simulations. We find that the lifetime of the saturated wave is considerably longer in the Vlasov simulation, due to differences in how these simulation methods approximate the plasma. Electron surfing acceleration which requires a stable saturated wave may thus be more important for electron acceleration at shocks than previously thought. For beam speeds above a critical value, which we estimate here, both simulation codes exclude surfing acceleration due to a rapid wave collapse.     

一种求解三维热辐射输运方程的整体预处理迭代方法及并行计算

为三维灰体热辐射输运方程的隐式离散纵标方法发展一个整体预处理迭代方法并研制并行程序.该方法采用组装线性代数方程组策略,同时求出所有离散方向上的辐射强度.借助预处理的Krylov子空间迭代法,避免复杂网格上扫描算法可能遇到的死锁问题,能够提高健壮性和计算效率.空间离散上采用一阶迎风有限体积格式.数值实验测试变形六面体网格...     

Comparison of the PIC model and the Lie algebraic metnod in the simulation of intense continuous beam transport

Both the PIC(Particle-In-Cell) model and the Lie algebraic method can be used to simulate the transport of intense continuous beams.The PIC model is to calculate the space charge field,which is blended into the external field,and then simulate the trajectories of particles in the total field;the Lie algebraic method is to simulate the intense continuous beam transport with transport matrixes.Two simulation codes based on the two methods are developed respectively,and the simulated results of transport in a set of electrostatic lenses are compared.It is found that the results from the two codes are in agreement with each other.and both approaches have their own merits.     

Maximum Speedup in Quantum Search: O(1) Running Time

We consider the running time of the generalized quantum search Hamiltonian. We provide the surprising result that the maximum speedup of quantum search in the generalized Hamiltonian is an O(1) running time regardless of the number of total states. This seems to violate the proof of Zalka that the quadratic speedup is optimal in quantum search. However the argument of Giovannetti et al. that a quantum speedup comes from the interaction between subsystems (or, equivalently entanglement) (and is concerned with the Margolus and Levitin theorem) supports our result.     

Inclusion of the Full Electromagnetic Interaction in Correlations and Physical Quantities

Density-density correlations in a homogeneous, isotropic, classical plasma are calculated with inclusion of the full electromagnetic interaction. This leads to isotropic homogeneous field limits, contrary to the electrostatic approximation, that is consequently not valid for small wavenumbers. Application of the theory of generalized functions, combined with Kramer Kronig relations makes it possible to deal with the integrals that occur in the transverse part, provided a nonrelativistic approximation has been applied. As a side result Kubo's formula, applied to the electric conductivity, and generally derived under restriction to the electrostatic approximation, reappears for the case that the full electromagnetic interaction is included.     

A low order flow/acoustics interaction method for the prediction of sound propagation using 3D adaptive hybrid grids

A low-order flow/acoustics interaction method for the prediction of sound propagation and diffraction in unsteady subsonic compressible flow using adaptive 3-D hybrid grids is investigated. The total field is decomposed into the flow field described by the Euler equations, and the acoustics part described by the Nonlinear Perturbation Equations. The method is shown capable of predicting monopole sound propagation, while employment of acoustics-guided adapted grid refinement improves the accuracy of capturing the acoustic field. Interaction of sound with solid boundaries is also examined in terms of reflection, and diffraction. Sound propagation through an unsteady flow field is examined using static and dynamic flow/acoustics coupling demonstrating the importance of the latter.