一种基于正弦变换的三维泊松方程并行求解算法

泊松方程的数值解法在许多物理或者工程问题上得到广泛应用,但是由于大部分三维泊松方程的离散化格式不具有明显的并行性,实际中使用整体迭代的思想,这使得计算效率和稳定性受到了限制。摒弃了传统数值解法中整体迭代的思想,结合离散正弦变换理论(DST),基于27点四阶差分格式,将三维泊松方程求解算法在算法级进行修改和并行优化,把整个求解问题转化成多个独立的问题进行求解,稳定性和并行性能得到大幅提升。对于确定的离散化形式,可以使用同一套参数解决不同的泊松方程,大大提高了编程效率。基于共享存储并行模型实现了该算法,实验结果显示,对于给出的实例,新算法具有较好的加速效果,计算结果精度误差约为10e-5,在可接受范围内,并且计算精度随着维数的升高具有一定提升。     

基于卷积金字塔的屏蔽泊松方程快速求解算法

屏蔽泊松方程在图像视频处理和图形学领域有着广泛应用,传统上一般借助离散余弦变换或快速傅里叶变换求解,计算复杂度为O(nlogn).提出了一种基于卷积金字塔的快速近似求解算法,将屏蔽泊松方程求解视为一个"大核"卷积过程,利用卷积金字塔,将"大核"卷积分解为若干个"小核"卷积,从而将计算复杂度改善至线性.实验发现,在图像无缝拼合和梯度域绘制的应用中,对于千万像素级别图像,所提算法能获得5～6倍的性能提升.进一步,屏蔽泊松方程求解也是许多图像迭代算法的中间步骤,以加权最小二乘图像光滑和基于总变差正则化的图像重建算法为例,运用所提算法,在视觉效果和均方误差上都有着很好的近似,在速度上有显著的提升.     

采用离散正弦变换的快速可控的烟雾动画

在制作可控的烟雾动画时常常需要反复调试各种参数.以得到满意的效果.为此,提出一种针对可控烟雾动画的计算框架,以节省动画师设计时间.该框架采用基于离散正弦变换的泊松方程求解算法取代传统的预处理共轭梯度求解,并改进了现有的涡旋粒子算法,使其在增强湍流细节的同时完成流体方程中对流项的求解.实验结果表明,文中的框架能得到与传统框架十分相似的结果,但计算速度可提高20倍以上.     

复用拉普拉斯算子的高效网格融合方法

针对传统基于测地线的泊松融合方法中插值旋转场与尺度场计算量大而影响交互建模的应用,提出了基于复用拉普拉斯算子的高效融合方法.该方法将几何融合、旋转场与尺度场的插值问题均转化为拉普拉斯（泊松）方程进行求解,仅需一次Cholesky分解和多次回代计算,得到融合所需的8个标量场,比起传统基于测地线的插值方法快两个数量级;随后,运用基于约束Delaunay三角化与离散极小曲面的鲁棒方法对融合边界处的网格进行优化,实现网格的高效融合.同时,再次复用拉普拉斯算子,在进行几何融合的同时,实现了纹理坐标的快速融合.该算法不仅能够处理具有复杂拓扑与多个边界模型,并获得与传统泊松融合方法相媲美的实验结果,而且显著地提高了效率,能够满足交互响应的需求.     

二维泊松问题的并行共轭梯度算法

该论文研究了利用并行共轭梯度算法求解二维泊松方程的方法,在由24台微机组成的机群上进行了实验。实验数据表明并行共轭梯度算法适用于求解二维泊松方程,它具有收敛快,可扩展性强的特点。在实验的基础上提出并验证了适用于并行共轭梯度算法的合理计算节点数的选择函数。     

大尺度图像编辑的泊松方程并行多重网格求解算法

随着获取设备的发展,大尺度、高分辫率数字图像已逐步进入人们的生活,大尺度图像的梯度域编辑显得更为重要,求解大规模未知数的泊松方程是大尺度图像梯度域编辑的关键。传统多重网格算法的迭代、约束和插值操作单独进行,内存和外存间通讯量大,算法效率低,为此提出了一种面向大尺度图像梯度域编辑的并行多重网格求解泊松方程的算法。该算法利用多重网格的迭代、约束和插值过程的内存数据访问局部性和更新相关性,构造滑动工作窗口,使迭代、约束和插值操作并行运行,提高了多重网格算法求解泊松方程的计算效率。全景图拼接实验表明,所提算法的运行效率高于超松弛迭代、高斯塞德尔迭代和传统多重网格算法。     

三维Navier-Stokes方程分步法的并行算法在异构平台上实现初探

本文选取了三维不可压缩流动方程的分步法(fractional-step method),其中动量方程使用BiCGSTAB算法进行迭代求解,而压力泊松方程使用Fourier变换法进行直接求解。本文研究该算法在集群平台上的并行算法,从区域分解入手,分析一维、两维、三维区域划分三种情况下,各并行处理器上的计算量与通讯量,根据分析结果使用两维区域分解。分析BiCGSTAB算法和泊松Fourier变换法在GPGPU异构平台上的移植方法。最后,本文分析了BiCGSTAB和泊松方程Fourier变换法两种算法在CPU集群和GPGPU异构平台上的并行性能结果。     

基于FFT的泊松方程快速求解器的硬件实现

针对传统的泊松方程求解算法执行效率低、功耗大,很难满足实际需要的缺点,设计了一种FPGA硬件平台的泊松方程快速求解器。设计采用软件与硬件结合的方式,由软件执行控制复杂、计算量较小的任务,而由硬件完成控制简单、计算量大的任务,从而达到硬件加速的目的。在FPGA平台上,独立设计的FFT协处理器可以流水和高度并行化的处理数据,提高了求解器的性能。实验结果表明,硬件实现的基于FFT的泊松方程快速求解器具有较高的计算性能和良好的可扩展性。     

基于抠像技术的图像无缝融合算法

为了更快更好的进行图像融合,在用自由融合算法对泊松图像编辑与抠像技术进行改进的基础上,实现了一套图像自由无缝融合算法。该融合算法首先利用抠像技术来提取图像中的物体边缘,然后迭代求解泊松方程,从而取得了自然的融合效果。该算法与最新的无缝融合算法相比,其优点在于:①对于前景与背景的颜色变化复杂的图像,仍可得到准确的边缘;②可以使用户能引导前景映射图的走向;③无论图像前景层中存在多少个洞,融合效果将不受任何影响。这使得该算法的应用具有相当的灵活性。对于目前的图像融合技术而言,该算法可以获得更高的融合质量、更快的处理速度以及多样的融合效果。     

基于泊松方程实现点云的表面重构

根据测量的数据点集,由梯度关系得到采样点和指示函数的积分关系,根据积分关系用划分块的方法获得点集的向量场,计算指示函数梯度场的逼近,构成泊松方程.根据泊松方程使用矩阵迭代求出近似解,采用移动立方体算法提取等值面,对所测数据点集重构出被测物体的模型,泊松方程在边界处的误差为零,因此得到的模型不会存在假的表面框.     

非线性演化方程的新Jacobi椭圆函数解

基于sinh-Gordon方程的椭圆函数解,构造新的试探解来扩展sinh-Gordon方程展开法.利用该方法研究了KdV-mKdV方程,双sine-Gordon方程和BBM方程,获得了这些方程的新Jacobi椭圆函数解.该方法也能用来求解其他数学物理中的非线性演化方程.     

A Class of Sparse Spectral Operators for Inversion of Powers of the Laplacian in N Dimensions

For inversion of the Laplacian subject to Dirichlet boundary conditions and, more generally, for the kth power of the Laplacian subject to boundary conditions on the function and its first k – 1 derivatives in the normal coordinate, there is a sparse symmetric, well-conditioned, projection of the operator that results from an expansion in associated Legendre polynomials.     

Zakai equation of nonlinear filtering with unbounded coefficients. The case of dependent noises

In this paper, we prove that the unnormalized filter associated with nonlinear filtering problems with dependent noises and a one-dimensional observations process the coefficients of which are unbounded solves a parabolic stochastic partial differential equation, the Zakai equation. The robust form of the Zakai equation is also computed.     

Numerical solution of GRLW equation using Sinc-collocation method

In this paper, we present a meshfree technique for the numerical solution of the generalized regularized long wave (GRLW) equation. This approach is based on a global collocation method using Sinc basis functions. The propagation of single solitons and the interaction of two solitary waves are used to validate the method which is found to be accurate and efficient. The three invariants of the motion are evaluated to determine the conservation properties of the method.     

Matrix algorithms for solving (in)homogeneous bound state equations

In the functional approach to quantum chromodynamics, the properties of hadronic bound states are accessible via covariant integral equations, e.g. the Bethe–Salpeter equation for mesons. In particular, one has to deal with linear, homogeneous integral equations which, in sophisticated model setups, use numerical representations of the solutions of other integral equations as part of their input. Analogously, inhomogeneous equations can be constructed to obtain off-shell information in addition to bound-state masses and other properties obtained from the covariant analogue to a wave function of the bound state. These can be solved very efficiently using well-known matrix algorithms for eigenvalues (in the homogeneous case) and the solution of linear systems (in the inhomogeneous case). We demonstrate this by solving the homogeneous and inhomogeneous Bethe–Salpeter equations and find, e.g. that for the calculation of the mass spectrum it is as efficient or even advantageous to use the inhomogeneous equation as compared to the homogeneous. This is valuable insight, in particular for the study of baryons in a three-quark setup and more involved systems.     

On the quantum master equation under feedback control

The nature of the quantum trajectories, described by stochastic master equations, may be jump-like or diffusive, depending upon different measurement processes. There are many different unravelings corresponding to different types of stochastic master equations for a given master equation. In this paper, we study the relationship between the quantum stochastic master equations and the quantum master equations in the Markovian case under feedback control. We show that the corresponding unraveling no longer exists when we further consider feedback control besides measurement. It is due to the fact that the information gained by the measurement plays an important role in the control process. The master equation governing the evolution of ensemble average cannot be restored simply by eliminating the noise term unlike the case without a control term. By establishing a fundamental limit on performance of the master equation with feedback control, we demonstrate the differences between the stochastic master equation and the master equation via theoretical proof and simulation, and show the superiority of the stochastic master equation for feedback control.     

求一类非线性偏微分方程解析解的简洁构造算法

通过引入一个变换,利用齐次平衡原理和选准一个待定函数来构造求解一类非线性偏微分方程解析解的算法.作为实例,我们将该算法应用到了mKdV方程,KdV-Burgers方程和KdV-Burgers-Kuramoto方程.借助符号计算软件Mathematica获得了这些方程的解析解.不难看出,该方法不仅简洁,而且有望进一步扩展.     

Numerical scheme based on similarity reductions for the regularized long wave equation

Numerical solution based on similarity reductions for partial differential equations used to get the numerical scheme for the regularized long wave (RLW) equation. The similarity reductions for RLW equation are obtained locally on subdomains defined by the classical three-point stencil. The ordinary differential equation, which deduced from the similarity reduction can be linearized, integrated analytically and then obtain the solution. This approch eliminates the difficulties associated with boundary conditions for the similarity reduction over the whole solution domain. Numerical results are obtained for test problem. The computed results using our scheme confirm the accuracy of our scheme.     

An analytic algorithm of Lane-Emden type equations arising in astrophysics using modified Homotopy analysis method

Lane-Emden type equation models many phenomena in mathematical physics and astrophysics. It is a nonlinear differential equation which describes the equilibrium density distribution in self-gravitating sphere of polytropic isothermal gas, has a singularity at the origin, and is of fundamental importance in the field of stellar structure, radiative cooling, modeling of clusters of galaxies. An efficient analytic algorithm is provided for Lane-Emden type equations using modified homotopy analysis method, which is different from other analytic techniques as it itself provides us with a convenient way to adjust convergence regions even without Pade technique. Some examples are given to show its validity.     

正倒向随机微分方程与一类线性二次随机最优控制问题

讨论一类正倒向随机微分方程解的存在唯一性及其对应的一类线性二次随机最优控制 问题,利用单调性方法证明了一类特殊的正倒向随机微分方程解的存在唯一性定理,利用该结果 研究一类耦合了一个倒向随机微分方程的线性随机控制系统广义最优指标随机控制问题,得到 由正倒向随机微分方程的解所表示的唯一最优控制的显式表达式,并得到精确的线性反馈及其 对应的Riccati方程.