求解PageRank问题改进的多分裂迭代法

引用两种加速计算PageRank的算法,分别为内外迭代法和多分裂迭代算法.从这两种方法中,得到改进的多分裂迭代方法.首先,详细介绍了算法实施过程.然后,对此算法的收敛性进行证明,并且将此算法的谱半径与原有的多分裂迭代算法的谱半径进行比较.最后,数值实验说明我们的算法的计算速度比原有的多分裂迭代法要快.     

关于实矩阵束广义特征值排序的一点注记

在[1]中对广义特征值问题的排序给出了矩阵对的推广Rube算法,它很便于实际计算.由于排序算法的应用日益广泛,我们在实际应用中发现原算法可稍作改变而大大改进其数值稳定性,现列举如下(参见[1]中Ⅱ§1.):     

快速三维散乱数据点重建算法

提出了一种快速三维散乱数据点重建算法.我们对包围盒算法进行了改进,减少了建立散乱点近邻关系所需的计算量;同时针对位置相对平坦的数据点,结合最小二乘法给出了一种计算法矢的混合方法.实例表明,该算法是有效的.     

求解M-张量方程的两种新型算法

多重线性系统在当今的工程计算和数据挖掘等领域有很多实际应用,许多问题可以转化为多重线性系统求解问题.在本文中,我们首先提出了一种新的迭代算法来求解系数张量为M-张量的多重线性系统,在此基础上又提出了一种新的改进算法,并对两种算法的收敛性进行了分析.数值算例的结果表明,本文提出的两种算法是有效的并且改进算法的迭代时间更少.     

横肋管内流场的计算方法

利用Usawa-惩罚算法,本文计算了横肋管内的流场。其主要思想是将Usawa算法与惩罚算法相结合(参见[5]),以改进Usawa算法的收敛性。该方法在步长因子的选取上比Usawa算法有更大的余地。我们利用此方法得到的结果与文献[4]中给出结果符合很好。同时,本文讨论了Usawa-惩罚算法的收敛性条件,并讨论原问题的解的存在性和唯一性。一、引言横肋管在工业上广泛用于换热管,以强化传热,这种管内流场的计算,对了解传热强化的机理,改进设计方法有重要意义,因而引起研究者的重视[1]～[4],但还缺乏较好的计算方法。我们在本文中将[5]中的Usawa算法与惩罚算法相结合,以改进Usawa算法的收敛性。同时,我们对该算法的收敛性条件作了研究,从而在理论上证明了横肋管内     

解微分方程组的改进尤拉方法的改进

对改进尤拉方法解微分方程组的方法作了改进,改进的算法与原来算法的计算量一样,但精度比较高.     

一种改进的FastICA算法

FastICA算法是一种快速独立分量分析(Independent Component Analysis:ICA)算法,但它是基于牛顿迭代方法和合理近似的一种算法,所以具有改进空间.近年来提出了许多改进的具有更高阶收敛性质的牛顿迭代方法.将一种3阶收敛的牛顿迭代方法引入ICA算法的推导中,在合理近似的基础上,提出了一种改进的两步迭代FastICA算法.与传统FastICA算法相比,提出的改进的FastICA算法一次迭代的计算量有所增加.但是,实验结果表明,新提出的改进的FastICA算法更稳健、具有更快的收敛速度.     

第一阶段原有单纯形和对偶单纯形算法的计算比较

线性最优化广泛应用于经济与管理的各个领域.在线性规划问题的求解中,如果一个初始基本可行解没有直接给出,则常采用经典的两阶段法求解.对含有"≥"不等式约束的线性规划问题,讨论了第一阶段原有单纯形法和对偶单纯形法两种算法形式,并根据第一阶段问题的特点提出了改进的对偶单纯形枢轴准则.最后,通过大规模数值试验对两种算法进行计算比较,结果表明,改进后的对偶单纯形算法在计算效率上明显优于原有单纯形算法.     

解微分方程组的Adams预测校正算法的改进方法

针对Adams预测校正算法求解微分方程组作了改进.算法的主要改进之处是在校正环节用已经计算出来的"新值"取代"旧值".最后,基于提出的改进方法和传统的Adams预测校正算法对同一微分方程组问题作了数值比较实验.数值实验结果表明改进的算法与传统的Adams预测校正算法的计算量一样,但精度大大提高.     

识别含热源瞬态热传导问题的热扩散系数

针对含有热源的瞬态热传导反问题,引入一个变换将含热源热传导问题转换为无热源热传导问题,采用改进布谷鸟算法反演热扩散系数.正问题由边界元法求解.将热扩散系数作为优化变量,以计算温度和测量温度之间的接近程度为目标函数,通过改进布谷鸟算法极小化目标函数来优化估计热扩散系数.比较共轭梯度法、布谷鸟算法和改进布谷鸟算法的反演结果.与共轭梯度法相比,改进布谷鸟算法对迭代初值不敏感;与布谷鸟算法相比,改进布谷鸟算法收敛速度更快.算例讨论了测点数量、鸟巢数量、测量误差对计算结果的影响.增加测点数量,反演结果精度降低;增加鸟巢数量,迭代次数减少;随着测量误差的增大,结果精度降低.数值算例验证了改进布谷鸟算法反演热扩散系数的准确性和有效性.     

一种守恒型间断跟踪法用于处理一维Euler方程组中的若干问题

1引言间断跟踪法(front-tracking)是数值求解双曲型守恒型方程(组)的一种重要的数值方法,其主要特点是把间断作为移动的内边界来处理,光滑区域中的数值解用计算光滑解有效的数值方法来求解,而间断的移动和间断两侧的数值解的修正要满足Rankine-Hugoniot条件．我们称这样的跟踪法为传统的间断跟踪法(见[3],[14])．本文的第二作者多年来研究设计了一种基于解的守恒性质的间断跟踪法(见[11],[12]),其最主要的特点是以解的守恒性作为跟踪的机制,而不是象传统的间断跟踪法那样利用     

Unconditionally Stable Splitting Methods for the Shallow Water Equations

The front-tracking method for hyperbolic conservation laws is combined with operator splitting to study the shallow water equations. Furthermore, the method includes adaptive grid refinement in multidimensions and shock tracking in one dimension. The front-tracking method is unconditionally stable, but for practical computations feasible CFL numbers are moderately above unity (typically between 1 and 5). The method resolves shocks sharply and is highly efficient. The numerical technique is applied to four test cases, the first being an expanding bore with rotational symmetry. The second problem addresses the question of describing the time development of two constant water levels separated by a dam that breaks instantaneously. The third problem compares the front-tracking method with an explicit analytic solution of water waves rotating over a parabolic bottom profile. Finally, we study flow over an obstacle in one dimension.     

一种守恒型间断跟踪法中对任意多个间断的移动和相互作用的处理

对一种守恒型间断跟踪法设计了一种技巧来处理任意多个间断的移动和相互作用．由此技巧我们就可以建立一个“一般的强健的”间断跟踪法．由于采用了此技巧就会使得算法在某时刻在某网格上会存在非相容性并且会产生O(1)-强度的误差．但不管怎样,这些误差在后续的计算中会被算法的守恒性所消除．还给出了几个数值例子来显示这一技巧的有效性．     

A singular-perturbed two-phase Stefan problem

The asymptotic behavior of a singular-perturbed two-phase Stefan problem due to slow diffusion in one of the two phases is investigated. In the limit, the model equations reduce to a one-phase Stefan problem. A boundary layer at the moving interface makes it necessary to use a corrected interface condition obtained from matched asymptotic expansions. The approach is validated by numerical experiments using a front-tracking method.     

Additive splitting methods for elliptic-parabolic problems

Three finite-difference splitting schemes are proposed for numerical solution of the nonlinear 3D parabolic-elliptic problem. Adaptive front-tracking and time-stepping strategies are included into the algorithms. Parallelization of the algorithms is done using the domain decomposition method. The 1D decomposition of the computational domain is used in order to obtain the optimal computational load balancing among processors and to minimize the frequency of data communications. A redistribution of the computational domain among processors is done dynamically during computations.

---

Sunto In questo lavoro vengono proposti tre diversi schemi di decomposizione alle differenze finite per la risoluzione numerica di problemi ellittico-parabolici non lineari in 3D. Negli algoritmi sono incluse strategie adattative front-tracking e time-stepping. La parallelizzazione degli algoritmi è realizzata usando il metodo della decomposizione di domini. Viene impiegata la decomposizione 1D del dominio computazionale per ottenere il bilanciameto ottimale del carico computazionale tra i processori e per minizzare la frequenza della comunicazione dei dati. Durante le computazioni, infine, viene realizzata dinamicamente la ridistribuzione dei domini computazionali.

     

The numerical solution of Stefan problems with front-tracking and smoothing methods

The numerical performance of some computer methods for heat transfer with change of phase is discussed. For one-dimensional problems the application of invariant imbedding to time-discretized problems is suggested. For some multidimensional problems an absorption of the phase transition process into the diffusion equation through the so-called enthalpy transformation is advocated. If this transformation is not applicable, a locally one-dimensional Gauss-Seidel-type front-tracking method coupled with invariant imbedding is effective.     

A front-tracking model to predict solidification macrostructures and columnar to equiaxed transitions in alloy castings

The solidified grain structure (macrostructure) of castings is predicted by process simulation using a newly extended front-tracking technique which models the growth of solid dendritic fronts through undercooled liquid during metallic alloy solidification. Such fronts are either constrained, as is the case with directed columnar growth from mould walls, or unconstrained, as is the case for multiple equiaxed growth from individual nucleating particles distributed throughout the liquid. Non-linear latent heat evolution is treated by incorporating the Scheil equation. Thermal conductivity changes with the solid fraction. A log-normal distribution of activation undercooling to initiate free growth from equiaxed nuclei is used, and the routines to deal with such growth followed by impingement of dendritic grains upon one another are verified by comparison with the results of analytical studies of simplified systems. The extensions to the model enable the predictions of equiaxed grain structure and, importantly, the columnar to equiaxed transition in inoculated alloy castings. The model is validated via comparison with experimental results. The front-tracking method is proposed as a suitable formulation for modelling alloy castings that solidify with a dendritic structure, either columnar, equiaxed, or both.     

Dynamic adaptation method in gasdynamic simulations with nonlinear heat conduction

A dynamic adaptation method is applied to gas dynamics problems with nonlinear heat conduction. The adaptation function is determined by the condition that the energy equation is quasi-stationary and the grid point distribution is quasi-uniform. The dynamic adaptation method with the adaptation function thus determined and a front-tracking technique are used to solve the model problem of a piston moving in a heat-conducting gas. It is shown that the results significantly depend on the thermal conductivity chosen. The numerical results obtained on a 40-node grid are compared with self-similar solutions to this problem.     

Mathematical modeling and numerical simulation of two-phase flows using Fourier pseudospectral and front-tracking methods: The proposition of a new method

In the present work, an extension of the Fourier Pseudospectral Method coupled with the Immersed Boudary Method for non-periodic problems (IMERSPEC) applied to numerical simulation of two-phase flow was developed. The proposed method was originally developed for single-phase, incompressible flow. Here, the method is extended to two-phase flows using the front-tracking method (IMERSPEC-FT) to model fluid-fluid interfaces. The proposed method was verified and validated through results involving spurious currents, mass conservation and numerical experiments analysis for rising bubbles. IMERSPEC-FT is shown to be a promising scheme for the two-phase computational fluid dynamics (CFD).     

A Finite-element Front-tracking Enthalpy Method for Stefan Problems

A finite-element front-tracking method, which avoids explicittreatment of the jump condition on the phase boundary, is proposed.This method is based on the discretization of a weak enthalpyformulation with isoparametric space—time finite elements.A one-dimensional two-phase problem and a two-dimensional one-phaseproblem are solved by this method. Then the method is appliedto a generalized Stefan problem—the spot-welding problemand extended to the alloy-solidification problem. Numerical experiments show that this method is convergent andunconditionally stable and that second-order convergence canbe expected if the biquadratic elements are used for one-dimensionalproblems. The effectiveness of this method is, in particular,shown in solving the last two problems.