MIMD计算机上的一个稳定并行算法

在MIMD计算机上解稠密线性方程组的问题,见[1]与[2].这两篇文章研究了基于高斯消去法和G-J消去法以及Givens变换法的实用并行算法,推得这三个并行算法的效率分别为2/3,4/7和4/9,且以并行高斯消去法为最佳.     

Computing generalized inverses of a rational matrix and applications

In this paper we investigate symbolic implementation of two modifications of the Leverrier-Faddeev algorithm, which are applicable in computation of the Moore-Penrose and the Drazin inverse of rational matrices. We introduce an algorithm for computation of the Drazin inverse of rational matrices. This algorithm represents an extension of the papers [11] and [14]. and a continuation of the papers [15, 16]. The symbolic implementation of these algorithms in the package mathEmatica is developed. A few matrix equations are solved by means of the Drazin inverse and the Moore-Penrose inverse of rational matrices.     

On systolic arrays for updating the Cholesky factorization

We have considered the systolic implementation of several methods for updating the Cholesky factorization. For positive rank-k changes there are simple one-pass arrays that implement algorithms based on elimination and plane rotations. In the case of negative rank-one changes, we do not feel that the standard algorithm [2] has a practical implementation. We have introduced a new algorithm for the case of a negative rank-k change and provided an attractive two-pass systolic implementation.     

On parallel algorithms for semidiscretized parabolic partial differential equations based on subdiagonal Padé approximations

It is well known that methods for solving semidiscretized parabolic partial differential equations based on the second-order diagonal [1/1] Padé approximation (the Crank–Nicolson or trapezoidal method) can produce poor numerical results when a time discretization is imposed with steps that are “too large” relative to the spatial discretization. A monotonicity property is established for all diagonal Padé approximants from which it is shown that corresponding higher-order methods suffer a similar time step restriction as the [1/1] Padé. Next, various high-order methods based on subdiagonal Padé approximations are presented which, through a partial fraction expansion, are no more complicated to implement than the first-order implicit Euler method based on the [0/1] Padé approximation; moreover, the resulting algorithms are free of a time step restriction intrinsic to those based on diagonal Padé approximations. Numerical results confirm this when various test problems from the literature are implemented on a Multiple Instruction Multiple Data (MIMD) machine such as an Alliant FX/8. © 1993 John Wiley & Sons, Inc.     

Solving Hamiltonian systems arising from ODE eigenproblems

This paper presents an algorithm for solving a linear Hamiltonian system arising in the study of certain ODE eigenproblems. The method follows the phase angles of an associated unitary matrix, which are essential for correct indexing of the eigenvalues of the ODE. Compared to the netlib code SL11F [11] the new method has the property that on many important problems – in particular, on matrix–vector Schrödinger equations – the cost of the integration is bounded independently of the eigenparameter λ. This allows large eigenvalues to be found much more efficiently. Numerical results show that our implementation of the new algorithm is substantially faster than the netlib code SL11F.     

A communication-less parallel algorithm for tridiagonal Toeplitz systems

Diagonally dominant tridiagonal Toeplitz systems of linear equations arise in many application areas and have been well studied in the past. Modern interest in numerical linear algebra is often focusing on solving classic problems in parallel. In McNally [Fast parallel algorithms for tri-diagonal symmetric Toeplitz systems, MCS Thesis, University of New Brunswick, Saint John, 1999], an m processor Split & Correct algorithm was presented for approximating the solution to a symmetric tridiagonal Toeplitz linear system of equations. Nemani [Perturbation methods for circulant-banded systems and their parallel implementation, Ph.D. Thesis, University of New Brunswick, Saint John, 2001] and McNally (2003) adapted the works of Rojo [A new method for solving symmetric circulant tri-diagonal system of linear equations, Comput. Math. Appl. 20 (1990) 61–67], Yan and Chung [A fast algorithm for solving special tri-diagonal systems, Computing 52 (1994) 203–211] and McNally et al. [A split-correct parallel algorithm for solving tri-diagonal symmetric Toeplitz systems, Internat. J. Comput. Math. 75 (2000) 303–313] to the non-symmetric case. In this paper we present relevant background from these methods and then introduce an m processor scalable communication-less approximation algorithm for solving a diagonally dominant tridiagonal Toeplitz system of linear equations.     

An optimal schedule for Gaussian elimination on an MIMD architecture

In this paper, we propose an algorithm for allocating the tasks of the well known Gaussian Elimination Algorithm on an MIMD architecture and prove that the schedule is optimal in order of magnitude, up to a polylog factor.     

A Parallel Preconditioned Iterative Realization of the Panel Method in 3D

The parallel version of precondition iterative techniques is developed for matrices arising from the panel boundary element method for three-dimensional simple connected domains with Dirichlet boundary conditions. Results were obtained on an nCube-2 parallel computer showing that preconditioned iterative methods are very well suited also in three-dimensional cases for implementation on an MIMD computer and that they are much more efficient than usual direct solution techniques.     

On parallel asynchronous high-order solutions of parabolic PDEs

In achieving significant speed-up on parallel machines, a major obstacle is the overhead associated with synchronizing the concurrent processes. This paper presents high-orderparallel asynchronous schemes, which are schemes that are specifically designed to minimize the associated synchronization overhead of a parallel machine in solving parabolic PDEs. They are asynchronous in the sense that each processor is allowed to advance at its own speed. Thus, these schemes are suitable for single (or multi) user shared memory or (message passing) MIMD multiprocessors. Our approach is demonstrated for the solution of the multidimensional heat equation, of which we present a spatial second-order Parametric Asynchronous Finite-Difference (PAFD) scheme. The well-known synchronous schemes are obtained as its special cases. This is a generalization and expansion of the results in [5] and [7]. The consistency, stability and convergence of this scheme are investigated in detail. Numerical tests show that although PAFD provides the desired order of accuracy, its efficiency is inadequate when performed on each grid point.In an alternative approach that uses domain decomposition, the problem domain is divided among the processors. Each processor computes its subdomain mostly independently, while the PAFD scheme provides the solutions at the subdomains' boundaries. We use high-order finite-difference implicit scheme within each subdomain and determine the values at subdomains' boundaries by the PAFD scheme. Moreover, in order to allow larger time-step, we use remote neighbors' values rather than those of the immediate neighbors. Numerical tests show that this approach provides high efficiency and in the case which uses remote neighbors' values an almost linear speedup is achieved. Schemes similar to the PAFD can be developed for other types of equations [3].This research was supported by the fund for promotion of research at the Technion.     

A modified nearly exact method for solving low-rank trust region subproblem

In this paper, we first discuss how the nearly exact (NE) method proposed by Moré and Sorensen [14] for solving trust region (TR) subproblems can be modified to solve large-scale “low-rank” TR subproblems efficiently. Our modified algorithm completely avoids computation of Cholesky factorizations by instead relying primarily on the Sherman–Morrison–Woodbury formula for computing inverses of “diagonal plus low-rank” type matrices. We also implement a specific version of the modified log-barrier (MLB) algorithm proposed by Polyak [17] where the generated log-barrier subproblems are solved by a trust region method. The corresponding direction finding TR subproblems are of the low-rank type and are then solved by our modified NE method. We finally discuss the computational results of our implementation of the MLB method and its comparison with a version of LANCELOT [5] based on a collection extracted from CUTEr [12] of nonlinear programming problems with simple bound constraints.      

一个MIMD上的多项式求值算法

设N次多项式 f(x)=sum from i=0 to N (a_ix~(N-i),(1)求 f(x)在某给定点的函数值. 熟知,串行计算(1)的最佳算法是Horner法,而并行计算(1)的算法目前有倍增法、分段-倍增法等.对于SIMD型计算机,完全倍增法已达到多项式求值的并行复杂性下界2「log(N+1)」,而对MIMD型并行机来说,结果还可以改进.[3]给出了一个证明:若有N台处理机,多项式求值的并行计算复杂性的一个上界为     

基于对称化原理的对称区域分裂法

本文利用对称化原理,讨论了一种只需在子区域上计算两个完全独立子问题就可得到原问题解的对称区域分裂法,并用此方法求解线性算子方程和线性透射问题.此方法可作为并行算法在MIMD计算机上使用.     

ON AN EFFICIENT IMPLEMENTATION OF THE FACE ALGORITHM FOR LINEAR PROGRAMMING＊

In this paper, we consider the solution of the standard linear programming [Lt＇）. A remarkable result in LP claims that all optimal solutions form an optimal face of the underlying polyhedron. In practice, many real-world problems have infinitely many optimal solutions and pursuing the optimal face, not just an optimal vertex, is quite desirable. The face algorithm proposed by Pan [19] targets at the optimal face by iterating from face to face, along an orthogonal projection of the negative objective gradient onto a relevant null space. The algorithm exhibits a favorable numerical performance by comparing the simplex method. In this paper, we further investigate the face algorithm by proposing an improved implementation. In exact arithmetic computation, the new algorithm generates the same sequence as Pan＇s face algorithm, but uses less computational costs per iteration, and enjoys favorable properties for sparse problems.     

An efficient implementation of a quasi-polynomial algorithm for generating hypergraph transversals and its application in joint generation

Given a finite set V, and a hypergraph H⊆2^V, the hypergraph transversal problem calls for enumerating all minimal hitting sets (transversals) for H. This problem plays an important role in practical applications as many other problems were shown to be polynomially equivalent to it. Fredman and Khachiyan [On the complexity of dualization of monotone disjunctive normal forms, J. Algorithms 21 (1996) 618-628] gave an incremental quasi-polynomial-time algorithm for solving the hypergraph transversal problem. In this paper, we present an efficient implementation of this algorithm. While we show that our implementation achieves the same theoretical worst-case bound, practical experience with this implementation shows that it can be substantially faster. We also show that a slight modification of the original algorithm can be used to obtain a stronger bound on the running time.More generally, we consider a monotone property π over a bounded n-dimensional integral box. As an important application of the above hypergraph transversal problem, pioneered by Bioch and Ibaraki [Complexity of identification and dualization of positive Boolean functions, Inform. and Comput. 123 (1995) 50-63], we consider the problem of incrementally generating simultaneously all minimal subsets satisfying π and all maximal subsets not satisfying π, for properties given by a polynomial-time satisfiability oracle. Problems of this type arise in many practical applications. It is known that the above joint generation problem can be solved in incremental quasi-polynomial time via a polynomial-time reduction to a generalization of the hypergraph transversal problem on integer boxes. In this paper we present an efficient implementation of this procedure, and present experimental results to evaluate our implementation for a number of interesting monotone properties π.     

Optimal Multipliers for Linear Congruential Pseudo-Random Number Generators with Prime Moduli: Parallel Computation and Properties

Two systematic search methods are employed to find multipliers for linear congruential pseudo-random number generation which are optimal with respect to an upper bound for the discrepancy of pairs of successive pseudo-random numbers. The efficiency of these search procedures when executed on parallel systems is assessed by experimental results of a MIMD parallel implementation on a Meiko CS-2 and a workstation cluster. Furthermore the quality of the computed multipliers is evaluated by using the spectral test in dimensions 2–8 and by calculating the actual discrepancy of pairs of the resulting full-period sequences.     

Computational experience using an edge search algorithm for linear reverse convex programs

This paper presents computational experience with a rather straight forward implementation of an edge search algorithm for obtaining the globally optimal solution for linear programs with an additional reverse convex constraint. The paper's purpose is to provide a collection of problems, with known optimal solutions, and performance information for an edge search implementation so that researchers may have some benchmarks with which to compare new methods for reverse convex programs or concave minimization problems. There appears to be nothing in the literature that provides computational experience with a basic edge search procedure. The edge search implementation uses a depth first strategy. As such, this paper's implementation of the edge search algorithm is a modification of Hillestad's algorithm [11]. A variety of test problems is generated by using a modification of the method of Sung and Rosen [20], as well as a new method that is presented in this paper. Test problems presented may be obtained at ftp://newton.ee.ucla.edu/nonconvex/pub/.     

Reducing systems of linear differential equations to a passive form

In this paper, an application of the Riquer-Thomas-Janet theory is described for the problem of transforming a system of partial differential equations into a passive form, i.e., to a special form which contains explicitly both the equations of the initial system and all their nontrivial differential consequences. This special representation of a system markedly facilitates the subsequent integration of the corresponding differential equations. In this paper, the modern approach to the indicated problem is presented. This is the approach adopted in the Knuth-Bendix procedure [13] for critical-pair/completion and then Buchberger's algorithm for completion of polynomial ideal bases [13] (or, alternatively, for the construction of Groebner bases for ideals in a differential operator ring [14]). The algorithm of reduction to the passive form for linear system of partial differential equations and its implementation in the algorithmic language REFAL, as well as the capabilities of the corresponding program, are outlined. Examples illustrating the power and efficiency of the system are presented.     

New ODE Methods for Equality Constrained Optimization (2) — Algorithms

As a continuation of [1], this paper considers implementation of ODE approaches. A modified Hamming's algorithm for integration of (ECP)-equation is suggested to obtain a local solution. In addition to the main algorithm, three supporting algorithms are also described: two are for evaluation of the right-hand side of (ECP)-equation, which may be especially suitable for certain kinds of (ECP)-equation when applied to large scale problems; the third one, with a convergence theorem, is for computing an initial feasible point. Our numerical results obtained by executing these algorithms on an example of (ECP)-equation given in [1] on five test problems indicate their remarkable superiority of performance to Tanabe's ODE version that is recently claimed to be much better than some well-known SQP techniques.     

A branch-and-cut algorithm for nonconvex quadratic programs with box constraints

We present the implementation of a branch-and-cut algorithm for bound constrained nonconvex quadratic programs. We use a class of inequalities developed in [12] as cutting planes. We present various branching strategies and compare the algorithm to several other methods to demonstrate its effectiveness.Mathematics Subject Classification (2000): 90C26, 90C27, 90C20