Generalization of Backward Differentiation Formulas for Parallel Computers

We analyze an extension of backward differentiation formulas, used as boundary value methods, that generates a class of methods with nice stability and convergence properties. These methods are obtained starting from the boundary value GBDFs class, and are in the class of EBDF-type methods. We discuss different ways of using these linear multistep formulas in order to have efficient parallel implementations. Numerical experiments show their effectiveness.     

Modified defect correction algorithms for ODEs. Part II: Stiff initial value problems

As shown in part I of this paper and references therein, the classical method of Iterated Defect Correction (IDeC) can be modified in several nontrivial ways, extending the flexibility and range of applications of this approach. The essential point is an adequate definition of the defect, resulting in a significantly more robust convergence behavior of the IDeC iteration, in particular, for nonequidistant grids. The present part II is devoted to the efficient high-order integration of stiff initial value problems. By means of model problem investigation and systematic numerical experiments with a set of stiff test problems, our new versions of defect correction are systematically evaluated, and further algorithmic measures are proposed for the stiff case. The performance of the different variants under consideration is compared, and it is shown how strong coupling between non-stiff and stiff components can be successfully handled. AMS subject classification 65L05 Supported by the Austrian Research Fund (FWF) grant P-15030.     

On the Convergence of Backward Differentiation Formulas for Stiff Initial Value Problems

The influence of a time-dependent transformation to a numerical method is studied. Thus convergence results of backward differentiation formulas applied to the non-autonomous stiff system y = A(t)y + (t) are given. The approach is based on a special decomposition of the companion matrix.This revised version was published online in October 2005 with corrections to the Cover Date.     

Parallel iteration of the extended backward differentiation formulas

The extended backward differentiation formulas (EBDFs) and theirmodified form (MEBDF) were proposed by Cash in the 1980s forsolving initial value problems (IVPs) for stiff systems of ordinarydifferential equations (ODEs). In a recent performance evaluationof various IVP solvers, including a variable-step-variable-orderimplementation of the MEBDF method by Cash, it turned out thatthe MEBDF code often performs more efficiently than codes likeRADAU5, DASSL and VODE. This motivated us to look at possibleparallel implementations of the MEBDF method. Each MEBDF stepessentially consists of successively solving three non-linearsystems by means of modified Newton iteration using the sameJacobian matrix. In a direct implementation of the MEBDF methodon a parallel computer system, the only scope for (coarse grain)parallelism consists of a number of parallel vector updates.However, all forward–backward substitutions and all right-hand-sideevaluations have to be done in sequence. In this paper, ourstarting point is the original (unmodified) EBDF method. Asa consequence, two different Jacobian matrices are involvedin the modified Newton method, but on a parallel computer system,the effective Jacobian-evaluation and the LU decomposition costsare not increased. Furthermore, we consider the simultaneoussolution, rather than the successive solution, of the threenon-linear systems, so that in each iteration the forward–backwardsubstitutions and the right-hand-side evaluations can be doneconcurrently. A mutual comparison of the performance of theparallel EBDF approach and the MEBDF approach shows that wecan expect a speed-up factor of about 2 on three processors.     

Families of methods for ordinary differential equations based on trigonometric polynomials

We consider the construction of methods based on trigonometric polynomials for the initial value problems whose solutions are known to be periodic. It is assumed that the frequency w can be estimated in advance. The resulting methods depend on a parameter ν = hw, where h is the step size, and reduce to classical multistep methods if ν → 0. Gautschi [4] developed Adams and Störmer type methods. In our paper we construct Nyström's and Milne-Simpson's type methods. Numerical experiments show that these methods are not sensitive to changes in w, but require the Jacobian matrix to have purely imaginary eigenvalues.     

Parallel initial-value algorithms for singularly perturbed boundary-value problems

Initial-value methods for linear and semilinear singularly perturbed boundary-value problems are examined with a view to designing and implementing algorithms on parallel architectures. Practical experiments on a CRAY Y-MP 8/432 multiprocessor have been performed, showing the reliability and performance of several proposed parallel schemes.This work was supported by CNR, Rome, Italy (Progetto Finalizzato Sistemi Informatici e Calcolo Parallelo, Sottoprogetto 1).The authors wish to thank Dr. A. Papini, who carried out most of the computations reported in this work.     

Trigonometric fitted modification of RADAU5

Implicit Runge-Kutta (RK) methods are in common use when addressing stiff initial value problems (IVP). They usually share the property of A-stability that is of crucial importance in solving the latter type of IVP. Radau IIA family of implicit RK methods is among the preferred ones. Especially its fifth-order representative named RADAU5 has received a lot of attention for use with lax accuracies. Here, we try the lesser possible perturbation of its coefficients. Then, we derive a trigonometric fitted modification that is intended to be applied in periodic IVPs. Numerical tests over a variety of problems with oscillatory solutions justify our effort.     

Parallel iterative methods using factorized preconditioning matrices for solving elliptic equations on triangular grids

Parallel analogs of the variants of the incomplete Cholesky-conjugate gradient method and the modified incomplete Cholesky-conjugate gradient method for solving elliptic equations on uniform triangular and unstructured triangular grids on parallel computer systems with the MIMD architecture are considered. The construction of parallel methods is based on the use of various variants of ordering the grid points depending on the decomposition of the computation domain. Results of the theoretic and experimental studies of the convergence rate of these methods are presented. The solution of model problems on a moderate number processors is used to examine the efficiency of the proposed parallel methods.     

Parallel asynchronous label-correcting methods for shortest paths

We develop parallel asynchronous implementations of some known and some new label-correcting methods for finding a shortest path from a single origin to all the other nodes of a directed graph. We compare these implementations on a shared-memory multiprocessor, the Alliant FX/80, using several types of randomly generated problems. Excellent (sometimes superlinear) speedup is achieved with some of the methods, and it is found that the asynchronous versions of these methods are substantially faster than their synchronous counterparts.The authors acknowledge the director and the staff of CERFACS, Toulouse, France for the use of the Alliant FX/80.This research was supported by the National Science Foundation under Grants 9108058-CCR, 9221293-INT, and 9300494-DMI.     

Parallel Asynchronous Schwarz and Multisplitting Methods for a Nonlinear Diffusion Problem

Parallel asynchronous subdomain algorithms with flexible communication for the numerical solution of nonlinear diffusion problems are presented. The discrete maximum principle is considered and the Schwarz alternating method and multisplitting methods are studied. A connection is made with M-functions for a classical nonlinear diffusion problem. Finally, computational experiments carried out on a shared memory multiprocessor are presented and analyzed.     

A Parallel Algorithm for the Estimation of the Global Error in Runge–Kutta Methods

The object of this work is the estimate of the global error in the numerical solution of the IVP for a system of ODE's. Given a Runge–Kutta formula of order q, which yields an approximation y _n to the true value y(x _n ), a general, parallel method is presented, that provides a second value y _n ^* of order q+2; the global error e _n =y _n –y(x _n ) is then estimated by the difference y _n –y _n ^*. The numerical tests reported, show the very good performance of the procedure proposed. A comparison with the code GEM90 is also appended.     

New construction of higher-order local continuous platforms for Error Correction Methods

Error correction method (ECM)~\cite{kim2011a,kim2011b} which has been recently developed, is based on the construction of a local approximation to the solution on each time step, and has the excellent convergence order $O(h^{2p+2})$, provided the local approximation has a local residual error $O(h^p)$. In this paper, we construct a higher-order continuous local platform to develop higher-order semi-explicit one-step ECM for solving initial value time dependent differential equations. It is shown that special choices of parameters for the local platform can lead to the improvement of the well-known explicit fourth and fifth order Runge-Kutta methods. Numerical experiments demonstrate the theoretical results     

Precisely A(α)-Stable One-Leg Multistep Methods

One-Leg Multistep (OLM) methods for initial value problems in ODEs use a nonlinear multistep formula to compute the solution at the next integration point. This paper shows that there exists an evaluation point t ^* which gives an OLM formula more precise than BDF's and (almost) precisely A()-stable for a k-step method (k6), and whose stability angle is essentially similar to BDF's. The stability region can be further improved by applying the corrector idea of Klopfenstein.This revised version was published online in October 2005 with corrections to the Cover Date.     

Spectral Deferred Correction Methods for Ordinary Differential Equations

We introduce a new class of methods for the Cauchy problem for ordinary differential equations (ODEs). We begin by converting the original ODE into the corresponding Picard equation and apply a deferred correction procedure in the integral formulation, driven by either the explicit or the implicit Euler marching scheme. The approach results in algorithms of essentially arbitrary order accuracy for both non-stiff and stiff problems; their performance is illustrated with several numerical examples. For non-stiff problems, the stability behavior of the obtained explicit schemes is very satisfactory and algorithms with orders between 8 and 20 should be competitive with the best existing ones. In our preliminary experiments with stiff problems, a simple adaptive implementation of the method demonstrates performance comparable to that of a state-of-the-art extrapolation code (at least, at moderate to high precision).Deferred correction methods based on the Picard equation appear to be promising candidates for further investigation.     

A family of explicit parallel Runge-Kutta-Nyström methods

A procedure for the construction of high-order explicit parallel Runge-Kutta-Nyström (RKN) methods for solving second-order nonstiff initial value problems (IVPs) is analyzed. The analysis reveals that starting the procedure with a reference symmetric RKN method it is possible to construct high-order RKN schemes which can be implemented in parallel on a small number of processors. These schemes are defined by means of a convex combination of k disjoint s_i-stage explicit RKN methods which are constructed by connecting s_i steps of a reference explicit symmetric method. Based on the reference second-order Störmer-Verlet methods we derive a family of high-order explicit parallel schemes which can be implemented in variable-step codes without additional cost. The numerical experiments carried out show that the new parallel schemes are more efficient than some sequential and parallel codes proposed in the scientific literature for solving second-order nonstiff IVPs.     

Runge-Kutta Methods for the Numerical Solution of Stiff Semilinear Systems

This paper studies the stability and convergence properties of general Runge-Kutta methods when they are applied to stiff semilinear systems y(t) = J(t)y(t) + g(t, y(t)) with the stiffness contained in the variable coefficient linear part.We consider two assumptions on the relative variation of the matrix J(t) and show that for each of them there is a family of implicit Runge-Kutta methods that is suitable for the numerical integration of the corresponding stiff semilinear systems, i.e. the methods of the family are stable, convergent and the stage equations possess a unique solution. The conditions on the coefficients of a method to belong to these families turn out to be essentially weaker than the usual algebraic stability condition which appears in connection with the B-stability and convergence for stiff nonlinear systems. Thus there are important RK methods which are not algebraically stable but, according to our theory, they are suitable for the numerical integration of semilinear problems.This paper also extends previous results of Burrage, Hundsdorfer and Verwer on the optimal convergence of implicit Runge-Kutta methods for stiff semilinear systems with a constant coefficients linear part.     

Parallel Calculation of Accurate Path Lines in Virtual Environments through Exploitation of Multi-Block CFD Data Set Topology

The use of Virtual Reality (VR) techniques for the investigation of complex flow phenomena offers distinct advantages in comparison to conventional visualization techniques. Especially for unsteady flows, VR methodology provides an intuitive approach for the exploration of simulated fluid flows. However, the visualization of Computational Fluid Dynamics (CFD) data is often too time-consuming to be carried out in real-time, and thus violates essential constraints concerning real-time interaction and visualization. To overcome this obstacle, we make use of the fact that typically a multi-block approach is employed for domain decomposition, and we use the corresponding data structures for the computation of path lines and for parallelization. In this paper, we present the synthesis of fragmented multi-block data sets and our implementation of an accurate path line integration scheme in order to speed up path line computations. We report on the results of our efforts and describe a combination of this algorithm with a highly efficient visualization approach of large amounts of particle traces, thus considerably improving interactivity when exploring large scale CFD data sets.Mathematics Subject Classifications (2000) 76Mxx, 76M27, 76M28, 65M55, 65L05, 65L06, 65D05, 65Y05, 68U05.     

Efficient parallel multivalue hybrid methods for stiff differential equations

A class of efficient parallel multivalue hybrid methods for stiff differential equations are presented, which are all extremely stable at infinity,A-stable for orders 1–3 and A(α)-stable for orders 4–8. Each method of the class can be performed parallelly using two processors with each processor having almost the same computational amount per integration step as a backward differentiation formula (BDF) of the same order with the same stepsize performed in serial, whereas the former has not only much better stability properties but also a computational accuracy higher than the corresponding BDF. Theoretical analysis and numerical experiments show that the methods constructed in this paper are superior in many respects not only to BDFs but also to some other commonly used methods.     

SERK2v2: A new second‐order stabilized explicit Runge‐Kutta method for stiff problems

Traditionally, explicit numerical algorithms have not been used with stiff ordinary differential equations (ODEs) due to their stability. Implicit schemes are usually very expensive when used to solve systems of ODEs with very large dimension. Stabilized Runge‐Kutta methods (also called Runge–Kutta–Chebyshev methods) were proposed to try to avoid these difficulties. The Runge–Kutta methods are explicit methods with extended stability domains, usually along the negative real axis. They can easily be applied to large problem classes with low memory demand, they do not require algebra routines or the solution of large and complicated systems of nonlinear equations, and they are especially suited for discretizations using the method of lines of two and three dimensional parabolic partial differential equations. In Martín‐Vaquero and Janssen [Comput Phys Commun 180 (2009), 1802–1810], we showed that previous codes based on stabilized Runge–Kutta algorithms have some difficulties in solving problems with very large eigenvalues and we derived a new code, SERK2, based on sixth‐order polynomials. Here, we develop a new method based on second‐order polynomials with up to 250 stages and good stability properties. These methods are efficient numerical integrators of very stiff ODEs. Numerical experiments with both smooth and nonsmooth data support the efficiency and accuracy of the new algorithms when compared to other well‐known second‐order methods such as RKC and ROCK2. © 2012 Wiley Periodicals, Inc. Numer Methods Partial Differential Eq 2013     

Parallel algorithms for nonlinear programming problems

This paper describes several parallel algorithms for solving nonlinear programming problems. Two approaches where parallelism can successfully be introduced have been explored: a quadratic approximation method based on penalty function and a dual method. These methods are improved by using two algorithms originally proposed for solving unconstrained problems: the parallel variable metric algorithm and the parallel Jacobson-Oksman algorithm. Even though general problems are dealt with, particular emphasis is placed on the potential of these parallel methods for separable programming problems. The numerical effectiveness of the algorithms is demonstrated on a set of test problems using a Cray-1S vector computer and serial computers (with respect to sequential versions of the same methods).These studies were sponsored in part by the CERT. The author would particularly like to thank Ph. Berger (LSI-ENSEEIHT), the researchers of the DERI (CERT) and of the Groupe Structures, Aerospatiale, for their assistance.