In this paper, we firstly introduce an-operator related with FIF. Then, we get an efficient method by this operator to solve the inverse problem of FIF and the inverse problem of piecewise FIF.

In this paper, a non-overlapping domain decomposition method is discussed for solving the exterior boundary value problem of plane elasticity equation. The exterior domain is naturally decomposed by a circle into a bounded domain and an unbounded domain. With the advantage of the natural boundary reduction, a D-N method is presented. This method is effective and geometric convergent. The convergence rate of this iteration is independent of the finite element mesh size, but dependent on the relaxation factor.

GPS, integrated with Inertial Navigation System, can improve the accuracy of navigation and positioning. In order to apply the GPS signal, the paper investigate the random error adaptive AR model of GPS positioning signal, and approve the model and the algorithm based on numerical analysis.

In this paper, we discuss the convergence properties of the memoryless quasi-Newton method proposed by Shanno (1978). In the two-dimensional quadratic case, we prove the global convergence of the method without any line search; if an exact line search is made at the first iteration, then the method gives the exact solution at most at the forth iteration. Numerical experiments further demonstrate these properties of the memoryless quasi-Newton method.

In this paper, a coupled method of spherical harmonic function expansion and streamline diffusion finite element method for solving 3-dimentional Boltzmann transport equation is presented, and the convergence of the method is obtained. Based on the method, we developed a power software for simulating neutron well-logging problem. The resuts are satisfied.

On solving differential equations, the classical numerical algorithms require us to have some knowledge of the solutions before we find them. For examples, to solve an ordinary differential equation of order 4, the classical Runge-Kutta method requires the initial data of the solution together with its first, second, and third derivatives; the generalized Newton's method depends severely on the initial guess; etc. This paper presents a variational algorithm, based on the Mountain Pass theory, for solving the traveling wave solutions of nonlinear suspension beam equations. Some results of the numerical experiments are shown to indicate that this algorithm does overcome some of the shortcomings of the classical methods.

A coarse-grid correction procedure coupled with patched grid method is designed to solve 2-D/3-D Navier-Stokes equations. It combines the acceleration machanism of coarse-grid correction procedure and the parallel efficiency in PVM. In the steadystate solutions of supersonic blunt body and base flowfields the wall time is shortened dramatically.

In this papers a class of algorithm for determining determinants of sparse banded matrix is suggested. The computing speed of this algorithm is 2--5 times faster than of column pivot Gaussian elimination method. Some munerical experiments are discussed. This algorithm is adapted for parallel evaluation too.

In this paper, we present a fast Fourier transform algorithm for the inverse of Rblock circulant matrices of order mn, its arithmetic complexity is o(mn log_2 mn).

In this note, we point out mistakes in both papers published in this journal about C~k-shape preserving interpolating spline function and provide a corrective method.

This paper presents a new method for trust region subproblems- Tangent Single Dogleg method, this method is proved by analysis and calculated results to be hotter than Powell's single dogleg method.

general method for constructing radial basis function (RBF) and a mul-tiscale approximate formula of an arbitrary function in L2(Rn) based on RBFexpansion are proposed. Two radial basis functions are introduced as examples.They are not only infinitely differentiable but also locally supported. As appli-cations in image processing, they are used for both edge detection and noisesreduction respectively. The computational complexity of the corresponding algo-rithms is O(N2) for N × N images, a linear order of the piXel number, which issignificantly better than that of O(N2 log N) using the preconditioned conjugategradient iterative algorithm for handling leaJst squares problem. The denoisingalgorithm given here can be implemented using less storage and computation thanthe orthogonal least squares learning algorithm. The test results show that theRBF with local support is suited to design fast algorithms in image processing.

Based on norm-minimization technique, a parallel sparse approximate inversepreconditioning method (PPAT method) is present for the unsymmetric sparselinear systems. The sparsity pattern of the approximate inverse is the same as thatof the transpose of the coefficient matrix. This keeps the amount of work and theneed of storage small. The computation of the preconditioner is inherently parallel.Some numerical experiments show that PPAT preconditioners can accelerate theconvergence.

numerical method for 2D-unsteady fluid flow is proposed in this paper,which is based on the common lagrange quadrilateral grid (primary grid) andreconstructed a series of secondary grid. The computational scheme on two-seriesgrid can resist effectively the intersection of grid.

In this paper, four inverse eigenproblems with given three eigenvalues and cor-responding eigenvectors are considered, some necessary and sufficient conditionsunder which there exists a unique solution for these problems are given. Further-more some numerical algorithms and some numerical experiments are given.

llowing for the limitations of LS energy function used in BP algorithm, thispaper proposes a robust learning algorithm based on the study of how cluster-ing puts down radom noise's effects and the consideration of intensified trainingfor high-quality examples. Some simulation results demonstrate that the robustalgorithm is clearly superior to BP algorithm in anti-disturbance and aJstringency.

In order to simulate the response of shafting with hysteresis nonlinear elementsa new improved BP algorithm is proposed. In the new method, the iteration steps,which are previously determined by experience, are here optimized, to prevent errorfunction from falling into a local minimum. At the same time, a new excitationfunction is adopted, so that the learning process of BP network will not oscillateand diverge under large iteration steps. The above two improvements increase theconvergent speed and stability of BP network. An illustrative example is given toshow the effectiveness of the new method.

computational procedure using three-dimensional Euler equations has beendeveloped to predict flows over integrated inlet systems of supersonic flighters ormissiles. The solution of 3-D Euler equations is based on finite volume spatialdiscretization and multi-step Runge-Kuttta time marching schemes proposed byJameson in 1981. In order to suppress its well-known tendency for odd-even pointdecoupling and to minimize pre- and post-shock oscillations, an adaptive dissipation term is added to the system. A non-overlapping domain decomposition technique is used to achieve the required geometric fiexibility and to relieve computermemory capacity requirement. Detailed method is given for the flux interpolationacross the interfaces between adjacent subdomains, and full solution procedure isalso presented. Numerical simulations at three flight states have been implementedfor a supersonic missile which has a complex outer shape. Satisfactory results areobtained.

Balance Mechanism in Jukebox was analyzed, the mathematical modelof optimal design was discussed, and parameter optimization of mechanism wasresolved by numerical integrate and simplex algorithm in this paper. A generalmethod to resolve this kind of problem was put forward in the end of this paper.

In the FEM dynamic analysis of large flexible space structure (LFSS), thesolution of frequencies and corresponding models is actually a general eigenvalueand eigenvector problem. In this article, Lanczos method was applied to thistype of problem. In numerical examples, the convergency, accuracy-time relationand multi-root problem of Lanczos method for beam structure were analyzed.Considering the characteristic of the stiffness matrix of beam structure, the authoralso present some ideas to improve Lanczos method by iteration method.

Parallel computation of FEM under systolic distributed network is a impor-tallt direction of FEM parallel computation. A program designing method and itsimplementation for FEM parallel analysis under network based on PVM is pre-sented. Substructure parallel analysis method with multi-front parallel processingis adopted in FEM parallel computation, the interface equations are solved withPreconditioned Conjugate Gradient (PCG) method. The implementation of thisdesigning method is easy. Example shows the designing method can obtain higherspeedup ratio.

A method is presented for tracking interfaces, which is MOCL (marker on cell line) employed in two-dimensional Eulerian code. To test it, five kinds of objects with different shapes being uniform motion are numerically simulated in a two- dimensional Eulerian hydrodynamics code that uses the MOCL technique to track interfaces. Results show that the method is simple and feasible.

In the processing of measured data, the number of operations of the algorithm for picking out outlier data in batches is very large. A large number of linear or nonlinear equations based on the parameter model built according to the characteristics of measuring equipment and measured object are to be sovied. This paper presents the criterion of picking out outlier data point by point and a parallel algorithm for picking out outlier data in batches, with regard to large-scale linear regression model. The scalability for the parallel algorithm is analyzed, and the results for the algorithm on a group of computers are given. High speed-up is obtained.

A parallel algorithm is developed for the two-dimensional time-dependent convective dominant-diffusion problem. An explicit alternating direction (EAD) method, which is based on the second-order compact upwind finite difference scheme, is studied. The algorithm is tested on a linear and a nonlinear differential equation using a parallel computer. Some numerical results show that the method has high accuracy and is ideally suitable for massively parallel computers.

A new sorting method, the method of code-transformation, graded computing and insertional relocations is presented. Its algorithm description, time complexity and experimental results in C are given. It's proved by algorithm analysis and ex- perimental results that its time complexity is O(N) and the new sorting algorithm is better than the quick sort etc. when data are in equidistribution.

The paper presents a posteriori error estimate of FD-SD method for one- dimension time-dependent convection- dominated diffusion equation 5 which can be used to adjust space mesh. Some numerical results and an algorithm of adaptive finite element method based on this a posteriori error estimate are given.

The practical method is proposed in this papers which is based on the iterative method for solving linear system of equations. The normal equations are solved directly by the algorithm and a minimum norm solution can be obtained when coefficient matrix of normal equation is singular, so that the sequence generated by the algorithm may be convergent. The numerical examples illustrate the algorithm is very efficient for singular problem or ill-condition problem.

An optimal multilevel preconditioner for plate problem was constructed in [7]. In this paper, the matrix presentation of the preconditioner is given and EBE technology is used to realize the saving and computing of global stiff matrix, interpolation matrix and transport matrix. An MPI algorithm is also given on Dawn 1000+.

A simply and flexible simplex identification method is presented in this paper for system identification on complex object function in control. This method has been applied to linearize complex nonlinear magnetic bearing system. This result is very important to analyze the state of magnetic bearing in small range and to explore the method to improve the performance of magnetic bearing.

This study examined the effects of missing Radon transform on single-orbit cone beam SPECT image quality, and employed statistical Kriging in the estimation of missing data on Grangeat algorithm for two simulation phantoms. Evaluation studies found that Kriging estimation improved reconstruction image quality for an object containing sharp edges.

This paper deals with the stability analysis of numerical solution of linear θ- methods for delay differential equations. We focus on the linear test equation X'(t) = ax(t)+bx([t]), where a,b are constants and [t] is the largest-integer function. Sufficent conditions are given for the numerical solution to be asymptotic stable.

In this paper, a finite difference method for a initial-boundary value problem of regularized long-wave equation was considered. A energy conservative finite difference scheme of three levels was proposed. Convergence and stability of difference solution were proved. The scheme needn't iterate, thus, requires less CPU time. Numerical experiment results demonstrate that the method is efficient and reliable.

Ray tracing is a basic aspect in tomography. To solve the caustic problem in inhomogeneous media using Maslov asymptotic theory, we need to calculate the position and slowness vector at every point. Therefore, ray tracing must rely on the ray equations in Hamiltonian form. In this paper, fourth order symplectic scheme and nonsymplectic Runge-Kutta scheme are compared in ray tracing for sinusoidal velocity model. The result indicates that ray paths obtained by two schemes are almost the same. But on keeping Hamilton quantities, the symplectic scheme is far better than the Runge-Kutta scheme. On computing travel time for Htamiltonian system with T parameter, we use trapezoid formula for numerical integration. The result coincides with that obtained using Hamiltonian system with t parameter.

Firstly, with the discusses of main ingredients to exert the peak float performance for currently high performance mirco-processors in detail, this paper analyzed the principal motivations for the speedup of parallel applied codes under the parallel computers consisted of the these micro-processors. Secondly, this paper presented a suite of performance evaluations rules for parallel codes, which can reveal the overall numerical and parallel performance with respect to the serial codes, pose the performance improving strategies, explain exactly the reasons for super-linear Speedup. The numerical experimential results of two realistic applied codes under two parallel computer are also given in this paper.

In tills paper, the numerical implementation of boundary element methods with overlapping domain decomposition method for solving the Navier equations of linear elastodynamics problems in Fourier transformed domain. The computer program is compiled with Fortran 77 and several numerical examples are presented with the test on the relation of convergence rate with the overlapping size.

We propose a virtual boundary condition forecast algorithm for multi grid parallel competing, and derive a forecast function formula in this paper. Numerical results of one and two-dimension boundary condition problems obtained with the algorithm in a PVM network competing environment show that the algorithm has high parallel efficiency.

The sorting method of base distribution and linking is presented. Its algorithm description, time complexity and experimental results in C are given. It's proved by algorithm analysis and experimental results that its time complexity is O(N) and it has nothing to do with data distribution. The method is obviously quicker than that of quick sort, base sort and linking sort etc.

In this paper, a new algorithm of heapsort, called second order heapsort, is suggested. It is proved that the average time complexity of the algorithm is O(n log2 n) and the speed of the algorithm, which has used optimum data processing method, is heightened 180%.

Algorithms for iteratively refining the parameter estimates and residuals from the fitting of a regression model using QR decomposition methods are described, It is shown that if square root free algorithms for performing the QR decomposition are used the related iterative refinement algorithms can also be square root free. Testing of the algorithms is carried out and comments made about accuracies of parameter estimates.

In this paper, some computational skills are presented to calculate the values of functions and gradients involved in entropy function methods so that the overflow phenomena can be avoided.