NUMERICAL VALIDATION OF PROBABILISTIC LAWS TO EVALUATE FINITE ELEMENT ERROR ESTIMATES

We propose a numerical validation of a probabilistic approach applied to estimate the relative accuracy between two Lagrange finite elements Pk and Pm,(k < m). In particular, we show practical cases where finite element Pk gives more accurate results than finite element Pm. This illustrates the theoretical probabilistic framework we recently derived in order to evaluate the actual accuracy. This also highlights the importance of the extra caution required when comparing two numerical methods, since the classical results of error estimates concerns only the asymptotic convergence rate.

Synchronized Dynamics of Hyperchaotic Flows via an Improved Finite-Time Adaptive Controller Design

We proposed a performance-improved finite-time adaptive synchronizing controllers and parameter update laws for coupling the dynamics of identical 4D hyperchaotic flows. The four-dimensional hyperchaotic flows consists of 12 terms and 11 system parameters and possessed very rich dynamics and larger parameter space. The performance of the proposed finite-time adaptive synchronizing controller was enhanced by the introduction of scalar quantities known as global controller strength coefficients and parameter update strength coefficients respectively, into the algebraically-derived control and parameter update structures, in order to constrained overshoots of the trajectories of the coupled systems and accelerate their rate of uniform convergence in finite time. Numerical simulation results obtained confirmed that the uniform asymptotic convergence rate of the coupling trajectories was faster, while the parameter update laws give a stable identification of the unknown system parameters in a global synchronizing time. A comparative analysis of the convergence time of the proposed adaptive controllers with recently published works indicated that the proposed controller has faster rates of uniform convergence of system trajectories.

Convergence Rates of a Class of Predictor-Corrector Iterations for the Nonsymmetric Algebraic Riccati Equation Arising in Transport Theory

In this paper, we analyse the convergence rates of several different predictor-corrector iterations for computing the minimal positive solution of the nonsymmetric algebraic Riccati equation arising in transport theory. We have shown theoretically that the new predictor-corrector iteration given in [Numer. Linear Algebra Appl., 21 (2014), pp. 761–780] will converge no faster than the simple predictor-corrector iteration and the nonlinear block Jacobi predictor-corrector iteration. Moreover the last two have the same asymptotic convergence rate with the nonlinear block Gauss-Seidel iteration given in [SIAM J. Sci. Comput., 30 (2008), pp. 804–818]. Preliminary numerical experiments have been reported for the validation of the developed comparison theory.

Asymptotic Convergence Properties of the EM Algorithm for Mixture of Experts

Mixture of experts (ME) is a modular neural network architecture for supervised classification. The double-loop expectation-maximization (EM) algorithm has been developed for learning the parameters of the ME architecture, and the iteratively reweighted least squares (IRLS) algorithm and the Newton-Raphson algorithm are two popular schemes for learning the parameters in the inner loop or gating network. In this letter, we investigate asymptotic convergence properties of the EM algorithm for ME using either the IRLS or Newton-Raphson approach. With the help of an overlap measure for the ME model, we obtain an upper bound of the asymptotic convergence rate of the EM algorithm in each case. Moreover, we find that for the Newton approach as a specific Newton-Raphson approach to learning the parameters in the inner loop, the upper bound of asymptotic convergence rate of the EM algorithm locally around the true solution Θ* is [Formula: see text], where ϵ>0 is an arbitrarily small number, o(x) means that it is a higher-order infinitesimal as x → 0, and e(Θ*) is a measure of the average overlap of the ME model. That is, as the average overlap of the true ME model with large sample tends to zero, the EM algorithm with the Newton approach to learning the parameters in the inner loop tends to be asymptotically superlinear. Finally, we substantiate our theoretical results by simulation experiments.

Asymptotic Convergence Rate of the EM Algorithm for Gaussian Mixtures

It is well known that the convergence rate of the expectation-maximization (EM) algorithm can be faster than those of convention first-order iterative algorithms when the overlap in the given mixture is small. But this argument has not been mathematically proved yet. This article studies this problem asymptotically in the setting of gaussian mixtures under the theoretical framework of Xu and Jordan (1996). It has been proved that the asymptotic convergence rate of the EM algorithm for gaussian mixtures locally around the true solution Θ* is o(e0.5−ε(Θ*)), where ε > 0 is an arbitrarily small number, o(x) means that it is a higher-order infinitesimal as x → 0, and e(Θ*) is a measure of the average overlap of gaussians in the mixture. In other words, the large sample local convergence rate for the EM algorithm tends to be asymptotically superlinear when e(Θ*) tends to zero.

Spectral Analysis of Large Finite Element Problems by Optimization Methods

Recently an efficient method for the solution of the partial symmetric eigenproblem (DACG, deflated-accelerated conjugate gradient) was developed, based on the conjugate gradient (CG) minimization of successive Rayleigh quotients over deflated subspaces of decreasing size. In this article four different choices of the coefficientβkrequired at each DACG iteration for the computation of the new search directionPkare discussed. The “optimal” choice is the one that yields the same asymptotic convergence rate as the CG scheme applied to the solution of linear systems. Numerical results point out that the optimalβkleads to a very cost effective algorithm in terms of CPU time in all the sample problems presented. Various preconditioners are also analyzed. It is found that DACG using the optimalβkand (LLT)−1as a preconditioner, L being the incomplete Cholesky factor of A, proves a very promising method for the partial eigensolution. It appears to be superior to the Lanczos method in the evaluation of the 40 leftmost eigenpairs of five finite element problems, and particularly for the largest problem, with size equal to 4560, for which the speed gain turns out to fall between 2.5 and 6.0, depending on the eigenpair level.