Viability of methods for generating stable reduced order models

Various methods for generating stable reduced order models are shown to have a serious disadvantage, in that the reduced model may approximate the nondominant poles of the system and hence lead to erroneous approximations. An example is given to compare these methods with the standard Pade approximation method for which this disadvantage does not exist.     

A generalized Padé approximation of time delay operator

This paper addresses the problem of fractional order Padé approximation for the time delay operator. Based on optimization method, a fitting polynomial with basic order α is constructed to approximate the time delay operator. This makes for two generalized Padé approximations in the form of normal and symmetric approximation, respectively. The advantage of the proposed method lies in the fact that the resulting approximate model is in the standard transform function form of any commensurate order α, making all of the existing stability analysis methods and control design tools for such systems amenable to the fractional order time delay systems. Some numerical examples are given to demonstrate the effectiveness of the proposed method.     

A family of homogeneous analysis models for the design of scalable structures

Research in optimum structural design has shown that mathematical programming techniques can be employed efficiently only in conjunction with explicit approximate constraints. In the course of time a well-established approximation for homogeneous functions (scalable structures) has emerged based on the linear Taylor expansion of the displacement functions in the compliance design space (Reciprocal approximation). It has been shown that the quality of this approximation is based on the property that homogeneity of the constraints is maintained and consequently the approximation is exact along the scaling line.The present paper presents a new family of approximations of homogenous functions which have the same properties as the Reciprocal approximation and which produce more accurate models in most of the tested cases. The approximations are obtained by mapping the direct linear Taylor expansion of the constraints unto a space spanned by intervening variables (original design variables to a powerm). Taking the envelope of these constraints along the scaling line yields a new family of approximations governed by the parameterm. It is shown that the Reciprocal approximation is a particular member of this family of approximations (m = –1).The new technique is illustrated with classical examples of truss optimization. An optimal plate design is also reported. A parametric study of the results for various values of the exponentm is presented. It is shown that for special values of the exponentm the new approximations usually yield better models than those based on the Reciprocal approximation.     

代数双曲空间中拟Legendre 基的应用

鉴于Legendre 基等正交基在代数多项式空间中的广泛应用,论文在深入 研究代数双曲空间的拟Legendre 基性质的基础上,给出了其在反函数逼近和等距曲线逼近 上的应用。利用多项式和双曲函数的混合多项式序列来逼近反函数,并通过实例证明给出方 法的有效性;对基曲线的法矢曲线进行逼近,构造H-Bézier 曲线的等距曲线的最佳逼近, 这种方法直接求得逼近曲线的控制顶点,计算简单,截断误差小。     

A preference aggregation rule approach for structural optimization

In this paper, preference aggregation rules are used to define overall design evaluation measures in optimal design problems. A methodology for the efficient solution of the corresponding design optimization problems is presented. Each design criterion as well as the constraints imposed on the design variables and problem parameters are characterized by preference functions. The nondifferentiable nature of the optimization problems which arise in this formulation is coped with using a first-order algorithm combined with approximation concepts. High-quality approximations for the system response functions are constructed using the concepts of intermediate response quantities and intermediate variables. These approximations are used to replace the original problem by a sequence of approximate problems. Example problems are presented to study the performance of the proposed optimization technique as well as the methodology based on approximation concepts.     

Efficient additive kernels via explicit feature maps

Large scale nonlinear support vector machines (SVMs) can be approximated by linear ones using a suitable feature map. The linear SVMs are in general much faster to learn and evaluate (test) than the original nonlinear SVMs. This work introduces explicit feature maps for the additive class of kernels, such as the intersection, Hellinger's, and χ2 kernels, commonly used in computer vision, and enables their use in large scale problems. In particular, we: 1) provide explicit feature maps for all additive homogeneous kernels along with closed form expression for all common kernels; 2) derive corresponding approximate finite-dimensional feature maps based on a spectral analysis; and 3) quantify the error of the approximation, showing that the error is independent of the data dimension and decays exponentially fast with the approximation order for selected kernels such as χ2. We demonstrate that the approximations have indistinguishable performance from the full kernels yet greatly reduce the train/test times of SVMs. We also compare with two other approximation methods: Nystrom's approximation of Perronnin et al., which is data dependent, and the explicit map of Maji and Berg for the intersection kernel, which, as in the case of our approximations, is data independent. The approximations are evaluated on a number of standard data sets, including Caltech-101, Daimler-Chrysler pedestrians, and INRIA pedestrians.     

Adaptive Quadrature for Maximum Likelihood Estimation of a Class of Dynamic Latent Variable Models

Maximum likelihood estimation of models based on continuous latent variables generally requires to solve integrals that are not analytically tractable. Numerical approximations represent a possible solution to this problem. We propose to use the adaptive Gaussian–Hermite (AGH) numerical quadrature approximation for a particular class of continuous latent variable models for time-series and longitudinal data. These dynamic models are based on time-varying latent variables that follow an autoregressive process of order 1, AR(1). Two examples are the stochastic volatility models for the analysis of financial time series and the limited dependent variable models for the analysis of panel data. A comparison between the performance of AGH methods and alternative approximation methods proposed in the literature is carried out by simulation. Empirical examples are also used to illustrate the proposed approach.     

Finite element methods for elliptic problems with singularities

It has been shown how singular isoparametric transformations defined on a triangular mesh can be used to provide finite element approximations that behave locally as r¹/m. The purpose of the present note is to show how this form of local approximation can be extended to quadrilateral meshes and how any approximations that behave as r^P/q can be obtained for arbitrary p and q.Examples are given of approximate solutions of elliptic partial differential equations using this form of finite element approximation. It is shown that accurate approximations to the coefficients of leading terms in the solution near the singularity are obtained as well as a good approximation to the solution itself.Finally, it is shown how curved elements with the desired singular behaviour can be constructed by means of a generalisation of the basic approach. Approximations based on transformations involving functions other than polynomials are also introduced.     

Using Economic Theory to Guide Numerical Analysis: Solving for Equilibria in Models of Asymmetric First-Price Auctions

In models of first-price auctions, when bidders are ex ante heterogeneous, deriving explicit equilibrium bid functions is typically impossible, so numerical methods are often employed to find approximate solutions. Recent theoretical research concerning asymmetric auctions has determined some properties these bid functions must satisfy when certain conditions are met. Plotting the relative expected pay-offs of bidders is a quick, informative way to decide whether the approximate solutions are consistent with theory. We approximate the (inverse-)bid functions by polynomials and employ theoretical results in two ways: to help solve for the polynomial coefficients and to evaluate qualitatively the appropriateness of a given approximation. We simulate auctions from the approximated solutions and find that, for the examples considered, low-degree polynomial approximations perform poorly and can lead to incorrect policy recommendations concerning auction design, suggesting researchers need to take care to obtain quality solutions.     

Incomplete series expansion for function approximation

We present an incomplete series expansion (ISE) as a basis for function approximation. The ISE is expressed in terms of an approximate Hessian matrix, which may contain second, third, and even higher order “main” or diagonal terms, but which excludes “interaction” or off-diagonal terms. From the ISE, a family of approximation functions may be derived. The approximation functions may be based on an arbitrary number of previously sampled points, and any of the function and gradient values at suitable previously sampled points may be enforced when deriving the approximation functions. When function values only are enforced, the storage requirements are minimal. However, irrespective of the conditions enforced, the approximate Hessian matrix is a sparse diagonal matrix. In addition, the resultant approximations are separable. Hence, the proposed approximation functions are very well-suited for use in gradient-based sequential approximate optimization requiring computationally expensive simulations; a typical example is structural design problems with many design variables and constraints. We derived a wide selection of approximations from the family of ISE approximating functions; these include approximations based on the substitution of reciprocal and exponential intervening variables. A comparison with popular approximating functions previously proposed illustrates the accuracy and flexibility of the new family of approximation functions. In fact, a number of popular approximating functions previously proposed for structural optimization applications derive from our ISE. Based on the similarly named paper presented at the Sixth World Congress on Structural and Multidisciplinary Optimization, Rio de Janeiro, Brazil, May 2005     

Approximated approximations for SAO

We propose to replace a number of popular approximations by their diagonal quadratic Taylor series expansions. The resulting separable quadratic approximations are easily convexified, and are well suited for use in dual sequential approximate optimization (SAO) algorithms. Global convergence of the resulting SAO algorithms may be enforced in a natural way using conservatism. The approximated approximation approach is explicitly illustrated for (i) reciprocal and exponential intervening variables, (ii) the intervening variables used in the method of moving asymptotes (MMA), (iii) the intervening variables used in CONLIN, and (iv) the TANA-3 approximations. The use of intermediate responses for use in, for example, truss and frame-like structures, is also discussed. Key advantages of replacing nonlinear approximations by their diagonal quadratic approximations are that these approximated approximations can all be used simultaneously in a single dual statement; the dual does not depend on the form of the original approximations. In addition, in a dual setting, the resulting subproblems yield simple analytical relationships between the primal and dual variables, which is often not the case with the original nonlinear approximations. An important example hereof is the exponential approximation. Although the diagonal quadratic approximations may differ notably from their original counterparts, they typically are quite similar in a sufficiently small search subregion, which relates to the move limits commonly used in SAO anyway.     

A compliance approximation method applied to variable stiffness composite optimisation

A way to approximate the compliance of composites for optimisation is described. A two-level approximation scheme is proposed inspired by traditional approximation concepts such as force approximations and convex linearisation. In level one, an approximation in terms of the reciprocal in-plane stiffness matrix is made. In level two, either the lamination parameters, or the nodal fibre angle distribution are used as design variables. A quadratic approximation is used to build the approximations in terms of the fibre angles. The method of conservative, convex separable approximations is used for the optimisation. Conservativeness is guaranteed by adding a convex damping function to the approximations. Two numerical examples, one optimisng the compliance of a plate clamped on the left, loaded downwards on the bottom right, another one optimising the compliance of a plate loaded with a shear force and a moment show the computational efficiency of the proposed optimisation algorithm.     

Four- and eight-node hybrid-Trefftz quadrilateral finite element models for helmholtz problem

In this paper, four- and eight-node quadrilateral finite element models which can readily be incorporated into the standard finite element program framework are devised for plane Helmholtz problems. In these models, frame (boundary) and domain approximations are defined. The former is obtained by nodal interpolation and the latter is truncated from Trefftz solution sets. The equality of the two approximations are enforced along the element boundary. Both the Bessel and plane wave solutions are employed to construct the domain approximation. For full rankness, a minimal of four and eight domain modes are required for the four- and eight-node elements, respectively. By using local coordinates and directions, rank sufficient and invariant elements with minimal and close to minimal numbers of domain approximation modes are devised. In most tests, the proposed hybrid-Trefftz elements with the same number of nodes yield close solutions. In absolute majority of the tests, the proposed elements are considerably more accurate than their single-field counterparts.     

Reconstruction of Piecewise Smooth Functions from Non-uniform Grid Point Data

Spectral series expansions of piecewise smooth functions are known to yield poor results, with spurious oscillations forming near the jump discontinuities and reduced convergence throughout the interval of approximation. The spectral reprojection method, most notably the Gegenbauer reconstruction method, can restore exponential convergence to piecewise smooth function approximations from their (pseudo-)spectral coefficients. Difficulties may arise due to numerical robustness and ill-conditioning of the reprojection basis polynomials, however. This paper considers non-classical orthogonal polynomials as reprojection bases for a general order (finite or spectral) reconstruction of piecewise smooth functions. Furthermore, when the given data are discrete grid point values, the reprojection polynomials are constructed to be orthogonal in the discrete sense, rather than by the usual continuous inner product. No calculation of optimal quadrature points is therefore needed. This adaptation suggests a method to approximate piecewise smooth functions from discrete non-uniform data, and results in a one-dimensional approximation that is accurate and numerically robust.      

双论域上的广义直觉模糊概率粗糙集模型及其应用

已有的双论域直觉模糊概率粗糙集模型通过设置两个阈值${\lambda _1}$、${\lambda _2} $,讨论了经典集合在直觉模糊二元关系下的概率粗糙下上近似。该模型不能计算直觉模糊集合在直觉模糊二元关系下的概率粗糙下上近似,这在一定程度上限制了该模型的应用。首先给出了直觉模糊条件概率的定义。在直觉模糊概率空间下构造了双论域广义直觉模糊概率粗糙集模型,讨论了模型的主要性质。最后,将模型应用到临床诊断系统中。与其他模型相比,所提出的广义直觉模糊概率粗糙集模型进一步丰富了概率粗糙集理论,更适合于实际应用。     

Linearization and Higher-Order Approximations: How Good are They?

The standard procedure for analyzing transitional dynamics in non-linear macro models has been to employ linear approximations. Recently quadratic approximations have been explored. This paper examines the accuracy of these and higher-order approximations in an endogenous growth model with public capital, thereby extending the work done in the current literature on the neoclassical growth model. We find that significant errors may persist in computed transition paths and welfare even after resorting to approximations as high as fourth order. Moreover, the accuracy of approximations may not increase monotonically with the increase in the order of approximation. Also, as in the previous literature, we find that achieving acceptable levels of accuracy when computing the welfare consequences of a policy change typically requires a higher order approximation than attaining similar levels of accuracy in the computation of the transition path: typically an increase in order of approximation by one is sufficient.     

Estimation of the mean squared error of predictors of small area linear parameters under a logistic mixed model

An accuracy measure (mean squared error, MSE) is necessary when small area estimators of linear parameters are provided. Even in the case when such estimators arise from the assumption of relatively simple models for the variable of interest, as linear mixed models, the analytic form of the MSE is not suitable to be calculated explicitly. Some good and widely used approximations are available for those models. For generalized linear mixed models, a rough approximation can be obtained by a linearization of the model and application of Prasad-Rao approximation for linear mixed models. Resampling methods, although computationally demanding, represent a conceptually simple alternative. Under a logistic mixed linear model for the characteristic of interest, the Prasad-Rao-type formula is compared with a bootstrap estimator obtained by a wild bootstrap designed for estimating under finite populations. A simulation study is developed in order to study the performance of both methods for estimating a small area proportion.     

Optimization of a sheet metal forming process using successive multipoint approximations

An automated optimization method based on multipoint approximations and applied to the design of a sheet metal forming process is presented. Due to the highly complex nature of the design functions, it was decided to focus on the characterization of the final product thickness distribution as a function of the preforming die shape variables. This was achieved by constructing linear approximations to the noisy responses usingresponse surface methodology (RSM). These approximations are used to obtain an approximate solution to an optimization problem. Successive approximations are constructed, which improve the solution. An automated panning-zooming scheme is used to resize and position the successive regions of approximation. The methodology is applied to optimally design the preforming die shape used in the manufacture of an automotive wheel centre pressing from a sheet metal blank. The die shape is based on a cubic spline interpolation and the objective is to minimize the blank weight, subject to minimum thickness constraints. A weight saving of up to 9.4% could be realized for four shape variables. Restart is introduced to escape local minima due to the presence of noise and to accelerate the progress of the optimization process.