A randomized method for one‐step extrapolation in reverse time migration

Reverse time migration has drawn great attention in exploration geophysics because it can be used successfully in areas with large structural and velocity complexity. But its computational cost is considerably high. This paper concerns the fast implementation of the optimized separable approximation of the two‐way propagator, which is the most computational expensive step in reverse time migration. On the basis of the low‐rank property of the propagator and the idea of randomized algorithm, a randomized method is introduced for optimized separable approximation‐based one‐step extrapolation. Numerical results of approximating the propagator show that the randomized method is more efficient than the conventional interpolation method. At the same time, numerical experiments of wavefield extrapolation show that the proposed method is much more accurate than the conventional finite‐difference plus pseudo‐spectrum scheme. Copyright © 2014 John Wiley & Sons, Ltd.     

An Approximation Method for Improving Dynamic Network Model Fitting

There has been a great deal of interest recently in the modeling and simulation of dynamic networks, that is, networks that change over time. One promising model is the separable temporal exponential-family random graph model (ERGM) of Krivitsky and Handcock, which treats the formation and dissolution of ties in parallel at each time step as independent ERGMs. However, the computational cost of fitting these models can be substantial, particularly for large, sparse networks. Fitting cross-sectional models for observations of a network at a single point in time, while still a nonnegligible computational burden, is much easier. This article examines model fitting when the available data consist of independent measures of cross-sectional network structure and the duration of relationships under the assumption of stationarity. We introduce a simple approximation to the dynamic parameters for sparse networks with relationships of moderate or long duration and show that the approximation method works best in precisely those cases where parameter estimation is most likely to fail—networks with very little change at each time step. We consider a variety of cases: Bernoulli formation and dissolution of ties, independent-tie formation and Bernoulli dissolution, independent-tie formation and dissolution, and dependent-tie formation models.     

Higher order approximations for tandem queueing networks

In this paper a higher order approximation for single server queues and tandem queueing networks is proposed and studied. Different from the most popular two-moment based approximations in the literature, the higher order approximation uses the higher moments of the interarrival and service distributions in evaluating the performance measures for queueing networks. It is built upon the MacLaurin series analysis, a method that is recently developed to analyze single-node queues, along with the idea of decomposition using higher orders of the moments matched to a distribution. The approximation is computationally flexible in that it can use as many moments of the interarrival and service distributions as desired and produce the corresponding moments for the waiting and interdeparture times. Therefore it can also be used to study several interesting issues that arise in the study of queueing network approximations, such as the effects of higher moments and correlations. Numerical results for single server queues and tandem queueing networks show that this approximation is better than the two-moment based approximations in most cases.     

ℋ︁‐matrices for the convection‐diffusion equation

Hierarchical matrices (ℋ︁‐matrices) provide a technique for the sparse approximation of large, fully populated matrices. This technique has been shown to be applicable to stiffness matrices arising in boundary element method applications where the kernel function displays certain smoothness properties. The error estimates for an approximation of the kernel function by a separable function can be carried over directly to error estimates for an approximation of the stiffness matrix by an ℋ︁‐matrix, using a certain standard partitioning and admissibility condition for matrix blocks. Similarly, ℋ︁‐matrix techniques can be applied in the finite element context where it is the inverse of the stiffness matrix that is fully populated. Here one needs a separable approximation of Green's function of the underlying boundary value problem in order to prove approximability by matrix blocks of low rank. Unfortunately, Green's function for the convection‐diffusion equation does not satisfy the required smoothness properties, hence prohibiting a straightforward generalization of the separable approximation through Taylor polynomials. We will use Green's function to motivate a modification in the (hierarchical) partitioning of the index set and as a consequence the resulting hierarchy of block partitionings as well as the admissibility condition. We will illustrate the effect of the proposed modifications by numerical results.     

An approximation method for general closed queuing networks with Fork/Join mechanisms

Queuing networks with Fork/Join mechanisms are encountered in modelling and analysis of parallel computer systems and computer/communication networks. Exact analytical solutions of such networks are not available. In particular, due to the Fork/Join mechanisms, these networks do not have a product-form solution. As a result, approximation methods that can provide accurate estimates of the performance parameters are of high interest. The purpose of this paper is to propose such an approximation method that applies to a fairly general class of closed queuing networks with Fork/Join mechanisms. The method is based on the use of a product-form approximation technique. Numerical results are provided that show that the accuracy of the method is fairly good.     

Approximating stationary points of stochastic optimization problems in Banach space

In this paper, we present a uniform strong law of large numbers for random set-valued mappings in separable Banach space and apply it to analyze the sample average approximation of Clarke stationary points of a nonsmooth one stage stochastic minimization problem in separable Banach space. Moreover, under Hausdorff continuity, we show that with probability approaching one exponentially fast with the increase of sample size, the sample average of a convex compact set-valued mapping converges to its expected value uniformly. The result is used to establish exponential convergence of stationary sequence under some metric regularity conditions.     

Decomposition methods for polyhedral approximation of the Edgeworth–Pareto hull

In the framework of multiobjective optimization (MOO) techniques, there is a group of methods that attract growing attention. These methods are based on the approximation of the Edgeworth–Pareto hull (EPH), that is, the largest set (in the sense of inclusion) that has the same Pareto frontier as the feasible objective set [1, 2]. The block separable structure of MOO problems is used in this paper to improve the efficiency of EPH approximation methods in the case when EPH is convex. In this case, polyhedral approximation can be applied to EPH. Its practical importance was shown, for example, in [3, 4]. In this paper, a two-level system is considered that consists of an upper (coordinating) level and a finite number of blocks interacting through the upper level. It is assumed that the optimization objectives are related only to the variables of the upper level. The approximation method is based on the polyhedral approximation of EPH for single blocks, followed by the application of these results for approximating the EPH for the whole MOO problem.     

A general performance analysis method for uniform memory architectures

The performance of a multiprocessor system greatly depends on the bandwidth of its memory architecture. In this paper, uniform memory architectures with various interconnection networks including crossbar, multiple-buses and generalized shuffle networks are studied. We propose a general method based on the Markov chain model by assuming that the blocked memory requests will be redistributed to the memory modules in the next memory cycle. This assumption results in an analysis with lower complexity where the number of states is linearly proportional to the number of processors. Moreover, it can provide excellent estimation on the system power and memory bandwidth for all three types of interconnection networks as compared with the simulation results in which the blocked memory requests are resubmitted to the same memory module. Comparisons also show that our method is more general and precise than most existing analysis methods. The method is further extended to estimate the performance of multiprocessor system with caches. The approximation results are also shown to be remarkably good.     

On the approximation of spectra of linear operators on Hilbert spaces

We present several new techniques for approximating spectra of linear operators (not necessarily bounded) on an infinite-dimensional, separable Hilbert space. Our approach is to take well-known techniques from finite-dimensional matrix analysis and show how they can be generalized to an infinite-dimensional setting to provide approximations of spectra of elements in a large class of operators. We conclude by proposing a solution to the general problem of approximating the spectrum of an arbitrary bounded operator by introducing the n-pseudospectrum and argue how that can be used as an approximation to the spectrum.     

Classical Solutions for a Nonlinear Fokker-Planck Equation Arising in Computational Neuroscience

In this paper we analyze the global existence of classical solutions to the initial boundary-value problem for a nonlinear parabolic equation describing the collective behavior of an ensemble of neurons. These equations were obtained as a diffusive approximation of the mean-field limit of a stochastic differential equation system. The resulting nonlocal Fokker-Planck equation presents a nonlinearity in the coefficients depending on the probability flux through the boundary. We show by an appropriate change of variables that this parabolic equation with nonlinear boundary conditions can be transformed into a non standard Stefan-like free boundary problem with a Dirac-delta source term. We prove that there are global classical solutions for inhibitory neural networks, while for excitatory networks we give local well-posedness of classical solutions together with a blow up criterium. Surprisingly, we will show that the spectrum for the operator in the linear case, that corresponding to a system of uncoupled networks, does not give any information about the large time asymptotic behavior.     

Approximation by neural networks with sigmoidal functions

In this paper, we introduce a type of approximation operators of neural networks with sigmodal functions on compact intervals, and obtain the pointwise and uniform estimates of the ap- proximation. To improve the approximation rate, we further introduce a type of combinations of neurM networks. Moreover, we show that the derivatives of functions can also be simultaneously approximated by the derivatives of the combinations. We also apply our method to construct approximation operators of neural networks with sigmodal functions on infinite intervals.     

A tighter variant of Jensen’s lower bound for stochastic programs and separable approximations to recourse functions

In this paper, we propose a new method to compute lower bounds on the optimal objective value of a stochastic program and show how this method can be used to construct separable approximations to the recourse functions. We show that our method yields tighter lower bounds than Jensen’s lower bound and it requires a reasonable amount of computational effort even for large problems. The fundamental idea behind our method is to relax certain constraints by associating dual multipliers with them. This yields a smaller stochastic program that is easier to solve. We particularly focus on the special case where we relax all but one of the constraints. In this case, the recourse functions of the smaller stochastic program are one dimensional functions. We use these one dimensional recourse functions to construct separable approximations to the original recourse functions. Computational experiments indicate that our lower bounds can significantly improve Jensen’s lower bound and our recourse function approximations can provide good solutions.     

On a class of approximations for closed queueing networks

This paper presents some analytical results concerning an approximation procedure for closed queueing networks. The procedure is well-known and has been found useful for product-form networks where large numbers of queues, jobs or job classes prohibit an exact analysis, as well as for networks which do not possess product-form. The procedure represents the mean sojourn time at a queue as a function of the throughput of the queue, and derives a set of fixed point equations for the throughputs of the various job classes. We begin by showing that under a mild regularity condition the fixed point equations have a unique solution. Then we show that derivatives of performance measures can be readily calculated, and that their simple form provides an interesting insight into capacity allocation in closed queueing networks.This work was supported in part by the Nuffield Foundation     

Decomposition in large system optimization using the method of multipliers

Although the method of multipliers can resolve the dual gaps which will often appear between the true optimum point and the saddle point of the Lagrangian in large system optimization using the Lagrangian approach, it is impossible to decompose the generalized Lagrangian in a straightforward manner because of its quadratic character. A technique using the linear approximation of the perturbed generalized Lagrangian has recently been proposed by Stephanopoulos and Westerberg for the decomposition. In this paper, another attractive decomposition technique which transforms the nonseparable crossproduct terms into the minimum of sum of separable terms is proposed. The computational efforts required for large system optimization can be much reduced by adopting the latter technique in place of the former, as illustrated by application of these two techniques to an optimization problem of a chemical reactor system.The authors would like to acknowledge the valuable comments given by Professor D. G. Luenberger of Stanford University.     

Synchronization analysis for complex networks with coupling delay based on T–S fuzzy theory

In this paper, synchronization problems for a general complex networks are investigated by Takagi–Sugeno (T–S) fuzzy theory. A novel concept named linear approximation method is firstly proposed to solve the synchronization problems for T–S fuzzy complex networks. This novel method can linearize the system into some time-delay subsystems, which can effectively simplify the complicated system and then be easy to acquire the synchronization results. Since the expression based on linear matrix inequality (LMI) is used, the synchronization criteria can be easily checked in practice. Numerical simulation examples are provided to verify the effectiveness and the applicability of the proposed method.     

The Completion Time of PERT Networks

This paper proposes an approximation for the completion time mean and variance of PERT networks in which the durations of individual activities are Normally and independently distributed. The proposed approximation involves the manipulation of distribution parameters only and hence can be computed manually, even for large networks.     

Approximate Separability of the Green's Function of the Helmholtz Equation in the High Frequency Limit

The minimum number of terms that are needed in a separable approximation for a Green's function reveals the intrinsic complexity of the solution space of the underlying differential equation. It also has implications for whether low‐rank structures exist in the linear system after numerical discretization. The Green's function for a coercive elliptic differential operator in divergence form was shown to be highly separable [2], and efficient numerical algorithms exploiting low‐rank structures of the discretized systems were developed. In this work, a new approach to study the approximate separability of the Green's function of the Helmholtz equation in the high‐frequency limit is developed. We show (1) lower bounds based on an explicit characterization of the correlation between two Green's functions and a tight dimension estimate for the best linear subspace to approximate a set of decorrelated Green's functions, (2) upper bounds based on constructing specific separable approximations, and (3) sharpness of these bounds for a few case studies of practical interest. © 2018 Wiley Periodicals, Inc.     

A comparative analysis of linear fitting for non-linear functions on optimization: A case study: Air pollution problems

A very frequent problem in advanced mathematical programming models is the linear approximation of convex and non-convex non-linear functions in either the constraints or the objective function of an otherwise linear programming problem. In this paper, based on a model that has been developed for the evaluation and selection of pollutant emission control policies and standards, we shall study several ways of representing non-linear functions of a single argument in mixed integer, separable and related programming terms. Thus we shall study the approximations based on piecewise constant, piecewise adjacent, piecewise non-adjacent additional and piecewise non-adjacent segmented functions. In each type of modelization we show the problem size and optimization results of using the following techniques: separable programming, mixed integer programming with Special Ordered Sets of type 1, linear programming with Special Ordered Sets of type 2 and mixed integer programming using strategies based on the quasi-integrality of the binary variables.     

On Stein?s method for infinite-dimensional Gaussian approximation in abstract Wiener spaces

In this paper, we generalize Stein?s method to “infinite-variate” normal approximation that is an infinite-dimensional approximation by abstract Wiener measures on a real separable Banach space. We first establish a Stein?s identity for abstract Wiener measures and solve the corresponding Stein?s equation. Then we will present a Gaussian approximation theorem using exchangeable pairs in an infinite-variate context. As an application, we will derive an explicit error bound of Gaussian approximation to the distribution of a sum of independent and identically distributed Banach space-valued random variables based on a Lindeberg-Lévy type limit theorem. In addition, an analogous of Berry-Esséen type estimate for abstract Wiener measures will be obtained.     

Nonlinear approximation using Gaussian kernels

It is well known that nonlinear approximation has an advantage over linear schemes in the sense that it provides comparable approximation rates to those of the linear schemes, but to a larger class of approximands. This was established for spline approximations and for wavelet approximations, and more recently by DeVore and Ron (in press) [2] for homogeneous radial basis function (surface spline) approximations. However, no such results are known for the Gaussian function, the preferred kernel in machine learning and several engineering problems. We introduce and analyze in this paper a new algorithm for approximating functions using translates of Gaussian functions with varying tension parameters. At heart it employs the strategy for nonlinear approximation of DeVore-Ron, but it selects kernels by a method that is not straightforward. The crux of the difficulty lies in the necessity to vary the tension parameter in the Gaussian function spatially according to local information about the approximand: error analysis of Gaussian approximation schemes with varying tension are, by and large, an elusive target for approximators. We show that our algorithm is suitably optimal in the sense that it provides approximation rates similar to other established nonlinear methodologies like spline and wavelet approximations. As expected and desired, the approximation rates can be as high as needed and are essentially saturated only by the smoothness of the approximand.