Parameter estimation with model order reduction and global measurements

We present a new method for parameter estimation for elliptic partial differential equations. Parameter estimation requires the evaluation of the partial differential equation for many different parameter sets. Therefore, model order reduction is reasonable. Model order reduction is composed of an offline phase and an online phase. In the offline phase the reduced model is constructed using snapshots. In this paper we use the given measurement as only snapshot. Hence, the computational costs of the offline phase are reduced. (© 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim)     

Implications of order reduction for implicit Runge-Kutta methods

Stability analysis of Runge-Kutta (RK) formulas was originally limited to linear ordinary differential equations (ODEs). More recently such analysis has been extended to include the behaviour of solutions to nonlinear problems. This extension led to additional stability requirements for RK methods. Although the class of problems has been widened, the analysis is still restricted to a fixed stepsize. In the case of differential algebraic equations (DAEs), additional order conditions must be satisfied [6] to achieve full classical ODE order and avoid possible order reduction. In this case too, a fixed stepsize analysis is employed. Such analysis may be of only limited use in quantifying the effectiveness of adaptive methods on stiff problems.In this paper we examine the phenomenon of order reduction and its implications on variable-step algorithms. We introduce a global measure of order referred to here as the observed order which is based on the average stepsize over the region of integration. This measure may be better suited to the study of stiff systems, where the stepsize selection algorithm will vary the stepsize considerably over the interval of integration. Observed order gives a better indication of the relationship between accuracy and cost. Using this measure, the observed order reduction will be seen to be less severe than that predicated by fixed stepsize order analysis.Supported by the Information Technology Research Centre of Ontario, and the Natural Science and Engineering Research Council of Canada.     

Index-aware model order reduction for differential-algebraic equations

We introduce a model order reduction (MOR) procedure for differential-algebraic equations, which is based on the intrinsic differential equation contained in the starting system and on the remaining algebraic constraints. The decoupling procedure in differential and algebraic part is based on the projector and matrix chain which leads to the definition of tractability index. The differential part can be reduced by using any MOR method, we use Krylov-based projection methods to illustrate our approach. The reduction on the differential part induces a reduction on the algebraic part. In this paper, we present the method for index-1 differential-algebraic equations. We implement numerically this procedure and show numerical evidence of its validity.     

Arobust algorithm for parametric model order reduction

A robust algorithm for computing reduced-order models of parametric systems is proposed. Theoretical considerations suggest that our algorithm is more robust than previous algorithms based on explicit multi-moment matching. Moreover, numerical simulation results show that the proposed algorithm yields more accurate approximations than traditional non-parametric model reduction methods and parametric model reduction methods based on explicitly computing moments. (© 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim)     

Artificial neural network for discrete model order reduction with substructure preservation

This paper presents a new technique for model order reduction (MOR) that is based on an artificial neural network (ANN) prediction. The ANN-based MOR can be applied for different scale systems with substructure preservation. In the proposed technique, the ANN is implemented for predicting the unknown elements of the reduced order model. Prediction of the ANN architecture is based on minimizing the cost function obtained by the difference between the actual and desired system behaviour. The ANN prediction process is pursued while maintaining the full order substructure in the reduced model. The proposed ANN-based model order reduction method is compared to recently published work on MOR techniques. Simulation results verify the validity of the new MOR technique.     

Two-sided model order reduction for two-directional difference system

This paper defines a two-directional difference system and constructs the projection matrix. Then the original system is projected into the smaller system, and we discuss its moment-matching properties. Next we define the dual system, and discuss the dual relation between the dual system and the original system. Then we can construct the projection matrix with the above mentioned dual relation, and project the dual system into the respectively smaller system, hence derive the moment-matching properties. Finally synthesizing the above two moment-matching properties we obtain the main results that the number of moments matched is twice as much as the number of the generating terms of the constructed projection subspace. We apply this result to the two-sided model order reduction for parameter time delay system, and obtain the result that the reduced system can preserve twice moments as the number of the generating terms of the constructed projection subspace. Finally we derive an algorithm to compute the basis of the subspace involved in the reduction process.     

Updating incomplete factorization preconditioners for model order reduction

When solving a sequence of related linear systems by iterative methods, it is common to reuse the preconditioner for several systems, and then to recompute the preconditioner when the matrix has changed significantly. Rather than recomputing the preconditioner from scratch, it is potentially more efficient to update the previous preconditioner. Unfortunately, it is not always known how to update a preconditioner, for example, when the preconditioner is an incomplete factorization. A recently proposed iterative algorithm for computing incomplete factorizations, however, is able to exploit an initial guess, unlike existing algorithms for incomplete factorizations. By treating a previous factorization as an initial guess to this algorithm, an incomplete factorization may thus be updated. We use a sequence of problems from model order reduction. Experimental results using an optimized GPU implementation show that updating a previous factorization can be inexpensive and effective, making solving sequences of linear systems a potential niche problem for the iterative incomplete factorization algorithm.     

Neural network closures for nonlinear model order reduction

Many reduced-order models are neither robust with respect to parameter changes nor cost-effective enough for handling the nonlinear dependence of complex dynamical systems. In this study, we put forth a robust machine learning framework for projection-based reduced-order modeling of such nonlinear and nonstationary systems. As a demonstration, we focus on a nonlinear advection-diffusion system given by the viscous Burgers equation, which is a prototypical setting of more realistic fluid dynamics applications due to its quadratic nonlinearity. In our proposed methodology the effects of truncated modes are modeled using a single layer feed-forward neural network architecture. The neural network architecture is trained by utilizing both the Bayesian regularization and extreme learning machine approaches, where the latter one is found to be more computationally efficient. A significant emphasis is laid on the selection of basis functions through the use of both Fourier bases and proper orthogonal decomposition. It is shown that the proposed model yields significant improvements in accuracy over the standard Galerkin projection methodology with a negligibly small computational overhead and provide reliable predictions with respect to parameter changes.     

A memory-efficient model order reduction for time-delay systems

Michiels et al. (SIAM J. Matrix Anal. Appl. 32(4):1399–1421, 2011) proposed a Krylov-based model order reduction (MOR) method for time-delay systems. In this paper, we present an efficient process, which requires less memory consumption, to accomplish the model reduction. Memory efficiency is achieved by replacing the classical Arnoldi process in the MOR method with a two-level orthogonalization Arnoldi (TOAR) process. The resulting memory requirement is reduced from quadratic dependency of the reduced order to linear dependency. Besides, this TOAR process can also be applied to reduce the original delay system into a reduced-order delay system. Numerical experiments are given to illustrate the feasibility and effectiveness of our method.     

Cascadic multigrid methods combined with sixth order compact scheme for poisson equation

Based on the extrapolation theory and a sixth order compact difference scheme, new extrapolation interpolation operator and extrapolation cascadic multigrid methods for two dimensional Poisson equation are presented. The new extrapolation interpolation operator is used to provide a better initial value on refined grid. The convergence of the new methods are given. Numerical experiments are shown to illustrate that the new methods have higher accuracy and efficiency.     

Stage reduction on P-stable Numerov type methods of eighth order

We present an implicit hybrid two step method for the solution of second order initial value problem. It costs only six function evaluations per step and attains eighth algebraic order. The method satisfy the P-stability property requiring one stage less. We conclude dealing with implementation issues for the methods of this type and give some first pleasant results from numerical tests.     

Clustering approach to model order reduction of power networks with distributed controllers

This paper considers the network structure preserving model reduction of power networks with distributed controllers. The studied system and controller are modeled as second-order and first-order ordinary differential equations, which are coupled to a closed-loop model for analyzing the dissimilarities of the power units. By transfer functions, we characterize the behavior of each node (generator or load) in the power network and define a novel notion of dissimilarity between two nodes by the \(\mathcal {H}_{2}\)-norm of the transfer function deviation. Then, the reduction methodology is developed based on separately clustering the generators and loads according to their behavior dissimilarities. The characteristic matrix of the resulting clustering is adopted for the Galerkin projection to derive explicit reduced-order power models and controllers. Finally, we illustrate the proposed method by the IEEE 30-bus system example.     

A model order reduction technique for systems with nonlinear frequency dependent damping

Frequency domain solution of systems with frequency dependent damping is a computationally expensive endeavour especially when dealing with large order three-dimensional systems. A moment-matching based reduced order model is proposed in this work which is capable of handling nonlinear frequency dependent damping in second-order systems. In the proposed approach, local linear systems with frequency independent matrices are derived from the original system, and using the principles of the Rational Krylov approach, orthogonal basis vectors are computed from these local systems through the second-order Arnoldi procedure. The system is then projected on to the basis set to obtain a numerically efficient reduced order model, accurate in the entire frequency domain of interest. The proposed approach is also shown to be more accurate than the popular modal projection based multi-model approach of the same order. The proposed tool is applied to the problem of determining the frequency response of an idealised centrifugal compressor impeller with non-viscous (frequency dependent) damping.     

Structure preserving model order reduction of shallow water equations

In this paper, we present two different approaches for constructing reduced‐order models (ROMs) for the two‐dimensional shallow water equation (SWE). The first one is based on the noncanonical Hamiltonian/Poisson form of the SWE. After integration in time by the fully implicit average vector field method, ROMs are constructed with proper orthogonal decomposition(POD)/discrete empirical interpolation method that preserves the Hamiltonian structure. In the second approach, the SWE as a partial differential equation with quadratic nonlinearity is integrated in time by the linearly implicit Kahan's method, and ROMs are constructed with the tensorial POD that preserves the linear‐quadratic structure of the SWE. We show that in both approaches, the invariants of the SWE such as the energy, enstrophy, mass and circulation are preserved over a long period of time, leading to stable solutions. We conclude by demonstrating the accuracy and the computational efficiency of the reduced solutions by a numerical test problem.     

Model order reduction methods and their possible application in compliant mechanisms

Model order reduction (MOR) is commonly used to approximate large-scale linear time-invariant dynamical systems. A new feed unit based on a compliant mechanism consisting of flexure hinges can be described by a discrete system of n ordinary differential equations. A projection framework using modal and Krylov subspace techniques is applied to reduce the order of the system to lower computational cost and make the model feasible for control, analysis and optimization. Single flexure hinges are investigated numerical, analytical and experimental and compared to reduced models via modal and tangential Krylov subspace methods regarding the first eigenfrequency. (© 2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim)     

Runge-Kutta methods without order reduction for linear initial boundary value problems

Summary. It is well-known the loss of accuracy when a Runge–Kutta method is used together with the method of lines for the full discretization of an initial boundary value problem. We show that this phenomenon, called order reduction, is caused by wrong boundary values in intermediate stages. With a right choice, the order reduction can be avoided and the optimal order of convergence in time is achieved. We prove this fact for time discretizations of abstract initial boundary value problems based on implicit Runge–Kutta methods. Moreover, we apply these results to the full discretization of parabolic problems by means of Galerkin finite element techniques. We present some numerical examples in order to confirm that the optimal order is actually achieved. Received July 10, 2000 / Revised version received March 13, 2001 / Published online October 17, 2001     

Iterative versus direct parallel substructuring methods in semiconductor device modelling

The numerical simulation of semiconductor devices is extremely demanding in term of computational time because it involves complex embedded numerical schemes. At the kernel of these schemes is the solution of very ill‐conditioned large linear systems. In this paper, we present the various ingredients of some hybrid iterative schemes that play a central role in the robustness of these solvers when they are embedded in other numerical procedures. On a set of two‐dimensional unstructured mixed finite element problems representative of semiconductor simulation, we perform a fair and detailed comparison between parallel iterative and direct linear solution techniques. We show that iterative solvers can be robust enough to solve the very challenging linear systems that arise in those simulations. Copyright © 2004 John Wiley & Sons, Ltd.     

A modified adaptive-order rational Arnoldi method for model order reduction

A Greedy-type expansion point selection for moment-matching methods in model order reduction mainly depends on the computation of a sequence of reduced order models. Typically, the adaptive-order rational Arnoldi (AORA) method resembles an efficient way for the computation of a Galerkin projection corresponding to a set of expansion points. We will provide an extension of the AORA method, in order to reuse the orthonormal basis from previous calls of the AORA method. (© 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim)