arXiv subjects

Emre Mengi

Publications and source records attributed to Emre Mengi.

18 recordsLinked to original sources

arXiv2025

Approximation of the Pseudospectral Abscissa via Eigenvalue Perturbation Theory

Reliable and efficient computation of the pseudospectral abscissa in the large-scale setting is still not settled. Unlike the small-scale setting where there are globally convergent criss-cross algorithms, all algorithms in the large-scale setting proposed to date are at best locally convergent. We first describe how eigenvalue perturbation theory can be put in use to estimate the globally rightmost point in the $\epsilon$-pseudospectrum if $\epsilon$ is small. Our treatment addresses both general nonlinear eigenvalue problems, and the standard eigenvalue problem as a special case. For small $\epsilon$, the estimates by eigenvalue perturbation theory are quite accurate. In the standard eigenvalue case, we even derive a formula with an ${\mathcal O}(\epsilon^3)$ error. For larger $\epsilon$, the estimates can be used to initialize the locally convergent algorithms. We also propose fixed-point iterations built on the the perturbation theory ideas for large $\epsilon$ that are suitable for the large-scale setting. The proposed fixed-point iterations initialized by using eigenvalue perturbation theory converge to the globally rightmost point in the pseudospectrum in a vast majority of the cases that we experiment with.

math.NA↗

arXiv2024

Uniform Approximation of Eigenproblems of a Large-Scale Parameter-Dependent Hermitian Matrix

We consider the uniform approximation of the smallest eigenvalue of a large parameter-dependent Hermitian matrix by that of a smaller counterpart obtained through projections. The projection subspaces are constructed iteratively by means of a greedy strategy; at each iteration the parameter where a surrogate error is maximal is computed and the eigenvectors associated with the smallest eigenvalues at the maximizing parameter value are added to the subspace. Unlike the classical approaches, such as the successive constraint method, that maximize such surrogate errors over a discrete and finite set, we maximize the surrogate error over the continuum of all permissible parameter values globally. We formally prove that the projected eigenvalue function converges to the actual eigenvalue function uniformly. In the second part, we focus on the uniform approximation of the smallest singular value of a large parameter-dependent matrix, in case it is non-Hermitian. The proposed frameworks on numerical examples, including those arising from discretizations of parametric PDEs, reduce the size of the large matrix-valued function drastically, while retaining a high accuracy over all permissible parameter values.

math.NA↗

arXiv2024

Minimization of the Pseudospectral Abscissa of a Quadratic Matrix Polynomial

For a quadratic matrix polynomial dependent on parameters and a given tolerance $\epsilon > 0$, the minimization of the $\epsilon$-pseudospectral abscissa over the set of permissible parameter values is discussed, with applications in damping optimization and brake squeal reductions in mind. An approach is introduced that is based on nonsmooth and global optimization (or smooth optimization techniques such as BFGS if there are many parameters) equipped with a globally convergent criss-cross algorithm to compute the $\epsilon$-pseudospectral abscissa objective when the matrix polynomial is of small size. For the setting when the matrix polynomial is large, a subspace framework is introduced, and it is argued formally that it solves the minimization problem globally. The subspace framework restricts the parameter-dependent matrix polynomial to small subspaces, and thus solves the minimization problem for such restricted small matrix polynomials. It then expands the subspaces using the minimizers for the restricted polynomials. The proposed approach makes the global minimization of the $\epsilon$-pseudospectral abscissa possible for a quadratic matrix polynomial dependent on a few parameters and for sizes up to at least a few hundreds. This is illustrated on several examples originating from damping optimization.

math.NA↗

arXiv2024

Large-Scale Minimization of the Pseudospectral Abscissa

This work concerns the minimization of the pseudospectral abscissa of a matrix-valued function dependent on parameters analytically. The problem is motivated by robust stability and transient behavior considerations for a linear control system that has optimization parameters. We describe a subspace procedure to cope with the setting when the matrix-valued function is of large size. The proposed subspace procedure solves a sequence of reduced problems obtained by restricting the matrix-valued function to small subspaces, whose dimensions increase gradually. It possesses desirable features such as a superlinear convergence exhibited by the decay in the errors of the minimizers of the reduced problems. In mathematical terms, the problem we consider is a large-scale nonconvex minimax eigenvalue optimization problem such that the eigenvalue function appears in the constraint of the inner maximization problem. Devising and analyzing a subspace framework for the minimax eigenvalue optimization problem at hand with the eigenvalue function in the constraint require special treatment that makes use of a Lagrangian and dual variables. There are notable advantages in minimizing the pseudospectral abscissa over maximizing the distance to instability or minimizing the $\mathcal{H}_\infty$ norm; the optimized pseudospectral abscissa provides quantitative information about the worst-case transient growth, and the initial guesses for the parameter values to optimize the pseudospectral abscissa can be arbitrary, unlike the case to optimize the distance to instability and $\mathcal{H}_\infty$ norm that would normally require initial guesses yielding asymptotically stable systems.

math.NA↗

arXiv2023

A Subspace Framework for ${\mathcal L}_\infty$ Model Reduction

We consider the problem of locating a nearest descriptor system of prescribed reduced order to a descriptor system with large order with respect to the ${\mathcal L}_\infty$ norm. Widely employed approaches such as the balanced truncation and best Hankel norm approximation for this ${\mathcal L}_\infty$ model reduction problem are usually expensive and yield solutions that are not optimal, not even locally. We propose approaches based on the minimization of the ${\mathcal L}_\infty$ objective by means of smooth optimization techniques. As we illustrate, direct applications of smooth optimization techniques are not feasible, since the optimization techniques converge at best at a linear rate requiring too many evaluations of the costly ${\mathcal L}_\infty$-norm objective to be practical. We replace the original large-scale system with a system of smaller order that interpolates the original system at points on the imaginary axis, and minimize the ${\mathcal L}_\infty$ objective after this replacement. The smaller system is refined by interpolating at additional imaginary points determined based on the local minimizer of the ${\mathcal L}_\infty$ objective, and the optimization is repeated. We argue the framework converges at a quadratic rate under smoothness and nondegeneracy assumptions, and describe how asymptotic stability constraints on the reduced system sought can be incorporated into our approach. The numerical experiments on benchmark examples illustrate that the approach leads to locally optimal solutions to the ${\mathcal L}_\infty$ model reduction problem, and the convergence occurs quickly for descriptors systems of order a few ten thousands.

math.NA↗

arXiv2022

Large-Scale Estimation of Dominant Poles of a Transfer Function by an Interpolatory Framework

We focus on the dominant poles of the transfer function of a descriptor system. The transfer function typically exhibits large norm at and near the imaginary parts of the dominant poles. Consequently, the dominant poles provide information about the points on the imaginary axis where the ${\mathcal L}_\infty$ norm of the system is attained, and they are also sometimes useful to obtain crude reduced-order models. For a large-scale descriptor system, we introduce a subspace framework to estimate a prescribed number of dominant poles. At every iteration, the large-scale system is projected into a small system, whose dominant poles can be computed at ease. Then the projection spaces are expanded so that the projected system after subspace expansion interpolates the large-scale system at the computed dominant poles. We prove an at-least-quadratic-convergence result for the framework, and provide numerical results confirming this. On real benchmark examples, the proposed framework appears to be more reliable than SAMDP [IEEE Trans. Power Syst. 21, 1471-1483, 2006], one of the widely used algorithms due to Rommes and Martins for the estimation of the dominant poles.

math.NA↗

arXiv2022

Derivative Interpolating Subspace Frameworks for Nonlinear Eigenvalue Problems

We first consider the problem of approximating a few eigenvalues of a rational matrix-valued function closest to a prescribed target. It is assumed that the proper rational part of the rational matrix-valued function is expressed in the transfer function form $H(s) = C (sI - A)^{-1} B$, where the middle factor is large, whereas the number of rows of $C$ and the number of columns of $B$ are equal and small. We propose a subspace framework that performs two-sided or one-sided projections on the state-space representation of $H(\cdot)$, commonly employed in model reduction and giving rise to a reduced transfer function. At every iteration, the projection subspaces are expanded to attain Hermite interpolation conditions at the eigenvalues of the reduced transfer function closest to the target, which in turn leads to a new reduced transfer function. We prove in theory that, when a sequence of eigenvalues of the reduced transfer functions converges to an eigenvalue of the full problem, it converges at least at a quadratic rate. In the second part, we extend the proposed framework to locate the eigenvalues of a general square large-scale nonlinear meromorphic matrix-valued function $T(\cdot)$, where we exploit a representation $\mathcal{R}(s) = C(s) A(s)^{-1} B(s) - D(s)$ defined in terms of the block components of $T(\cdot)$. The numerical experiments illustrate that the proposed framework is reliable in locating a few eigenvalues closest to the target point, and that, with respect to runtime, it is competitive to established methods for nonlinear eigenvalue problems.

math.NA↗

arXiv2019

A Subspace Framework for ${\mathcal H}_\infty$-Norm Minimization

We deal with the minimization of the ${\mathcal H}_\infty$-norm of the transfer function of a parameter-dependent descriptor system over the set of admissible parameter values. Subspace frameworks are proposed for such minimization problems where the involved systems are of large order. The proposed algorithms are greedy interpolatory approaches inspired by our recent work [Aliyev et al., SIAM J. Matrix Anal. Appl., 38(4):1496--1516, 2017] for the computation of the ${\mathcal H}_\infty$-norm. In this work, we minimize the ${\mathcal H}_\infty$-norm of a reduced-order parameter-dependent system obtained by two-sided restrictions onto certain subspaces. Then we expand the subspaces so that Hermite interpolation properties hold between the full and reduced-order system at the optimal parameter value for the reduced order system. We formally establish the superlinear convergence of the subspace frameworks under some smoothness assumptions. The fast convergence of the proposed frameworks in practice is illustrated by several large-scale systems.

math.NA↗

arXiv2018

Computation of Stability Radii for Large-Scale Dissipative Hamiltonian Systems

A linear time-invariant dissipative Hamiltonian (DH) system x' = (J-R)Q x, with a skew-Hermitian J, an Hermitian positive semi-definite R, and an Hermitian positive definite Q, is always Lyapunov stable and under weak further conditions even asymptotically stable. In various applications there is uncertainty on the system matrices J, R, Q, and it is desirable to know whether the system remains asymptotically stable uniformly against all possible uncertainties within a given perturbation set. Such robust stability considerations motivate the concept of stability radius for DH systems, i.e., what is the maximal perturbation permissible to the coefficients J, R, Q, while preserving the asymptotic stability. We consider two stability radii, the unstructured one where J, R, Q are subject to unstructured perturbation, and the structured one where the perturbations preserve the DH structure. We employ characterizations for these radii that have been derived recently in [SIAM J. Matrix Anal. Appl., 37, pp. 1625-1654, 2016] and propose new algorithms to compute these stability radii for large scale problems by tailoring subspace frameworks that are interpolatory and guaranteed to converge at a super-linear rate in theory. At every iteration, they first solve a reduced problem and then expand the subspaces in order to attain certain Hermite interpolation properties between the full and reduced problems. The reduced problems are solved by means of the adaptations of existing level-set algorithms for H-infinity norm computation in the unstructured case, while, for the structured radii, we benefit from algorithms that approximate the objective eigenvalue function with a piece-wise quadratic global underestimator. The performance of the new approaches is illustrated with several examples including a system that arises from a finite-element modeling of an industrial disk brake.

math.NA↗

arXiv2018

Nonsmooth Rate-of-Convergence Analyses of Algorithms for Eigenvalue Optimization

Non-smoothness at optimal points is a common phenomenon in many eigenvalue optimization problems. We consider two recent algorithms to minimize the largest eigenvalue of a Hermitian matrix dependent on one parameter, both proven to be globally convergent unaffected by non-smoothness. One of these models the eigenvalue function with a piece-wise quadratic function, and effective in dealing with non-convex problems. The other projects the Hermitian matrix into subspaces formed of eigenvectors, and effective in dealing with large-scale problems. We generalize the latter slightly to cope with non-smoothness. For both algorithms, we analyze the rate-of-convergence in the non-smooth setting, when the largest eigenvalue is multiple at the minimizer and zero is strictly in the interior of the generalized Clarke derivative, and prove that both algorithms converge rapidly. The algorithms are applied to, and the deduced results are illustrated on the computation of the inner numerical radius, the modulus of the point on the boundary of the field of values closest to the origin, which carries significance for instance for the numerical solution of a definite generalized symmetric eigenvalue problem.

math.NA↗

arXiv2018

Large-Scale and Global Maximization of the Distance to Instability

The larger the distance to instability from a matrix is, the more robustly stable the associated autonomous dynamical system is in the presence of uncertainties and typically the less severe transient behavior its solution exhibits. Motivated by these issues, we consider the maximization of the distance to instability of a matrix dependent on several parameters, a nonconvex optimization problem that is likely to be nonsmooth. In the first part we propose a globally convergent algorithm when the matrix is of small size and depends on a few parameters. In the second part we deal with the problems involving large matrices. We tailor a subspace framework that reduces the size of the matrix drastically. The strength of the tailored subspace framework is proven with a global convergence result as the subspaces grow and a superlinear rate-of-convergence result with respect to the subspace dimension.

math.NA↗

arXiv2018

Subspace Methods for 3-Parameter Eigenvalue Problems

We propose subspace methods for 3-parameter eigenvalue problems. Such problems arise when separation of variables is applied to separable boundary value problems; a particular example is the Helmholtz equation in ellipsoidal and paraboloidal coordinates. While several subspace methods for 2-parameter eigenvalue problems exist, their extensions to three parameter setting seem to be challenging. An inherent difficulty is that, while for 2-parameter eigenvalue problems we can exploit a relation to Sylvester equations to obtain a fast Arnoldi type method, such a relation does not seem to exist when there are three or more parameters. Instead, we introduce a subspace iteration method with projections onto generalized Krylov subspaces that are constructed from scratch at every iteration using certain Ritz vectors as the initial vectors. Another possibility is a Jacobi--Davidson type method for three or more parameters, which we generalize from its 2-parameter counterpart. For both approaches, we introduce a selection criterion for deflation that is based on the angles between left and right eigenvectors. The Jacobi--Davidson approach is devised to locate eigenvalues close to a prescribed target, yet it often also performs well when eigenvalues are sought based on the proximity of one of the components to a prescribed target. The subspace iteration method is devised specifically for the latter task. The proposed approaches are suitable especially for problems where the computation of several eigenvalues is required with high accuracy. Matlab implementations of both methods have been made available in the package MultiParEig.

math.NA↗

arXiv2017

Large-Scale Computation of ${\mathcal L}_\infty$-Norms by a Greedy Subspace Method

We are concerned with the computation of the ${\mathcal L}_\infty$-norm for an ${\mathcal L}_\infty$-function of the form $H(s) = C(s) D(s)^{-1} B(s)$, where the middle factor is the inverse of a meromorphic matrix-valued function, and $C(s),\, B(s)$ are meromorphic functions mapping to short-and-fat and tall-and-skinny matrices, respectively. For instance, transfer functions of descriptor systems and delay systems fall into this family. We focus on the case where the middle factor is large-scale. We propose a subspace projection method to obtain approximations of the function $H$ where the middle factor is of much smaller dimension. The ${\mathcal L}_\infty$-norms are computed for the resulting reduced functions, then the subspaces are refined by means of the optimal points on the imaginary axis where the ${\mathcal L}_\infty$-norm of the reduced function is attained. The subspace method is designed so that certain Hermite interpolation properties hold between the largest singular values of the original and reduced functions. This leads to a locally superlinearly convergent algorithm with respect to the subspace dimension, which we prove and illustrate on various numerical examples.

math.NA↗

arXiv2015

A Subspace Method for Large Scale Eigenvalue Optimization

We consider the minimization or maximization of the $J$th largest eigenvalue of an analytic and Hermitian matrix-valued function, and build on Mengi et al. (2014, SIAM J. Matrix Anal. Appl., 35, 699-724). This work addresses the setting when the matrix-valued function involved is very large. We describe subspace procedures that convert the original problem into a small-scale one by means of orthogonal projections and restrictions to certain subspaces, and that gradually expand these subspaces based on the optimal solutions of small-scale problems. Global convergence and superlinear rate-of-convergence results with respect to the dimensions of the subspaces are presented in the infinite dimensional setting, where the matrix-valued function is replaced by a compact operator depending on parameters. In practice, it suffices to solve eigenvalue optimization problems involving matrices with sizes on the scale of tens, instead of the original problem involving matrices with sizes on the scale of thousands.

math.NA↗

arXiv2013

A Support Based Algorithm for Optimization with Eigenvalue Constraints

Optimization of convex functions subject to eigenvalue constraints is intriguing because of peculiar analytical properties of eigenvalues, and is of practical interest because of wide range of applications in fields such as structural design and control theory. Here we focus on the optimization of a linear objective subject to a constraint on the smallest eigenvalue of an analytical and Hermitian matrix-valued function. We offer a quadratic support function based numerical solution. The quadratic support functions are derived utilizing the variational properties of an eigenvalue over a set of Hermitian matrices. Then we establish the local convergence of the algorithm under mild assumptions, and deduce a precise rate of convergence result by viewing the algorithm as a fixed point iteration. We illustrate its applicability in practice on the pseudospectral functions.

math.NA↗

arXiv2012

Matrix Polynomials with Specified Eigenvalues

This work concerns the distance in 2-norm from a matrix polynomial to a nearest polynomial with a specified number of its eigenvalues at specified locations in the complex plane. Perturbations are allowed only on the constant coefficient matrix. Singular value optimization formulas are derived for these distances facilitating their computation. The singular value optimization problems, when the number of specified eigenvalues is small, can be solved numerically by exploiting the Lipschitzness and piece-wise analyticity of the singular values with respect to the parameters.

math.NA↗

arXiv2011

Numerical Optimization of Eigenvalues of Hermitian Matrix Functions

This work concerns the global minimization of a prescribed eigenvalue or a weighted sum of prescribed eigenvalues of a Hermitian matrix-valued function depending on its parameters analytically in a box. We describe how the analytical properties of eigenvalue functions can be put into use to derive piece-wise quadratic functions that underestimate the eigenvalue functions. These piece-wise quadratic under-estimators lead us to a global minimization algorithm, originally due to Breiman and Cutler. We prove the global convergence of the algorithm, and show that it can be effectively used for the minimization of extreme eigenvalues, e.g., the largest eigenvalue or the sum of the largest specified number of eigenvalues. This is particularly facilitated by the analytical formulas for the first derivatives of eigenvalues, as well as analytical lower bounds on the second derivatives that can be deduced for extreme eigenvalue functions. The applications that we have in mind also include the ${\rm H}_\infty$-norm of a linear dynamical system, numerical radius, distance to uncontrollability and various other non-convex eigenvalue optimization problems, for which, generically, the eigenvalue function involved is simple at all points.

math.NA↗

arXiv2010

Generalized Eigenvalue Problems with Specified Eigenvalues

We consider the distance from a (square or rectangular) matrix pencil to the nearest matrix pencil in 2-norm that has a set of specified eigenvalues. We derive a singular value optimization characterization for this problem and illustrate its usefulness for two applications. First, the characterization yields a singular value formula for determining the nearest pencil whose eigenvalues lie in a specified region in the complex plane. For instance, this enables the numerical computation of the nearest stable descriptor system in control theory. Second, the characterization partially solves the problem posed in [Boutry et al. 2005] regarding the distance from a general rectangular pencil to the nearest pencil with a complete set of eigenvalues. The involved singular value optimization problems are solved by means of BFGS and Lipschitz-based global optimization algorithms.

math.NA↗