Model Order Reduction for Parametric Hermitian Eigenvalue Problems: Local Acceleration with Taylor-Reduced Basis Method

Benjamin Stamm † and Zhuoyao Zeng † · 2026

Read Model Order Reduction for Parametric Hermitian Eigenvalue Problems: Local Acceleration with Taylor-Reduced Basis Method by Benjamin Stamm † and Zhuoyao Zeng † (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-14620.txt.

Introduction

Stamm and Zeng speed up parametric eigenvalue problems with the Taylor-reduced basis method. Large Hermitian matrices with parameters need fast eigenspace estimates. Taylor-RBM builds a local approximation space near a reference point. Derivative data of the spectral projector shape that space. A compact error analysis backs the method. A cheap assembly route builds the reduced basis in practice. The study also compares the basis with raw truncated power series of the eigenprojector. Links to classic perturbation theory run through the analysis.

Catalog record

CREATORBenjamin Stamm † and Zhuoyao Zeng †
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE143919
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.14620

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Model Order Reduction for Parametric Hermitian Eigenvalue Problems: Local Acceleration with Taylor-Reduced Basis Method Benjamin Stamm † and Zhuoyao Zeng † Abstract . This paper is concerned with the Taylor-reduced basis method (Taylor-RBM) for the efficient approximation of eigenspaces of large scale parametric Hermitian matrices. The Taylor-RBM is a local model order reduction method, which constructs an approximation space by capturing derivatives information of the spectral projector at a reference point in the parameter domain. We perform a concise error analysis to justify the Taylor-RBM for eigenvalue problems, and we present a computationally efficient procedure to assemble the Taylor-reduced basis space. Since this method is tightly connected to the classical multivariate analytic perturbation theory, we also provide a detailed analysis of the spectral approximation using the truncated power series of the eigenprojector, and compare this with the approximation obtained from the Taylor-RBM. Keywords: Parametric eigenvalue problem, spectral approximation, model order reduction, reduced basis method, Rayleigh-Ritz method, perturbation theory. AMS subject classification: Primary 41A63, 65F15; secondary 47A55, 65N25, 65F50, 65K10, 81P68. 1. Introduction Let N\in\mathbb{N} with N\gg 1 and \mathcal{P}\subseteq\mathbb{R}^{d} be an open parameter domain. Consider a family of N\times N Hermitian matrices \{\mathbf{A}(\bm{\mu})\}_{\bm{\mu}\in\mathcal{P}} , i.e., for each \bm{\mu}\in\mathcal{P} , the matrix \mathbf{A}(\bm{\mu})\in\mathbb{C}^{N\times N} satisfies \mathbf{A}(\bm{\mu})=\mathbf{A}(\bm{\mu})^{*} . Throughout this work, we assume that the parameter dependency of \mathbf{A}(\bm{\mu}) on \bm{\mu} is (real-)analytic , in the sense specified later in ( 3.1 ). For practical relevance and for simplicity, this work is presented in the finite dimensional matrix setting. There are no fundamental obstacles when formulating the results to analytic families of unbounded self-adjoint linear operators after clarifications of technical details (type of operators, type of spectrum, free from spectral pollution, matrix representations, etc.). Suppose at a reference parameter \bm{\mu}_{0}\in\mathcal{P} , we have computed the smallest eigenvalues \{\lambda_{i}(\bm{\mu}_{0})\}_{i=1}^{K} counting without multiplicities, i.e., \lambda_{1}(\bm{\mu}_{0})<\lambda_{2}(\bm{\mu}_{0})<\cdots<\lambda_{K}(\bm{\mu}_{0})<\lambda_{K+1}(\bm{\mu}_{0})<\cdots , and their corresponding total eigenprojector \{\mathbf{P}_{\!i}(\bm{\mu}_{0})\}_{i=1}^{K} , each with multiplicity m_{i}(\bm{\mu}_{0})\coloneqq\mathrm{rank}\big(\mathbf{P}_{\!i}(\bm{\mu}_{0})\big) , i.e, \ker\big(\mathbf{A}(\bm{\mu}_{0})-\lambda_{i}(\bm{\mu}_{0})\mathbf{I}\big)\,=\,\mathrm{ran}\mathbf{P}_{\!i}(\bm{\mu}_{0})\qquad\text{for }i=1,\ldots,K. Let us denote \mathbf{P}(\bm{\mu}_{0})\coloneqq\sum_{i=1}^{K}\mathbf{P}_{\!i}(\bm{\mu}_{0}) and M\coloneqq\mathrm{rank}\big(\mathbf{P}(\bm{\mu}_{0})\big) . Note that M=m_{1}(\bm{\mu}_{0})+\cdots+m_{K}(\bm{\mu}_{0})\quad\text{and}\quad\sum_{i=1}^{K}\lambda_{i}(\bm{\mu}_{0})m_{i}(\bm{\mu}_{0})=\mathrm{tr}\big(\mathbf{A}(\bm{\mu}_{0})\mathbf{P}(\bm{\mu}_{0})\big). We are interested in approximating the eigenvalue sum and the total eigenprojectors at \bm{\mu} in a small neighborhood of \bm{\mu}_{0} . Equivalently by the variational characterization of the sum of the smallest eigenvalues [ 21 ] , we are interested in approximating the orthogonal projector \mathbf{P}(\bm{\mu}) and its Ritz-value sum \mathrm{tr}\big(\mathbf{A}(\bm{\mu})\mathbf{P}(\bm{\mu})\big) , where \mathbf{P}(\bm{\mu})\quad\text{ solves }\min_{\begin{subarray}{c}\mathbf{M}=\mathbf{M}^{2}=\mathbf{M}^{*}\in\mathbb{C}^{N\times N}\\ \mathrm{tr}\mathbf{M}=M\end{subarray}}\mathrm{tr}\big(\mathbf{A}(\bm{\mu})\,\mathbf{M}\big)\quad\text{for }\bm{\mu}\text{ near }\bm{\mu}_{0}. Since the mapping \bm{\mu}\mapsto\mathbf{A}(\bm{\mu}) is assumed to be analytic over \mathcal{P} , it is natural to expect that the total eigenprojector \mathbf{P}(\bm{\mu}) also depends analytically on \bm{\mu} . Locally, this is indeed the case as established in the classical works [ 34 , 5 ] . Exploiting this analyticity, we shall use the Taylor reduced basis method (Taylor-RBM) to approximate the total eigenprojector \mathbf{P}(\bm{\mu}) and the Ritz-value sum \mathrm{tr}\big(\mathbf{A}(\bm{\mu})\mathbf{P}(\bm{\mu})\big) for \bm{\mu} in a neighborhood of \bm{\mu}_{0} . The main idea is to construct a reduced basis (RB) space \cal{V} with r\coloneqq\dim\cal{V}\ll N and r\geq M , where \cal{V} is spanned by the ranges of the Taylor coefficients of \mathbf{P}(\bm{\mu}) at \bm{\mu}_{0} up to a certain order. Then, we perform a Galerkin projection of the original problem onto \cal{V} , i.e., for a matrix \mathbf{V}\in\mathbb{C}^{N\times r} whose columns form an orthonormal basis (ONB) of \cal{V} and \mathbf{A}^{\!\!\mathbf{V}}\!(\bm{\mu})\coloneqq\mathbf{V}^{*}\mathbf{A}(\bm{\mu})\mathbf{V} , we consider the reduced problem (1.1) \mathbf{Q}^{\!\mathbf{V}}\!(\bm{\mu})\ \text{ solves}\argmin_{\begin{subarray}{c}\mathbf{M}=\mathbf{M}^{2}=\mathbf{M}^{*}\in\mathbb{C}^{r\times r}\\ \mathrm{tr}\mathbf{M}=M\end{subarray}}\mathrm{tr}\big(\mathbf{A}^{\!\!\mathbf{V}}\!(\bm{\mu})\,\mathbf{M}\big), and we view \mathbf{P}(\bm{\mu})\approx\mathbf{Q}_{\mathcal{V}}(\bm{\mu})\coloneqq\mathbf{V}\mathbf{Q}^{\!\mathbf{V}}\!(\bm{\mu})\,\mathbf{V}^{*} for \bm{\mu}_{0} near \bm{\mu} . The motivation of our research is twofold. For one thing, Taylor-RBM is well-studied in RBM theory for parametric linear source problems in model order reduction (MOR). For another, the Taylor-RBM in case of univariate parameter space emerges independently in physics. Our goal is to provide a concise mathematical understanding of this method that bridges the two communities and to extend the method to the multivariate parameter case. We believe this method has great potential for applications involving parametric eigenvalue problems. Our focus is on quantum-mechanical applications in computational physics and chemistry, although engineering applications are also within its scope. This work provides a solid theoretical foundation for its future development. Before stating our main contributions, we summarize related existing works. 1.1. Literature review and state-of-the-art MOR is a vast research area with numerous applications in science and engineering [ 8 , 3 , 50 ] . The general idea of MOR for parametric systems is that given a large parametric full-order model (FOM) that is computationally expensive to solve, one construct in the offline phase a reliable and efficient reduced order model (ROM) that approximates the FOM well in a certain sense, and then use the ROM in the online phase for rapid repeated evaluations [ 10 ] . Among the various MOR techniques, we focus on projection-based MOR , which can be divided into two categories: System-theoretic MOR addresses more on control problems, see [ 7 ] for a detailed survey; reduced basis method (RBM) focuses on reducing the dimensionality of the state space through Galerkin-projection [ 9 ] . RBM is a class of projection-based MOR techniques, proposed originally for solving parametric source problems, i.e., parametric partial differential equations (PDEs) [ 42 , 45 , 49 ] . The key idea is to use information of the computed true solutions at selected parameter values, also called snapshot solutions , to construct a low-dimensional RB space, such that the Galerkin projection onto this space provides a good approximation of the solution set. Following [ 28 ] , there are two main ideas of constructing the RB space: • Lagrange-RBM constructs the RB space by collecting snapshot solutions at selected parameters. There are two common strategies: The proper orthogonal decomposition (POD) refers to computing snapshot solutions of many preselected parameters and then performing a singular value decomposition (SVD) of the resulting large snapshot matrix. Alternatively, one can use a greedy method to iteratively select the most “representative” snapshots from a training set of parameters based on a-posteriori error estimators. In both cases, solid mathematical framework and convergence analysis have been established for linear, coercive continuous and affinely decomposable problems [ 47 , 30 , 16 ] . • Taylor-RBM constructs the RB space by collecting the Taylor coefficients of the parametric solution at a given parameter up to a certain order, provided that the parametric solution is differentiable w.r.t.​ the parameter. Convergence analysis of Taylor-RBM for parametric source problems has been conducted in [ 22 ] . Parametric eigenvalue problems (pEVPs) arises naturally in numerical linear algebra [ 36 , 53 , 33 , 40 , 46 ] , in infinite dimensional setting (especially within finite element approximations of parametric PDE EVP) [ 37 , 2 , 1 , 24 , 31 ] , in uncertainty quantification [ 17 , 52 , 19 , 6 ] , as well as in quantum physics [ 15 , 29 , 23 , 41 , 20 ] . Compared to parametric source problems, pEVPs demonstrate a different nature. The main challenges include (i) the lack of Galerkin orthogonality, (ii) the absence of a reliable residual based error estimator, (iii) the complicated regularity behavior of the eigenvalues and eigenvectors w.r.t.​ the parameter, especially when eigenvalue crossings occur. The first two challenges are intrinsic problem of subspace methods for eigenvalue problems [ 13 , 26 , 4 , 57 ,

Read the complete text