Abstract:Given a nonlinear matrix-valued function $F(\lambda)$ and approximate eigenpairs $(\lambda_i, v_i)$, we discuss how to determine the smallest perturbation $\delta F$ such that $[F + \delta F](\lambda_i) v_i = 0$; we call the distance between the $F$ and $F + \delta F$ the backward error for this set of approximate eigenpairs. We focus on the case where $F(\lambda)$ is given as a linear combination of scalar functions multiplying matrix coefficients $F_i$, and the perturbation is done on the matrix coefficients. We provide inexpensive upper bounds, and a way to accurately compute the backward error by means of direct computations or through Riemannian optimization. We also discuss how the backward error can be determined when the $F_i$ have particular structures (such as symmetry, sparsity, or low-rank), and the perturbations are required to preserve them. For special cases (such as for symmetric coefficients), explicit and inexpensive formulas to compute the $\delta F_i$ are also given.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is about the calculation of the backward error of multiple eigenpairs in nonlinear eigenvalue problems (NEPs). Specifically, the author has studied how to determine the minimum perturbation $\delta F$ such that a given approximate eigenpair becomes an exact solution, and has defined the distance of this perturbation as the backward error. ### Core problems of the paper 1. **Definition and calculation of backward error**: For a given nonlinear matrix function $F(\lambda)$ and an approximate eigenpair $(\lambda_i, v_i)$, how to determine the minimum perturbation $\delta F$ such that $[F + \delta F](\lambda_i)v_i = 0$ holds. This distance is defined as the backward error. 2. **Handling structured and unstructured nonlinear eigenvalue problems**: The author has not only considered the general form of nonlinear eigenvalue problems, but also paid special attention to the situation when the matrix coefficients $F_i$ have specific structures (such as symmetry, sparsity or low - rank), and required that the perturbation also preserve these structures. 3. **Providing computable upper bounds and numerical methods**: In order to evaluate the backward error, the author has provided several inexpensive upper - bound estimation methods and proposed a numerical method based on Riemannian optimization to accurately calculate the backward error. ### Formula summary - **Definition of backward error**: \[ \eta := \min \left\{ \left\| [\delta F_1, \ldots, \delta F_k] \right\|_F \mid \sum_{j = 1}^k f_j(\hat{\lambda}_i) (F_j + \delta F_j) \hat{v}_i = 0, \quad 1\leq i\leq p \right\} \] - **Structured backward error**: \[ \eta_S := \min \left\{ \left\| [\delta F_1, \ldots, \delta F_k] \right\|_F \mid \sum_{j = 1}^k (F_j + \delta F_j) V f_j(\Lambda) = 0, \quad \delta F_j\in S_j, \quad j = 1, \ldots, k \right\} \] - **Special treatment for symmetric matrices**: For symmetric matrices $F_j$, specific perturbation formulas can be obtained through QR decomposition and solving some linear equations. ### Numerical methods - **Riemannian optimization**: For nonlinear structures (such as low - rank matrices), the author proposes to use the Riemannian optimization method to calculate the backward error. This involves minimizing an objective function on the manifold $S_j$, where $S_j$ is a set of matrices with specific structures. ### Summary The main contribution of this paper lies in providing a calculation method for the backward error of multiple eigenpairs in nonlinear eigenvalue problems, including computable upper bounds in general and structured cases as well as numerical methods. This is very useful for evaluating the accuracy of numerical algorithms, especially when dealing with matrices with specific structures.

Backward errors for multiple eigenpairs in structured and unstructured nonlinear eigenvalue problems

Structured backward error analysis for sparse polynomial eigenvalue problems.

Structured eigenvalue backward errors for rational matrix functions with symmetry structures

On backward errors of structured polynomial eigenproblems solved by structure preserving linearizations

Structured eigenvalue/eigenvector backward errors of matrix pencils arising in optimal control

Structured eigenvalue backward errors for rational matrix polynomials with symmetry structures

Structured Backward Errors for Eigenvalues of Linear Port-Hamiltonian Descriptor Systems

Eigenvalue backward errors of Rosenbrock systems and optimization of sums of Rayleigh quotient

Towards backward perturbation bounds for approximate dual Krylov subspaces

Pseudospectra, stability radii and their relationship with backward error for structured nonlinear eigenvlaue problems

An overview of a posteriori error estimation and post-processingmethods for nonlinear eigenvalue problems

A posteriori error control for finite element approximations of elliptic eigenvalue problems

Structured singular values and their application in computing eigenvalue backward errors of the Rosenbrock system matrix

An Inexact Cayley Transform Method for Inverse Eigenvalue Problems with Multiple Eigenvalues

Entrywise error bounds for low-rank approximations of kernel matrices

Riemannian Linearized Proximal Algorithms for Nonnegative Inverse Eigenvalue Problem.

Avoiding discretization issues for nonlinear eigenvalue problems

Structured backward error analysis of linearized structured polynomial eigenvalue problems

Backward error analysis for variational discretisations of partial differential equations

Disguised and new Quasi-Newton methods for nonlinear eigenvalue problems

On matrix balancing and eigenvector computation