Abstract:This paper is concerned with the extraction of the smallest eigenvalue and the corresponding eigenvector of a symmetric positive definite matrix pencil. We reveal implicit convexity of the eigenvalue problem in Euclidean space. A provable accelerated eigensolver based on preconditioning and implicit convexity (EPIC) is proposed. Theoretical analysis shows the acceleration of EPIC with the rate of convergence resembling the expected rate of convergence of the well-known locally optimal preconditioned conjugate gradient (LOPCG). A complete proof of the expected rate of convergence of LOPCG is elusive so far. Numerical results confirm our theoretical findings of EPIC.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to extract the minimum eigenvalue of a symmetric positive - definite matrix pencil and its corresponding eigenvector. Specifically, the authors revealed the implicit convexity of the eigenvalue problem in Euclidean space and proposed a provably accelerated eigenvalue solver (EPIC) based on pre - conditioning and implicit convexity. They hope that through this method, the solution efficiency can be significantly improved, achieving a convergence rate similar to that of the locally optimal pre - conditioned conjugate gradient (LOBPCG) method. ### Problem Description The generalized eigenvalue problem considered in the paper is: \[ Au = Mu\lambda, \] where \( A \) and \( M \) are given \( n\times n \) symmetric positive - definite matrices, and \((\lambda, u)\) is the required eigenpair. This problem has an important position in scientific and engineering computing, and there are already multiple algorithms for calculating eigenvalues and their corresponding eigenvectors. ### Main Contributions 1. **Revealing Implicit Convexity**: The authors revealed the implicit convexity of the eigenvalue problem in Euclidean space, which provides a theoretical basis for accelerating eigenvalue solving. 2. **Proposing the EPIC Algorithm**: Based on pre - conditioning and implicit convexity, a new accelerated eigenvalue solver EPIC was proposed. 3. **Theoretical Analysis and Verification**: Through strict theoretical analysis, it was proved that the convergence rate of EPIC is similar to that of LOBPCG, and these theoretical results were verified through numerical experiments. ### Method Overview - **Implicit Convexity**: By constructing an auxiliary problem for the eigenvalue problem, its implicit convexity was revealed in Euclidean space. - **Acceleration Mechanism**: Combining the Nesterov Accelerated Gradient (NAG) flow and pre - conditioning techniques, a new Locally Optimal Nesterov Accelerated Gradient Descent method (LONAG) was designed and further developed into the EPIC algorithm. - **Convergence Analysis**: By introducing a discrete Lyapunov function, the convergence and acceleration effects of the LONAG and EPIC algorithms were proved. ### Numerical Results The numerical results provided in the paper show that the EPIC algorithm does achieve the expected acceleration effect in practical applications, verifying the correctness of the theoretical analysis. ### Summary The main purpose of this paper is to propose an efficient accelerated eigenvalue solver EPIC by revealing the implicit convexity of the eigenvalue problem, and to verify its effectiveness through strict theoretical analysis and numerical experiments. This result is expected to provide new ideas and methods for the efficient solution of large - scale eigenvalue problems.

EPIC: a provable accelerated Eigensolver based on Preconditioning and Implicit Convexity

Riemannian Linearized Proximal Algorithms for Nonnegative Inverse Eigenvalue Problem.

Convergence analysis of a block preconditioned steepest descent eigensolver with implicit deflation

Riemannian Acceleration with Preconditioning for symmetric eigenvalue problems

An Indefinite Variant of LOBPCG for Definite Matrix Pencils

On Substochastic Inverse Eigenvalue Problems with the Corresponding Eigenvector Constraints

IMPLICIT DETERMINANT METHOD FOR SOLVING AN HERMITIAN EIGENVALUE OPTIMIZATION PROBLEM

Fast and backward stable computation of the eigenvalues and eigenvectors of matrix polynomials

A Refined Block Conjugate Gradient Method for Symmetric Eigenvalue Problems

A Feasible Method for Optimization with Orthogonality Constraints

Accelerating Convergence by Augmented Rayleigh-Ritz Projections for Large-Scale Eigenpair Computation

Matrix Pencil Optimal Iterative Algorithms and Restarted Versions for Linear Matrix Equation and Pseudoinverse

Optimal Convergence Rate of Self-Consistent Field Iteration for Solving Eigenvector-dependent Nonlinear Eigenvalue Problems

Orthogonal Constrained Neural Networks for Solving Structured Inverse Eigenvalue Problems

An Eigenvalue Decomposition Based Branch-and-bound Algorithm for Nonconvex Quadratic Programming Problems with Convex Quadratic Constraints

Advances in Solution of Classical Generalized Eigenvalue Problem

A mixed precision LOBPCG algorithm

On a General Solution of Partially Described Inverse Quadratic Eigenvalue Problems and Its Applications

A practical, fast method for solving sum-of-squares problems for very large polynomials

Fast Convex Quadratic Optimization Solvers with Adaptive Sketching-based Preconditioners