Abstract:We exhibit a randomized algorithm which given a matrix $A\in \mathbb{C}^{n\times n}$ with $\|A\|\le 1$ and $\delta>0$, computes with high probability an invertible $V$ and diagonal $D$ such that $\|A-VDV^{-1}\|\le \delta$ using $O(T_{MM}(n)\log^2(n/\delta))$ arithmetic operations, in finite arithmetic with $O(\log^4(n/\delta)\log n)$ bits of precision. Here $T_{MM}(n)$ is the number of arithmetic operations required to multiply two $n\times n$ complex matrices numerically stably, known to satisfy $T_{MM}(n)=O(n^{\omega+\eta})$ for every $\eta>0$ where $\omega$ is the exponent of matrix multiplication (Demmel et al., Numer. Math., 2007). Our result significantly improves the previously best known provable running times of $O(n^{10}/\delta^2)$ arithmetic operations for diagonalization of general matrices (Armentano et al., J. Eur. Math. Soc., 2018), and (with regards to the dependence on $n$) $O(n^3)$ arithmetic operations for Hermitian matrices (Dekker and Traub, Lin. Alg. Appl., 1971). It is the first algorithm to achieve nearly matrix multiplication time for diagonalization in any model of computation (real arithmetic, rational arithmetic, or finite arithmetic), thereby matching the complexity of other dense linear algebra operations such as inversion and $QR$ factorization up to polylogarithmic factors. The proof rests on two new ingredients. (1) We show that adding a small complex Gaussian perturbation to any matrix splits its pseudospectrum into $n$ small well-separated components. In particular, this implies that the eigenvalues of the perturbed matrix have a large minimum gap, a property of independent interest in random matrix theory. (2) We give a rigorous analysis of Roberts' Newton iteration method (Roberts, Int. J. Control, 1980) for computing the sign function of a matrix in finite arithmetic, itself an open problem in numerical analysis since at least 1986.

Finding Eigenvectors: Fast and Nontraditional Approach

Online Dominant Generalized Eigenvectors Extraction Via a Randomized Algorithm

Fast computation of spectral densities for generalized eigenvalue problems

Online Generalized Eigenvectors Extraction Via A Fixed-Point Approach

A New Strategy for Directly Calculating the Minimum Eigenvector of Matrices Without Diagonalization.

A greedy algorithm for computing eigenvalues of a symmetric matrix

A Greedy Algorithm for Computing Eigenvalues of a Symmetric Matrix with Localized Eigenvectors

Exact Eigenvalues and Eigenvectors for Some n-Dimensional Matrices

A projected preconditioned conjugate gradient algorithm for computing many extreme eigenpairs of a Hermitian matrix

Locating Eigenvalues of Symmetric Matrices - A Survey

An Inverse Extreme Eigenpair Problem and Its Parallel Iterative Solution for Symmetric Tridiagonal Matrices

Computing Eigenvectors of Normal Matrices with Simple Inverse Iteration

More on the eigenvectors-from-eigenvalues identity

Orthogonal layers of parallelism in large-scale eigenvalue computations

On Eigenvector Approximation of Diagonalizable Random Matrices with Random Perturbations: Properties and Applications

Pseudospectral Shattering, the Sign Function, and Diagonalization in Nearly Matrix Multiplication Time

A new method for the computation of eigenvalues

The Power of Two Matrices in Spectral Algorithms for Community Recovery

Sublinear Time Eigenvalue Approximation via Random Sampling

The Projection Methods for Computing the Pseudospectra of Large Scale Matrices

Randomized methods for computing joint eigenvalues, with applications to multiparameter eigenvalue problems and root finding