Abstract:We revisit the problem of perturbing a large, i.i.d. random matrix by a finite rank error. It is known that when elements of the i.i.d. matrix have finite fourth moment, then the outlier eigenvalues of the perturbed matrix are close to the outlier eigenvalues of the error, as long as the perturbation is relatively small. We first prove that under a merely second moment condition, for a large class of perturbation matrix with bounded rank and bounded operator norm, the outlier eigenvalues of perturbed matrix still converge to that of the perturbation. We then prove that for a matrix with i.i.d. Bernoulli $(d/n)$ entries or Bernoulli $(d_n/n)$ entries with $d_n=n^{o(1)}$, the same result holds for perturbation matrices with a bounded number of nonzero elements.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is about the eigenvalue behavior of non - Hermitian random matrices under finite - rank perturbations, especially in the cases of heavy - tailed distributions and sparse matrices. Specifically, the paper explores the influence of the finite - rank perturbation matrix $C_n$ on the eigenvalues of $A_n$ when the elements of the random matrix $A_n$ have a finite second - order moment but may not have a finite fourth - order moment. ### Background and Problem Description 1. **Background**: - The non - Hermitian random matrix $A_n$ is an $n\times n$ matrix, and its elements $a_{ij}$ are independently and identically distributed, with a mean of zero and a variance of one. - When $a_{ij}$ has a finite fourth - order moment, it is known that the modulus of the eigenvalues of $n^{-1/2}A_n$ converges almost surely to 1, and its empirical spectral density obeys the circular law. - When $a_{ij}$ has only a finite second - order moment but no finite fourth - order moment, the operator norm of $n^{-1/2}A_n$ tends to infinity, but the modulus of its largest eigenvalue still converges almost surely to 1. 2. **Problem**: - The paper focuses on the eigenvalue behavior of the non - Hermitian random matrix $A_n$ under the finite - rank perturbation $C_n$. - In particular, the paper studies the influence of the perturbation matrix $C_n$ on the eigenvalues of $A_n$ when the elements of $A_n$ have an infinite fourth - order moment or the matrix $A_n$ is a sparse matrix. ### Main Contributions 1. **Main Results**: - The author proves that under the condition of only assuming a finite second - order moment, the influence of the finite - rank perturbation $C_n$ on the eigenvalues of $A_n$ is consistent with the results of previous assumptions of a finite fourth - order moment. - Specifically, for the finite - rank perturbation $C_n$, if the operator norm of $C_n$ is bounded and its rank is finite, then the outlier eigenvalues (i.e., eigenvalues with a modulus greater than 1) of $n^{-1/2}A_n + C_n$ converge almost surely to the eigenvalues of $C_n$. 2. **Technical Challenges**: - Since the elements of $A_n$ have only a finite second - order moment, the operator norm $n^{-1/2}A_n$ is unbounded, so methods based on the operator norm cannot be directly used. - The author uses the characteristic function method to deal with this problem and proves the convergence of the characteristic function. 3. **Sparse Matrix Case**: - The paper also studies the finite - rank perturbation of the sparse non - Hermitian random matrix $A_n$. - The author proves that in the case of sparse matrices, the behavior of outlier eigenvalues is similar to the general case, but there are some specific differences. ### Conclusion The main contribution of this paper lies in extending the finite - rank perturbation theory to make it applicable to non - Hermitian random matrices with infinite fourth - order moments and sparsity. By using the characteristic function method, the author successfully proves that in these cases, the influence of the finite - rank perturbation on the eigenvalues is consistent with previous research results. This provides new tools and perspectives for understanding and analyzing a wider range of random matrix models.

Finite rank perturbation of non-Hermitian random matrices: heavy tail and sparse regimes

Outliers and bounded rank perturbation for non-Hermitian random band matrices

Approximation of Sums of Locally Dependent Random Variables via Perturbation of Stein Operator

Outliers for deformed inhomogeneous random matrices

Rank-one Perturbation Bounds for Singular Values of Arbitrary Matrices

Finite rank perturbations of normal operators: hyperinvariant subspaces and a problem of Pearcy

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

Outliers of perturbations of banded Toeplitz matrices

Deviation of top eigenvalue for some tridiagonal matrices under various moment assumptions

Matrix Perturbation: Davis-Kahan in the Infinity Norm

Perturbation and Inverse Problems of Stochastic Matrices

The rank of sparse random matrices

Estimation of high-dimensional low-rank matrices

New matrix perturbation bounds via combinatorial expansion I: Perturbation of eigenspaces

A Riemann--Hilbert approach to the perturbation theory for orthogonal polynomials: Applications to numerical linear algebra and random matrix theory

A Schatten-$q$ Low-rank Matrix Perturbation Analysis via Perturbation Projection Error Bound

Large deviations for the largest singular value of sparse non-Hermitian matrices

Wegner estimate and upper bound on the eigenvalue condition number of non-Hermitian random matrices

Expansion of eigenvalues of rank-one perturbations of the discrete bilaplacian

Eigenvalue and pseudospectrum processes generated by nonnormal Toeplitz matrices with rank 1 perturbations

Probabilistic perturbation bounds of matrix decompositions