Abstract:In this paper, we introduce a randomized algorithm for solving the non-symmetric eigenvalue problem, referred to as randomized Implicitly Restarted Arnoldi (rIRA). This method relies on using a sketch-orthogonal basis during the Arnoldi process while maintaining the Arnoldi relation and exploiting a restarting scheme to focus on a specific part of the spectrum. We analyze this method and show that it retains useful properties of the Implicitly Restarted Arnoldi (IRA) method, such as restarting without adding errors to the Ritz pairs and implicitly applying polynomial filtering. Experiments are presented to validate the numerical efficiency of the proposed randomized eigenvalue solver.
What problem does this paper attempt to address?
This paper aims to solve the non - symmetric eigenvalue problem. Specifically, the paper introduces a randomized Implicitly Restarted Arnoldi (rIRA) method for computing a small number of eigenpairs of non - symmetric matrices. This method uses a sketch - orthogonal basis in the Arnoldi process, while maintaining the Arnoldi relationship, and utilizes a restarting scheme to focus on a specific part of the spectrum.
### Background of the Paper and Problem Definition
**Eigenvalue Problem**: Given a linear transformation \(A\in\mathbb{R}^{n\times n}\), the goal of the eigenvalue problem is to find a set of eigenpairs \((u,\lambda)\), where \(u\in\mathbb{R}^n\), \(\|u\| = 1\), and \(\lambda\in\mathbb{C}\), satisfying:
\[Au=\lambda u.\]
**Eigenvalue Problem for Non - symmetric Matrices**: For a non - symmetric matrix \(A\), computing its eigenpairs is a complex problem, especially when the matrix is large and sparse. Although the traditional Arnoldi method is effective, it encounters problems of memory and computational cost when dealing with large - scale problems.
### Main Contributions of the Paper
1. **Randomized Implicitly Restarted Arnoldi (rIRA) Method**
- **Sketch - Orthogonal Basis**: Use sketching techniques to generate an orthogonal basis, reducing memory and computational cost.
- **Restarting Scheme**: Improve computational efficiency by focusing on a specific part of the spectrum through the restarting scheme.
- **Theoretical Analysis**: Prove that rIRA retains some important properties of the traditional Implicitly Restarted Arnoldi (IRA) method, such as not adding errors to the Ritz pairs and implicitly applying polynomial filtering.
2. **Numerical Experiments**: Verify the numerical efficiency of rIRA through experiments, showing that its performance on different matrices is better than or close to the IRA method implemented in ARPACK, and the time per iteration is shorter.
### Key Technologies and Methods
- **Arnoldi Process**: Construct an orthogonal basis for the Krylov subspace.
- **Restarting Technique**: Restart the Arnoldi process by selecting specific Ritz vectors as new starting vectors.
- **Sketching Technique**: Use a randomized method to generate a sketch - orthogonal basis, reducing computational and storage costs.
### Mathematical Formulas
- **Arnoldi Decomposition**:
\[
AV = VH_k+r_ke_k^T,
\]
where \(V\in\mathbb{R}^{n\times k}\) is the orthogonal basis of the Krylov subspace, \(H_k\in\mathbb{C}^{k\times k}\) is an upper Hessenberg matrix, and \(r_k = h_{k + 1,k}v_{k+1}\) is the residual vector.
- **Sketch - Orthogonal Basis**:
\[
(\Omega V)^T(\Omega V)=I_k,
\]
where \(\Omega\) is a subspace embedding matrix such that \(\Omega V\) maintains orthogonality.
- **Ritz Pairs**:
\[
A\tilde{u}_i-\tilde{\lambda}_i\tilde{u}_i\perp K_k,
\]
where \(\tilde{u}_i = Vy_i\) and \(\tilde{\lambda}_i\) are obtained from the eigenpairs \((y_i,\tilde{\lambda}_i)\) of \(H_k\).
### Conclusion
The paper proposes an effective randomized Implicitly Restarted Arnoldi method (rIR