Abstract:Gradient-type iterative methods for solving Hermitian eigenvalue problems can be accelerated by using preconditioning and deflation techniques. A preconditioned steepest descent iteration with implicit deflation (PSD-id) is one of such methods. The convergence behavior of the PSD-id is recently investigated based on the pioneering work of Samokish on the preconditioned steepest descent method (PSD). The resulting non-asymptotic estimates indicate a superlinear convergence of the PSD-id under strong assumptions on the initial guess. The present paper utilizes an alternative convergence analysis of the PSD by Neymeyr under much weaker assumptions. We embed Neymeyr's approach into the analysis of the PSD-id using a restricted formulation of the PSD-id. More importantly, we extend the new convergence analysis of the PSD-id to a practically preferred block version of the PSD-id, or BPSD-id, and show the cluster robustness of the BPSD-id. Numerical examples are provided to validate the theoretical estimates.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: **How to analyze and improve the convergence behavior of the Block Preconditioned Steepest Descent Eigensolver with Implicit Deflation (BPSD - id), especially the convergence and robustness in computing the generalized Hermitian eigenvalue problem**.
Specifically, the paper focuses on:
1. **Background of the eigenvalue problem**: In many application fields, determining the smallest eigenvalue of a self - adjoint elliptic partial differential operator and its corresponding eigenfunction is a common task. Due to the large dimension of the discretized problem, the computational cost is usually high.
2. **Limitations of existing methods**: Although existing gradient - type iterative methods (such as the Preconditioned Steepest Descent method, PSD) can be accelerated by preconditioning and deflation techniques, in some cases (for example, when the target eigenvalue is closely clustered with other spectral lines), the convergence speed of these methods may be affected.
3. **Research objectives**: By introducing new convergence analysis methods, especially the convergence analysis under Neymeyr's weak assumptions, to improve the convergence behavior of PSD - id and BPSD - id, and to demonstrate the robustness of BPSD - id in dealing with clustered eigenvalue problems.
### Main contributions of the paper
- **New convergence analysis**: Using the convergence analysis method under Neymeyr's weak assumptions, the convergence behavior of PSD - id is re - analyzed.
- **Extension to the block version**: The new convergence analysis is extended to the block version of BPSD - id, and its robustness in dealing with clustered eigenvalue problems is demonstrated.
- **Theoretical and numerical verification**: Numerical experiments are provided to verify the accuracy of the theoretical estimates, thus gaining a deeper understanding of the convergence behavior of PSD - id and BPSD - id.
### Formula summary
The key formulas involved in the paper include:
- **Rayleigh quotient**:
\[
\rho(z)=\frac{z^{*}Hz}{z^{*}Sz}
\]
- **Preconditioned Steepest Descent iteration**:
\[
z^{(\ell + 1)}=z^{(\ell)}-\omega^{(\ell)}Kr^{(\ell)}
\]
where \(K\) is the preconditioning matrix, and \(r^{(\ell)} = Hz^{(\ell)}-\rho(z^{(\ell)})Sz^{(\ell)}\) is the residual.
- **Convergence factor**:
\[
\kappa=\left(\frac{\lambda_{i}-\sigma}{\lambda_{i + 1}-\sigma}\right)\left(\frac{\lambda_{n}-\lambda_{i + 1}}{\lambda_{n}-\lambda_{i}}\right)
\]
where \(\sigma\) is the shift parameter.
Through these analyses, the paper provides a more efficient and robust method for solving large - scale eigenvalue problems.