Abstract:Thanks to its great potential in reducing both computational cost and memory requirements, combining sketching and Krylov subspace techniques has attracted a lot of attention in the recent literature on projection methods for linear systems, matrix function approximations, and eigenvalue problems. Applying this appealing strategy in the context of linear matrix equations turns out to be far more involved than a straightforward generalization. These difficulties include analyzing well-posedness of the projected problem and deriving possible error estimates depending on the sketching properties. Further computational complications include the lack of a natural residual norm estimate and of an explicit basis for the generated subspace. In this paper we propose a new sketched-and-truncated polynomial Krylov subspace method for Sylvester equations that aims to address all these issues. The potential of our novel approach, in terms of both computational time and storage demand, is illustrated with numerical experiments. Comparisons with a state-of-the-art projection scheme based on rational Krylov subspaces are also included.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is to improve the efficiency of solving large - scale Sylvester matrix equations. Specifically, the author aims to improve the performance of the polynomial Krylov subspace method in solving Sylvester equations by combining the randomized sketching technique and the truncated Krylov subspace method.
### Background problems
The general form of the Sylvester matrix equation is:
\[ AX + XB = C_1C_2^T, \]
where \( A \in \mathbb{R}^{n_1 \times n_1} \), \( B \in \mathbb{R}^{n_2 \times n_2} \), and the right - hand side term is assumed to be low - rank, that is, \( C_1 \in \mathbb{R}^{n_1 \times r} \), \( C_2 \in \mathbb{R}^{n_2 \times r} \), where \( r \ll \max\{n_1, n_2\} \). Such equations are widely used in model order reduction, discretization of partial differential equations, and linearization of nonlinear problems.
### Limitations of existing methods
The traditional polynomial Krylov subspace method has the problem of slow convergence when solving Sylvester equations. It requires a large number of iterations to converge, resulting in high computational costs and large memory requirements. In addition, the cost of the orthogonalization step increases significantly with the number of iterations, becoming a performance bottleneck.
### The method proposed in the paper
To solve these problems, the author proposes a new "sketched - and - truncated" polynomial Krylov subspace method. The main innovations include:
1. **Combining the sketching technique**: By using a randomized sketching matrix (such as ε - subspace embedding), large - scale data can be effectively processed while maintaining a relatively low computational cost.
2. **Truncating the Krylov subspace**: Only the most recent several blocks are orthogonalized instead of all previous blocks, thereby reducing storage requirements and computational complexity.
3. **Two - step strategy**: Avoid storing the entire basis vector. After reaching the predetermined accuracy, the final solution is constructed through two passes, further saving memory.
### Performance improvement
Through these improvements, the new method not only performs well in terms of computational time and storage requirements but also can effectively deal with the convergence delay problem in traditional methods. Numerical experiments show that this method outperforms the existing projection methods based on rational Krylov subspaces in various test scenarios.
In conclusion, the goal of this paper is to design an efficient and reliable projection process to significantly improve the performance of the polynomial Krylov subspace method in solving large - scale Sylvester matrix equations.