Abstract:In view of the advantages of simplicity and effectiveness of the Kaczmarz method, which was originally employed to solve the large-scale system of linear equations $Ax=b$, we study the greedy randomized block Kaczmarz method (ME-GRBK) and its relaxation and deterministic versions to solve the matrix equation $AXB=C$, which is commonly encountered in the applications of engineering sciences. It is demonstrated that our algorithms converge to the unique least-norm solution of the matrix equation when it is consistent and their convergence rate is faster than that of the randomized block Kaczmarz method (ME-RBK). Moreover, the block Kaczmarz method (ME-BK) for solving the matrix equation $AXB=C$ is investigated and it is found that the ME-BK method converges to the solution $A^{+}CB^{+}+X^{0}-A^{+}AX^{0}BB^{+}$ when it is consistent. The numerical tests verify the theoretical results and the methods presented in this paper are applied to the color image restoration problem to obtain satisfactory restored images.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the solution of the large - scale matrix equation \(AXB = C\), especially the large - scale computational problems encountered in engineering science applications. Specifically, the paper proposes the Greedy Randomized Block Kaczmarz method (ME - GRBK) and its relaxed and deterministic versions for solving the matrix equation \(AXB = C\). These methods aim to overcome the problem of excessive storage and computational capacity requirements of traditional centralized algorithms in large - scale computational problems, and are superior to the Randomized Block Kaczmarz method (ME - RBK) in terms of convergence speed.
### Main contributions:
1. **Convergence of the ME - BK method**: For the first time, the paper studies the convergence of the Block Kaczmarz method (ME - BK) when solving the matrix equation \(AXB = C\), and proves that when the matrix equation is consistent, the ME - BK method converges to the solution \(A^+CB^+ + X_0 - A^+AX_0BB^+\).
2. **Improvement of the ME - GRBK method**: The ME - GRBK method is the greedy version of the ME - RBK method, and the ME - RGRBK and ME - MWRBK methods can be regarded as the relaxed and deterministic versions of the ME - GRBK method. The paper proves that these methods converge to the unique minimum - norm solution \(A^+CB^+\) when the matrix equation is consistent and exhibit superior performance.
3. **Numerical experiment verification**: Numerical experiments verify the theoretical results, indicating that the ME - GRBK, ME - RGRBK and ME - MWRBK methods are more efficient than the ME - RBK method. In addition, the application of these methods in color image restoration problems has also achieved satisfactory results.
### Formula representation:
- Matrix \(A\in\mathbb{R}^{m\times p}\), \(B\in\mathbb{R}^{q\times n}\), \(C\in\mathbb{R}^{m\times n}\).
- Matrix equation \(AXB = C\).
- Iterative formula of the ME - BK method:
\[
X_{k + 1}=X_k+\frac{\alpha}{\|A_{i_k,:}\|^2_2}A_{i_k,:}^T(C_{i_k,:}-A_{i_k,:}X_kB)B^T
\]
- Selection probability of the ME - GRBK method:
\[
\zeta_k=\frac{1}{2}\left(\frac{1}{\|R_k\|^2_F}\max_{1\leq i\leq m}\left(\frac{\|R_{k,i,:}\|^2_2}{\|A_{i,:}\|^2_2}\right)+\frac{1}{\|A\|^2_F}\right)
\]
- Estimation of the convergence rate:
\[
E[\|X_{k + 1}-X^*\|^2_F]\leq\left(1-\frac{(2\alpha-\alpha^2\|B\|^2_2)\phi_k}{\kappa^2_{F,2}(B^T\otimes A)}\right)\|X_k-X^*\|^2_F
\]
Through these methods, the paper provides new and effective tools for solving large - scale matrix equations and demonstrates their superiority in practical applications.