Kaczmarz-Type Methods for Solving Matrix Equations

Weiguo Li,Wendi Bao,Lili Xing,Zhiwei Guo

2023-05-31

Abstract:In this paper, several Kaczmarz-type numerical methods for solving the matrix equation $AX=B$ and $XA=C$ are proposed, where the coefficient matrix $A$ may be full rank or rank deficient. These methods are iterative methods without matrix multiplication. Theoretically, the convergence of these methods is proved. The numerical results show that these methods are more efficient than iterative methods involving matrix multiplication for high-dimensional matrices.

Numerical Analysis

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the solution of matrix equations, especially for matrix equations in the forms of \(AX = B\) and \(XA = C\). The coefficient matrix \(A\) in these equations may be full - rank or rank - deficient. The paper proposes a series of numerical methods based on the Kaczmarz method and the Coordinate Descent method to solve these problems. These methods do not require matrix multiplication operations, and thus are more efficient than traditional iterative methods when dealing with high - dimensional matrices.

Specifically, the main contributions of the paper include:

1. **Proposing new iterative algorithms**: The paper proposes several Kaczmarz - type iterative algorithms for solving matrix equations \(AX = B\) and \(XA = C\). These algorithms gradually approximate the solutions through row - by - row or column - by - column orthogonal projections without directly performing matrix multiplication operations.

2. **Theoretical convergence proof**: The paper theoretically proves the convergence of the proposed algorithms. For consistent matrix equations, the paper proves that the algorithms can linearly converge to the minimum F - norm solution; for inconsistent matrix equations, the paper proves that the algorithms can linearly converge to the least - squares solution.

3. **Extended Kaczmarz method and extended coordinate descent method**: The paper also proposes an extended Kaczmarz method and an extended coordinate descent method, which are applicable to various types of linear equations (consistent or inconsistent, over - determined or under - determined, and the coefficient matrix may be full - rank or rank - deficient). These methods can be used to calculate the Moore - Penrose generalized inverse of any matrix.

4. **Recursive methods**: When new data is added to the existing matrix equations, the paper proposes some recursive methods to update the solutions without having to recalculate from the beginning. These methods are particularly useful when dealing with dynamic data sets.

5. **Numerical experiments**: To verify the effectiveness of the proposed algorithms, the paper conducts a large number of numerical experiments and compares them with existing methods. The experimental results show that the proposed algorithms are more efficient when dealing with high - dimensional matrices.

In summary, the main objective of this paper is to provide a series of efficient numerical methods to solve the problem of matrix equation solutions, especially when dealing with high - dimensional and complex data, these methods can significantly improve the computational efficiency.