Abstract:The randomized Kaczmarz methods are a popular and effective family of iterative methods for solving large-scale linear systems of equations, which have also been applied to linear feasibility problems. In this work, we propose a new block variant of the randomized Kaczmarz method, B-MRK, for solving linear feasibility problems defined by matrices. We show that B-MRK converges linearly in expectation to the feasible region.Furthermore, we extend the method to solve tensor linear feasibility problems defined under the tensor t-product. A tensor randomized Kaczmarz (TRK) method, TRK-L, is proposed for solving linear feasibility problems that involve mixed equality and inequality constraints. Additionally, we introduce another TRK method, TRK-LB, specifically tailored for cases where the feasible region is defined by linear equality constraints coupled with bound constraints on the variables. We show that both of the TRK methods converge linearly in expectation to the feasible region. Moreover, the effectiveness of our methods is demonstrated through numerical experiments on various Gaussian random data and applications in image deblurring.
What problem does this paper attempt to address?
### Problems the paper attempts to solve
The paper aims to solve linear feasibility problems, especially for linear constraint systems in matrix and tensor forms. Specifically, the paper proposes several new randomized Kaczmarz methods to solve the following types of problems:
1. **Matrix linear feasibility problems**:
- In the form of:
\[
\begin{cases}
a_i^T X \leq b_i & (i \in I_{\leq}) \\
a_i^T X = b_i & (i \in I_{=})
\end{cases}
\]
- Where \( a_i^T \) represents the \( i \)-th row of matrix \( A \in \mathbb{R}^{m \times n} \), \( b_i \) represents the \( i \)-th row of matrix \( B \in \mathbb{R}^{m \times p} \), \( X \in \mathbb{R}^{n \times p} \), and the disjoint index sets \( I_{\leq} \) and \( I_{=} \) partition the set \( \{1, 2, \ldots, m\} \).
2. **Tensor linear feasibility problems**:
- In the form of:
\[
\begin{cases}
A_i:: * X \leq B_i:: & (i \in I_{\leq}) \\
A_i:: * X = B_i:: & (i \in I_{=})
\end{cases}
\]
- Where \( A \in \mathbb{R}^{m \times l \times n} \), \( X \in \mathbb{R}^{l \times p \times n} \), \( B \in \mathbb{R}^{m \times p \times n} \), and \( A_i:: \) and \( B_i:: \) respectively represent the \( i \)-th row slices of \( A \) and \( B \), and the disjoint index sets \( I_{\leq} \) and \( I_{=} \) partition the set \( I = \{1, 2, \ldots, m\} \).
3. **Tensor linear equality problems with boundary constraints**:
- In the form of:
\[
A * X = B, \quad X \leq \tilde{B}
\]
- Where \( A \in \mathbb{R}^{m \times l \times n} \), \( X \in \mathbb{R}^{l \times p \times n} \), \( B \in \mathbb{R}^{m \times p \times n} \), \( \tilde{B} \in \mathbb{R}^{l \times p \times n} \).
### Main contributions
1. **Block - matrix randomized Kaczmarz method (B - MRK)**:
- A new block - matrix randomized Kaczmarz method is proposed to solve matrix - form linear feasibility problems. This method does not need to calculate the pseudo - inverse of each block matrix, thus being more computationally efficient and numerically more stable.
- It is proved that the B - MRK method converges to the feasible region under linear expectation without relying on special row partitions.
2. **Tensor randomized Kaczmarz method (TRK - L)**:
- The B - MRK method is extended to solve tensor - form linear feasibility problems.