Abstract:The singular value decomposition (SVD) of a matrix is a powerful tool for many matrix computation problems. In this paper, we consider generalizing the standard SVD to analyze and compute the regularized solution of linear ill-posed problems that arise from discretizing the first kind Fredholm integral equations. For the commonly used quadrature method for discretization, a regularizer of the form $\|x\|_{M}^2:=x^TMx$ should be exploited, where $M$ is symmetric positive definite. To handle this regularizer, we give the weighted SVD (WSVD) of a matrix under the $M$-inner product. Several important applications of WSVD, such as low-rank approximation and solving the least squares problems with minimum $\|\cdot\|_M$-norm, are studied. We propose the weighted Golub-Kahan bidiagonalization (WGKB) to compute several dominant WSVD components and a corresponding weighted LSQR algorithm to iteratively solve the least squares problem. All the above tools and methods are used to analyze and solve linear ill-posed problems with the regularizer $\|x\|_{M}^2$. A WGKB-based subspace projection regularization method is proposed to efficiently compute a good regularized solution, which can incorporate the prior information about $x$ encoded by the regularizer $\|x\|_{M}^2$. Several numerical experiments are performed to illustrate the fruitfulness of our methods.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the matrix singular value decomposition (SVD) under non - standard inner products and its application in solving linear ill - posed problems. Specifically, the paper focuses on how to generalize the standard SVD to analyze and calculate the regularized solutions of linear ill - posed problems arising from the discretization of Fredholm integral equations of the first kind. When discretizing by commonly used quadrature methods, a regularization term of the form \(\|x\|^2_M := x^T M x\) needs to be utilized, where \(M\) is a symmetric positive - definite matrix. To handle this regularization term, the paper proposes the weighted SVD (WSVD), that is, the singular value decomposition of a matrix under the \(M\)-inner product.
### Main contributions of the paper:
1. **Proposal of weighted SVD**: For any symmetric positive - definite matrix \(M\in\mathbb{R}^{n\times n}\), the paper generalizes a new form of SVD such that the right singular vectors form an \(M\)-orthogonal basis of \((\mathbb{R}^n, \langle\cdot,\cdot\rangle_M)\). This new generalized SVD is called weighted SVD (WSVD) and has many good properties similar to the standard SVD.
2. **Basic properties and applications of WSVD**: Some basic properties of WSVD are studied, including its relationship with the null space and column space of the matrix, the low - rank approximation form based on WSVD, and the WSVD representation form of the solution to the least - squares problem with the minimum \(\|x\|_M\) norm.
3. **Weighted Golub - Kahan bidiagonalization (WGKB)**: A weighted Golub - Kahan bidiagonalization process is proposed, which can be used to calculate several dominant WSVD components. Based on the WGKB process, an iterative algorithm is also proposed to solve the least - squares problem. This algorithm corresponds to the standard LSQR algorithm, so it is called weighted LSQR (WLSQR).
4. **Analysis of Tikhonov regularization problems**: The solution of the Tikhonov regularization problem with the regularization term \(\|x\|^2_M\) is analyzed using WSVD. To utilize the information of the regularization term and avoid pre - selecting the regularization parameter \(\lambda\), the paper proposes the subspace projection regularization (SPR) method. It is proved that the SPR method based on WGKB is the WLSQR with an early - stopping rule and can efficiently calculate a satisfactory regularized solution.
### Numerical experiment part:
The paper uses four test examples of Fredholm integral equations of the first kind to verify the proposed algorithms. By using Simpson's rule for integral discretization, the results show that the proposed algorithms perform better than the standard LSQR solver in regularizing the original problems.
### Paper structure:
- **Section 1**: Introduction, introducing the basic concepts and background of SVD.
- **Section 2**: Reviewing the basic properties and applications of SVD.
- **Section 3**: Proposing weighted SVD and studying its properties.
- **Section 4**: Proposing the WGKB process and WLSQR algorithm for solving the least - squares problem.
- **Section 5**: Using WSVD to analyze and develop iterative regularization methods for linear ill - posed problems.
- **Section 6**: Verifying the effectiveness of the new methods through numerical experiments.
- **Section 7**: Conclusion.
In conclusion, by generalizing SVD to weighted SVD under non - standard inner products, this paper provides an effective method to handle linear ill - posed problems with specific regularization terms.