Abstract:A new variant of the GMRES method is presented for solving linear systems with the same matrix and subsequently obtained multiple right-hand sides. The new method keeps such properties of the classical GMRES algorithm as follows. Both bases of the search space and its image are maintained orthonormal that increases the robustness of the method. Moreover there is no need to store both bases since they are effectively represented within a common basis. Along with it our method is theoretically equivalent to the GCR method extended for a case of multiple right-hand sides but is more numerically robust and requires less memory.
The main result of the paper is a mechanism of adding an arbitrary direction vector to the search space that can be easily adopted for flexible GMRES or GMRES with deflated restarting.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: **How to effectively extend the Generalized Minimal Residual Method (GMRES method) to handle linear systems with the same matrix but multiple right - hand - side vectors without restarting the method**.
Specifically, the paper aims to develop a new GMRES variant that can:
1. **Add new right - hand - side vectors without restarting the method**: When solving linear systems with multiple right - hand - side vectors, the existing GMRES method usually needs to be restarted, which leads to a waste of computational resources. The new method can continue to process new right - hand - side vectors without losing previous information.
2. **Maintain numerical stability and reduce memory requirements**: The new method not only maintains the numerical stability of the classical GMRES algorithm, but also reduces memory usage by effectively representing the common basis.
3. **Provide flexible support for direction vectors**: The new method can easily add arbitrary direction vectors to the search space, thus being applicable to the flexible GMRES method or the GMRES method with deflation - restart techniques.
### Background and Motivation
In many practical applications, such as wave scattering problems, time - marching partial differential equations (PDEs) and structural mechanics problems, it is often necessary to solve linear systems with the same matrix but different right - hand - side vectors. These right - hand - side vectors may not be available simultaneously, but are generated step by step depending on the solution of the previous system. Therefore, direct solvers (such as LU decomposition), although efficient, are not ideal when dealing with multiple right - hand - side vectors because they cannot utilize the information obtained during previous solution processes.
### Main Contributions
The main contribution of the paper is to propose a new GMRES variant that:
- **Is theoretically equivalent to the extended GCR method**: The new method is mathematically equivalent to the extended GCR method, but is more numerically stable and occupies less memory.
- **Maintains the orthogonality of the search space and its image space**: The new method ensures that the bases of the search space and its image space always remain orthogonal, thereby improving the robustness of the method.
- **Introduces an effective representation of the common basis**: The new method reduces the storage requirements by constructing a common basis to represent the search space and its image space.
### Summary
In summary, this paper solves the problems existing in the existing methods when dealing with multiple right - hand - side vectors by proposing a new GMRES variant, provides a more efficient solution, and lays the foundation for future related research.