Abstract:The Bunch-Kaufman algorithm and Aasen's algorithm are two of the most widely used methods for solving symmetric indefinite linear systems, yet they both are known to suffer from occasional numerical instability due to potentially exponential element growth or unbounded entries in the matrix factorization. In this work, we develop a randomized complete pivoting (RCP) algorithm for solving symmetric indefinite linear systems. RCP is comparable to the Bunch-Kaufman algorithm and Aasen's algorithm in computational efficiency, yet enjoys theoretical element growth and bounded entries in the factorization comparable to that of complete-pivoting, up to a theoretical failure probability that exponentially decays with an oversampling parameter. Our finite precision analysis shows that RCP is as numerically stable as Gaussian elimination with complete pivoting, and RCP has been observed to be numerically stable in our extensive numerical experiments.
What problem does this paper attempt to address?
This paper attempts to solve the numerical instability problems that may occur during the solution process of symmetric indefinite linear systems. Specifically, although the Bunch - Kaufman algorithm and the Aasen algorithm are widely used to solve such systems, they both have the problem of occasional numerical instability due to excessive element growth or unbounded entries in matrix decomposition.
To overcome these problems, the author proposes a new Randomized Complete Pivoting (RCP) algorithm. The RCP algorithm is comparable in computational efficiency to the existing Bunch - Kaufman algorithm and the Aasen algorithm, but theoretically has better element growth control and bounded matrix decomposition entries. In addition, the numerical stability of the RCP algorithm is close to that of Gaussian elimination with complete pivoting, and it shows good numerical stability in a wide range of numerical experiments.
### Specific Problem Description
The solution of the symmetric indefinite linear system \(Ax = b\) is very important in many application fields such as acoustics, electromagnetics, and structural physics. However, since a symmetric indefinite matrix may contain both positive and negative eigenvalues simultaneously, it is prone to encounter numerical instability problems during the solution process. Although existing methods such as the Bunch - Kaufman algorithm and the Aasen algorithm are effective, in some cases they may lead to exponential growth of elements or unbounded matrix decomposition entries, thus affecting numerical stability.
### Solution
The RCP algorithm proposed in the paper selects pivots by introducing random projection techniques, ensuring that element growth can be effectively controlled and numerical stability can be maintained in most cases. Specifically:
1. **Random Projection**: The RCP algorithm uses random projection techniques to select pivot columns instead of looking for the largest element in the entire matrix. This not only reduces the computational complexity but also ensures that the 2 - norm of the selected pivot column is close to optimal.
2. **Simplified Bunch - Kaufman Pivoting Strategy (SBKP)**: After selecting the pivot column, the RCP algorithm applies a simplified Bunch - Kaufman pivoting strategy to perform specific pivot selection.
3. **Theoretical Analysis**: The paper proves through strict mathematical analysis that the RCP algorithm theoretically has better element growth control and bounded matrix decomposition entries, thus ensuring numerical stability.
### Numerical Experiments
Through extensive numerical experiments, the paper verifies the effectiveness and reliability of the RCP algorithm, indicating that it can provide stable solutions in practical applications.
### Summary
The main contribution of this paper is to propose a new RCP algorithm, which solves the numerical instability problems that may occur in existing methods when solving symmetric indefinite linear systems, and provides a more reliable tool for research in related fields.