Convergence Analysis and Parameter Selection for the AGSOR-like Iteration for Absolute Value Equations
The accelerated generalized SOR-like (AGSOR-like) iteration method for solving the absolute value equation is re-investigated. As pointed out by Guo [12], a convergence argument based only on a \(k\)-dependent iteration matrix proposed in [38] is not sufficient to justify the global convergence of the AGSOR-like iteration, and the statements about the optimal parameters proposed in [38, Theorem 3.3] are not valid. Rather than relying on a state-dependent iteration matrix, we analyze the method through successive differences of the iterates and establish a correct global convergence result by the Cauchy convergence criterion in a general \(p\)-norm. This approach leads to a new sufficient convergence domain for the iteration parameters. We further investigate norm-based parameter selection and derive explicit optimal parameters under the \(\infty\)-, 1-, and 2-norms. Numerical experiments show that the new convergence domain aligns more closely with the empirical convergence region, illustrate the practical effectiveness of the norm-based parameter choices, and demonstrate the competitive performance of the AGSOR-like iteration. These results provide a mathematically justified and computationally effective framework for using the AGSOR-like method for absolute value equations.
Authors
- Xujie Guo
- Pengru Li
Institutions
- Beihang University (CN)
Publication Details
- Journal
- Numerical Mathematics Theory Methods and Applications
- Published
- 2026-11-16
- DOI
- https://doi.org/10.4208/nmtma.oa-2026-0051
- Primary Topic
- Iterative Methods for Nonlinear Equations
- Type
- article
- Field-Weighted Citation Impact
- 0.00