Open Access

ARTICLE

A Revision of Relaxed Steepest Descent Method from the Dynamics on an Invariant Manifold

Chein-Shan Liu¹

1 Department of Civil Engineering, National Taiwan University, Taipei, Taiwan. E-mail: liucs@ntu.edu.tw

Computer Modeling in Engineering & Sciences 2011, 80(1), 57-86. https://doi.org/10.3970/cmes.2011.080.057

Abstract

Based-on the ordinary differential equations defined on an invariant manifold, we propose a theoretical procedure to derive a Relaxed Steepest Descent Method (RSDM) for numerically solving an ill-posed system of linear equations when the data are polluted by random noise. The invariant manifold is defined in terms of a squared-residual-norm and a fictitious time-like variable, and in the final stage we can derive an iterative algorithm including a parameter, which is known as the relaxation parameter. Through a Hopf bifurcation, this parameter indeed plays a major role to switch the situation of slow convergence to a new situation with faster convergence. Several numerical examples, including the first-kind Fredholm integral equation and backward heat conduction problem, are examined and compared with exact solutions, revealing that the RSDM has superior computational efficiency and accuracy even for the highly ill-conditioned linear equations with a large noise imposed on the given data.

---

Keywords

---