Core Concepts
The proposed extension of Newton's iteration quadratically converges to semiregular nonisolated solutions of smooth nonlinear systems, even when the Jacobian is rank-deficient. The iteration also serves as a regularization mechanism for computing approximate solutions when the system is perturbed.
Abstract
The content discusses an extension of the standard Newton's iteration method that enables quadratic convergence to nonisolated, semiregular solutions of smooth nonlinear systems. Key highlights:
The standard Newton's iteration (1.1) loses its quadratic convergence rate when applied to nonisolated solutions, where the Jacobian is rank-deficient or non-invertible.
The authors introduce the concept of "semiregular" zeros, where the dimension of the zero set matches the nullity of the Jacobian. They prove that the proposed rank-r Newton's iteration (1.2) quadratically converges to semiregular zeros under minimal assumptions (Theorem 4.1).
When the nonlinear system is perturbed or given through empirical data, the nonisolated solution may disappear. The authors show that the rank-r Newton's iteration still converges linearly to a stationary point that approximates an exact solution, with an error bound proportional to the data perturbation (Theorem 5.1).
Geometrically, the iteration approximately converges to the nearest point on the solution manifold. This simplifies nonlinear system modeling by eliminating the need to isolate solutions.
The extension enables a wide range of applications in algebraic computation where nonisolated solutions frequently arise.