A Shamanskii-like Levenberg-Marquardt method for nonlinear equations (Q360386)

The author provides a trust-region-based Shamanskii-like Levenberg-Marquardt (LM) method for nonlinear equations \(F(x)= 0\), where \(F: \mathbb{R}^n\to \mathbb{R}^n\) is a continuously differentiable function. At every iteration of the presented algorithm, first the exact LM step and in addition \(m-1\) (where \(m\) is a fixed positive integer) further approximate LM steps are computed. Thus the Jacobian evaluation and the matrix factorization are done after every \(m\) computations. It is shown that the provided algorithm is (under mild conditions) global convergent and converges (under a local error bound condition which is weaker than nonsingularity) with \(Q\)-order \(m+1\). At the end of the paper, the author gives some numerical results and compares the method with the general LM method \((m=1)\) and the modified LM method \((m=2)\), respectively.