A globally convergent Levenberg-Marquardt method for the least \(l_2\)-norm solution of nonlinear inequalities (Q2379033)

The author studies the least \(l_2\)-norm solution to the problem of solving a system of possibly inconsistent nonlinear inequalities. Since the objective function is usually not twice continuously differentiable, a variety of methods for solving the usual least squares problem are not applicable. Based on the ideas of smoothing Newton methods, the author constructs a smoothing function with twice continuously differentiability to approximate the objective function. A smoothing Levenberg-Marquardt method is used to solve the smoothed problems successively. The global convergence of the algorithm is established.