On the convegence of a sequential penalty function method for constrained minimization
SIAM Journal on Numerical Analysis
Numerical stability and efficiency of penalty algorithms
SIAM Journal on Numerical Analysis
Primal-dual interior-point methods
Primal-dual interior-point methods
Iterative solution of nonlinear equations in several variables
Iterative solution of nonlinear equations in several variables
On the Superlinear Convergence Order of the Logarithmic Barrier Algorithm
Computational Optimization and Applications
| Hi-index | 7.29 |
Recent efforts in differentiable non-linear programming have been focused on interior point methods, akin to penalty and barrier algorithms. In this paper, we address the classical equality constrained program solved using the simple quadratic loss penalty function/algorithm. The suggestion to use extrapolations to track the differentiable trajectory associated with penalized subproblems goes back to the classic monograph of Fiacco & McCormick. This idea was further developed by Gould who obtained a two-steps quadratically convergent algorithm using prediction steps and Newton correction. Dussault interpreted the prediction step as a combined extrapolation with respect to the penalty parameter and the residual of the first order optimality conditions. Extrapolation with respect to the residual coincides with a Newton step. We explore here higher-order extrapolations, thus higher-order Newton-like methods. We first consider high-order variants of the Newton-Raphson method applied to non-linear systems of equations. Next, we obtain improved asymptotic convergence results for the quadratic loss penalty algorithm by using high-order extrapolation steps.