Error-tolerant Multisecant Method for Nonlinearly Constrained Optimization

Abstract

We present a derivative-based algorithm for nonlinearly constrained optimization problems that is tolerant of inaccuracies in the data. The algorithm solves a semi-smooth set of nonlinear equations that are equivalent to the first-order optimality conditions, and it is matrix-free in the sense that it does not require the explicit Lagrangian Hessian or Jacobian of the constraints. The solution method is quasi-Newton, but rather than approximating only the Hessian or constraint Jacobian, the Jacobian of the entire nonlinear set of equations is approximated using a multisecant method. We show how preconditioning can be incorporated into the multisecant update in order to improve the performance of the method. For nonconvex problems, we propose a simple modification of the secant conditions to regularize the Hessian. Numerical experiments suggest that the algorithm is a promising alternative to conventional gradient-based algorithms, particularly when errors are present in the data.