A multiplier-free, reduced Hessian method for process optimization

Abstract: "Process optimization problems typically consist of large systems of algebraic equations with relatively few degrees of freedom. For these problems the equation system is generally constructed by linking smaller submodels and solution of these models is frequently effected by calculation procedures that exploit their equation structure. In this paper we describe a tailored optimization strategy based on reduced Hessian Successive Quadratic Programming (SQP). In particular, this approach only requires Newton steps and their 'sensitivities' from structured process submodels and does not require the calculation of Lagrange multipliers for the equality constraints. It can also be extended to large-scale systems through the use of sparse matrix factorizations. The algorithm has the same superlinear and global properties as the reduced Hessian method developed in [4]. Here we summarize these properties and demonstrate the performance of the multiplier-free SQP method through numerical experiments.