Separation process optimization calculations (Q1099788)

The paper reports about the application of a modified successive quadratic programming (SQP) algorithm to separation process optimization problems. The modification of the usual SQP methods of Wilson, Han and Powell contains a new way for the determination of the quasi-Newton approximations to the Hessian matrix of the Lagrangian function. In the described method these approximating matrices are split into two symmetric parts in the form \(B=C+A\). In this representation C contains readily available second derivatives of the Lagrangian, and A contains the second derivative information that is difficult or expensive to obtain. The quasi-Newton update is performed only with respect to the matrix A using the Powell-symmetric Broyden formula. At this update thermodynamic constraints are used as side conditions, and therefore, in general, the matrix A will not satisfy the secant condition for quasi- Newton methods. The numerical results show that, independently from this drawback, this hybrid method gives better results for the considered separation process problems than the original methods of Han and Powell.