The LFOPC leap-frog algorithm for constrained optimization

This paper describes an accurate and reliable new algorithm (LFOPC) for sol ving constrained optimization problems, through a three-phase application o f the well-established leap-frog method for unconstrained optimization, to penalty function formulations of the original constrained problems. The alg orithm represents a considerable improvement over an earlier version (LFOPC ON) which requires the judicious choice of parameter settings for efficient use. The current algorithm automatically executes normalization and scalin g operations on the gradients of the constraints. This results in a robust algorithm that, apart from convergence tolerances, requires virtually no pa rameter settings. The method has been well tested, on both standard analyti cal test problems and practical engineering design problems. (C) 2000 Elsev ier Science Ltd. All rights reserved.