An Improved Watchdog Line Search for Successive Quadratic Programming


Mark Hoza and Mark A. Stadtherr


In process optimization using successive quadratic programming, an important step in determining the overall computational efficiency and reliability is the line search used to determine the correction step size taken after solution of the quadratic programming subproblem. Several line search techniques are tested here, including strategies based on augmented Lagrangians and on the watchdog method. Based on the results of this study an improved watchdog line search strategy is suggested, based on an additional sufficient decrease condition and relaxed line search criterion. Tests indicate that the new strategy is the most efficient and reliable of the methods considered here.

Comput. Chem. Eng., 17, 943- 947 (1993)

Return to Publications Page

Return to Home Page