PADMOS und PADFIT — Automatisches Differenzieren und nichtlineare Optimierung in MS-DOS

Summary

The MS-DOS program PADMOS solves nonlinear programming problems with up to 40 constraints and at most 15 variables. The menu-oriented user interfaœ with built-in editor enables a convenient input and modification of problems. Gradients (and when needed Hessians) are computed accurately by automatic differentiation. These features make PADMOS a comfortable tool for research applications where the exact model is selected after several changes of the problem functions. As well the program is suited for teaching purposes. Tedious introductions can be skipped because the Turbo-Pascal like environment is usually familiar to students.

For introductory courses in nonlinear optimization there is a simplified tutorial (PADTUT) containing most relevant algorithms from steepest descent and BFGS to Newton’s method. PADMOS itself uses trust region approaches including directions of negative curvature. Nonlinear constraints can be tackled by SQP, augmented Lagrange or Robinson-type methods which simplify considerably in case of linear constraints. The important case of data fitting is supported by a suitable adapted syntax and graphic display of the fitted function versus sampled data points.