1998 Nonlinear Programming Software Survey
Page 6
| Product | Publisher | Restrictions |
Object or Source Code | Derivative Calculation Requirements |
| AIMMS | Paragon Decision Technology B.V. | Differentiable constraints | No | None; Analytic derivatives are automatically computed by AIMMS. |
| CONOPT | ARKI Consulting & Development A/S | The model must be smooth and differentiable. It is assumed to be large and sparse. | Object code (subroutine library);Executable modules w/ modeling languages. | User must provide derivatives; they must be returned in sparse format and must be accurate. |
| CONOPT for AMPL | Compass Modeling Solutions | Differentiable and continuous (preferred) | Object | None - automatic differentiation |
| DFNLP | K. Schittkowski | Differentiable model functions | Fortran source code | Analytical or numerical |
| DOC/DOT | Vanderplants R&D Inc. | Continuous with continuous first derivatives | Object | user may provide; otherwise, finite differentiation is used |
| FANPAC/NLP | Aptech Systems Inc. | Twice differential objective function, P.D. Hession | Source | Numerical or user-supplied analytical |
| GRG2 | Optimal Methods Inc. | None, but best with a differentiable function | Source code | None |
| GRG2 for AMPL and AMPL Plus | Compass Modeling Solutions | Differentiable and continuous (preferred) | Object | None - automatic differentiation |
| IMSL Libraries | Visual Numetrics | None | Both | Either user-supplied or finite difference approximations are applied. |
| INTPT | Optimal Methods Inc. | None, but best with differentiable functions | Source | None |
| LANCELOT | P. Toint | Differentiability | Source | First derivatives (second if possible) |
| LGO, for Continuous Global Optimization | Pinter Consulting Services | Only continuity is assumed; applicable even to stand-alone, black box models | Object | None |
| LINGO | LINDO Systems Inc. | All standard math. functions and probability/queuing functions supported. Convexity & differentiability help, but not required. | PC versions include DLL & OLE interfaces. | None; Derivatives are calculated automatically; user can override defaults. |
| LSGRG for AMPL and AMPL Plus | Compass Modeling Solutions | Differentiable and continuous (preferred) | Object | None - automatic differentiation |
| LSGRG2 | Optimal Methods Inc. | None, but best with a differentiable function | Source | None |
| LSSOL | Stanford Business Software | Positive definite or semi-definite QP (including LP) linear constraints | Source | — |
| Mathcad | MathSoft Inc. | Differentiable functions | Object | None |
| Microsoft Excel 97 - Solver | Microsoft Corporation | None, but convergence results depend on differentiability | Object | None |
| MINOS for AMPL | Compass Modeling Solutions | Differentiable and continuous (preferred) | Object | None - automatic differentiation |
| MINOS 5.5 | Stanford Business Software | Nonlinear objectives and constraint functions must be smooth, local optimum obtained for nonconvex problems. | Source, Mex files for MATLAB | Automatic or user supply |
| NAG C Library | Numerical Algorithms Group | Will use first derivatives if provided, but will estimate otherwise | Both | May provide first derivative, but not required |
| NAG Fortran Library | Numerical Algorithms Group | Will use first derivatives if provided, but will estimate otherwise | Both | May provide first derivative, but not required |
| NLPQL | K. Schittkowski | Differentiable model functions | Fortran source code | Analytical or numerical |
| NPSOL 5.0 | Stanford Business Software | Non-linear objective and constraints functions must be smooth. Local optimum obtaines for non-convex problems. | Source, Mex files for MATLAB | Automatic or user supply |
| Optimal Engineer� | Transpower Corporation | None | Only under very special conditions | None - program does it |
| Premium Solver Platform for Excel | Frontline Systems Inc. | None, but convergence results depend on differentiability | Object | None |
| Premium Solver, Premium Solver Plus for Excel | Frontline Systems Inc. | None, but convergence results depend on differentiability | Object | None |
| SAS Software | SAS Institute Inc. | Continuous objective with continuous 1st-order deriv. (except N-M simplex) Some techniques require continuous 2nd-order deriv. | n | Can compute deriv. via analysis or finite differentiation approx., or user can supply exact or approx. numerical functions. |
| SCIENTIST for Windows | MicroMath Research | n/a | n/a | n/a |
| SLP/GRG | Optimal Methods Inc. | None, but best with differentiable functions | Source code | None |
| SOCS and NLPSPR | Boeing Co. | Differentiable (C squared) | Object code | Analytic or finite difference derivatives |
| Solver DLL V3.0, Solver DLL Plus | Frontline Systems Inc. | Problem functions should be differentiable. | Object code (Dynamic Link Library) | User may optionally write jacobian subroutine to compute derivatives |
| Solver for Lotus 1-2-3 97/98 | Frontline Systems Inc. | None, but convergence results depend on differentiability | Object code | None |
| SOPT-CP | SAITECH Inc. | For convex problems, SOPT finds a global optimum. Otherwise, local optimum. | Yes | Without AMPL, need to set up Hession and/or Jacobian. |
| SQP | Optimal Methods Inc. | None, but best with differentiable functions | Source code | None |
| What's Best! | LINDO Systems Inc. | All standard math. functions & probability/queuing functions supported. Convexity & differentiability help, but not required. | — | None; Derivatives are calculated automatically; user can override defaults. |
| XPRESS Barrier QP Solver | Dash Associates Ltd. | Convex quadratic objective and constraints allowed | Object code only | None |
| X Solver 2.0 | Exatech Corporation | None | No | No derivative calculations are used. |
Nonlinear Programming Software Survey Pages:
Introduction | Page 1 | Page 2 | Page 3 | Page 4 | Page 5 | Page 6 | Page 7 | Page 8 | Accompanying Article