1998 Nonlinear Programming Software Survey
Page 5
| Product | Publisher | Algorithms | Large Problems |
Linear Constraints and Bounds |
| AIMMS | Paragon Decision Technology B.V. | Conopt | y | no |
| CONOPT | ARKI Consulting & Development A/S | Sparse GRG algorithm with many enhancements | y | Bounds are handled implicitly. Linear approximations to linear constraints are known to be exact. |
| CONOPT for AMPL | Compass Modeling Solutions | GRG | y | — |
| DFNLP | K. Schittkowski | Sequential quadratic programming | — | Passed directly to QP-solver |
| DOC/DOT | Vanderplants R&D Inc. | Modified Feasible Directions, SLP, SQP | y | Bounds handled explicitly |
| FANPAC/NLP | Aptech Systems Inc. | SQP | y | None |
| GRG2 | Optimal Methods Inc. | GRG | n | Bounds handled implicitly |
| GRG2 for AMPL and AMPL Plus | Compass Modeling Solutions | GRG | n | — |
| IMSL Libraries | Visual Numetrics | Successive quadratic programming | y | No |
| INTPT | Optimal Methods Inc. | Primal-Dual Interior Point | y | Bounds handled implicitly |
| LANCELOT | P. Toint | — | y | yes, for bounds; no for linear constraints |
| LGO, for Continuous Global Optimization | Pinter Consulting Services | A (proprietary) combination of global adaptive partition and search, and unconstrained/constrained convex programming algorithms | y | Handled explicitly; Linear constraints can be directly embedded into objective function or handled exactly in local search phase |
| LINGO | LINDO Systems Inc. | GRG and Successive Linear Programming (SLP) are used for nonlinear models. Branch & Bound is used for NL and LP models with integer restrictions. | y | System automatically identifies linear constraints & calculates their derivatives only once. Bounds handled implicitly. |
| LSGRG for AMPL and AMPL Plus | Compass Modeling Solutions | GRG | y | — |
| LSGRG2 | Optimal Methods Inc. | GRG | y | Bounds handled implicitly |
| LSSOL | Stanford Business Software | Active-set method for convex QP and linear least squares with constraints | y | Yes, More efficient if more bounds are active. |
| Mathcad | MathSoft Inc. | GRG2 | n | Linear and partially linear constraints are recognized, and bounds are handled directly by the algorithm. |
| Microsoft Excel 97 - Solver | Microsoft Corporation | GRG2 | n | Bounds are handled directly by the algorithm. |
| MINOS for AMPL | Compass Modeling Solutions | Quasi-Newton, Reduced Gradient and projected Lagrangian | y | — |
| MINOS 5.5 | Stanford Business Software | Primal simplex, reduced gradient protected lagrangiers | y | Yes, similar to primal simplex |
| NAG C Library | Numerical Algorithms Group | Sequential Quadratic Programming | y, large dense | Yes, passed to subroutine in separate structures |
| NAG Fortran Library | Numerical Algorithms Group | Sequential Quadratic Programming | y, large dense | Yes, passed to subroutine in separate structures |
| NLPQL | K. Schittkowski | Sequential Quadratic Programming | n | Passed directly to QP-solver |
| NPSOL 5.0 | Stanford Business Software | SQP method with quasi-Newton approximation of full Hessian | y | Yes, They improve warm starts on the QP subproblems. More efficient if more such constraints are active. |
| Optimal Engineer� | Transpower Corporation | Sequential Quadratic Programming | y | No different than non-linear constraints |
| Premium Solver Platform for Excel | Frontline Systems Inc. | Large-scale (sparse), GRG | y | Linear and partially linear constraints are specially recognized. Bounds are handled directly by the algorithm. |
| Premium Solver, Premium Solver Plus for Excel | Frontline Systems Inc. | GRG2 | n | Bounds are handled directly by the algorithm. |
| SAS Software | SAS Institute Inc. | Quasi-Newton, Newton-Raphson, trust-region, conjugate gradient | y | There are distinct program statements for specifying both boundary and general linear constraints. |
| SCIENTIST for Windows | MicroMath Research | Modified Powell algorithm for least squares | n | n/a |
| SLP/GRG | Optimal Methods Inc. | Successive Linear Programming | y | Bounds handled implicitly |
| SOCS and NLPSPR | Boeing Co. | SQP | y | Simple bounds and general nonlinear constraints permitted |
| Solver DLL V3.0, Solver DLL Plus | Frontline Systems Inc. | GRG2 | n | Bounds are handled directly by the algorithm. |
| Solver for Lotus 1-2-3 97/98 | Frontline Systems Inc. | GRG2 | n | Bounds are handled directly by the algorithm. |
| SOPT-CP | SAITECH Inc. | Primal-dual interior-point algorithms | y | No |
| SQP | Optimal Methods Inc. | Successive Quadratic Programming | y | Bounds handled implicitly |
| What's Best! | LINDO Systems Inc. | GRG and SLP are used for nonlinear models. Branch & Bound is used for NL and LP models with integer restrictions. | y | System automatically identifies linear constraints & calculates their derivatives once. Bounds are handled implicitly |
| XPRESS Barrier QP Solver | Dash Associates Ltd. | Homogeneous interior point | y | Yes, automatically exploited in the linear algebra |
| X Solver 2.0 | Exatech Corporation | Simulated annealing and genetic algorithm | y | Constraints are entered in C language syntax or can be modeled as Excel spreadsheets. |
Nonlinear Programming Software Survey Pages:
Introduction | Page 1 | Page 2 | Page 3 | Page 4 | Page 5 | Page 6 | Page 7 | Page 8 | Accompanying Article