Solvers that
Work with AMPL

Algorithm types listed in the table are distinguished by the problems they solve and the methods they use, as follows:

• Linear (simplex): Linear objective and constraints, by some version of the simplex method.

• Linear (interior): Linear objective and constraints, by some version of an interior (or barrier) method.

• Network: Linear objective and network flow constraints, by some version of the network simplex method.

• Quadratic: Convex or concave quadratic objective and linear constraints, by either a simplex-type or interior-type method.

• Nonlinear convex: Convex or concave but not all-linear objective, with linear (and possibly certain nonlinear) constraints, by an interior-type method.

• Nonlinear: Continuous but not all-linear objective and constraints, by any of several methods including reduced gradient, quasi-newton, augmented lagrangian and interior-point.

• Complementarity: Linear or nonlinear as above, with additional complementarity conditions.

• Integer linear: Linear objective and constraints and some or all integer-valued variables, by a branch-and-bound approach that applies a linear solver to successive subproblems.

• Integer nonlinear: Continuous but not all-linear objective and constraints and some or all integer-valued variables, by a branch-and-bound approach that applies a nonlinear solver to successive subproblems.

If your favorite solver does not appear here, we encourage you to explore the possibility of adding it to the list. See our discussion of hooking your solver to AMPL following the solver listing.

Solver listing

Algorithm Types. See above for definitions.

Vendor or Download Site. Sources of further information for obtaining the solver. This entry may be a link to a developer's or vendor's web site, or (where indicated) to a site from which the solver can be downloaded.

Driver Code. A link to a directory in netlib/ampl/solvers from which you can download C source code to make an AMPL driver for the solver. The driver provides an interface from AMPL's general description of an optimization problem to a solver's particular algorithms and options. Hence a different driver is needed for each solver. Some solvers are available bundled with the appropriate AMPL driver, while others require that you download and compile the driver yourself; details are available from the vendor or from the README file in the driver's netlib directory.

Documentation. The README files are text copied from the netlib/ampl/solvers directory mentioned above. The Using . . . booklets can be downloaded in PDF or postscript format.

Problem analysis tools

MProbe

Shape means whether the function is linear or almost linear, convex or almost convex, concave or almost concave, or concave and convex. Knowledge of function shape is crucial when developing nonlinear optimization models, or when selecting the nonlinear solver for a nonlinear optimization problem. Determining function shape is difficult for nonlinear functions having more than two variables. MProbe is specifically designed to operate on nonlinear functions having many variables.

MProbe 3.0 offers numerous new plotting options, sampling inside any convex enclosure (for improved accuracy of conclusions), and identification of redundant constraints and bounds.

MProbe has been developed by Prof. John Chinneck of Carleton University. A student/demo version bundled with the AMPL Student Edition is available for downloading through the web.

Hooking your own solver to AMPL

Detailed instructions and examples are available to help you (or your solver's developer) to write an AMPL driver. AMPL can then switch to your solver, set up its algorithmic options, send it a problem to be solved and retrieve the results, all in the same way that you work with currently supported solvers. See our instructions on hooking your solver to AMPL for further details.

Return to the AMPL home page.