Implementing a specific solver interface

The Mosel Native Interface (NI) defines a set of conventions that make it possible to extend the Mosel language with constants, subroutines, types, I/O drivers, and control parameters. The Mosel NI also publishes a special set of functionality that provides access to the matrix-based representation of optimization problems formulated using the Mosel types mpvar, linctr, mpproblem, that is, LP and MIP problems. These NI functions can be used for implementing interfaces to optimization solvers that are available in the form of a C/C++ library.

A minimal implementation of a solver module needs to do the following:

• Modeling functionality:
  • define subroutines to start an optimization run and retrieve solution values
  • provide access to solver parameters
  • if supported by the solver, provide support for handling multiple problems
• NI functionality:
  • implement a reset and an unload service
  • initialize the module and the required interface structures

Example

The following Mosel model highlights the modelling features that are provided through the example solver module 'myxprs'.

model "Problem solving and solution retrieval"
  uses "myxprs"                       ! Load the solver module

  declarations
    x,y: mpvar                        ! Some decision variables
    pb: mpproblem                     ! (Sub)problem
  end-declarations

  procedure printsol
    if getprobstat = MYXP_OPT then
      writeln("Solution: ", getobjval, ";", x.sol, ";", y.sol)
    else
      writeln("No solution found")
    end-if
  end-procedure

  Ctr1:= 3*x + 2*y <= 400
  Ctr2:= x + 3*y <= 310
  MyObj:= 5*x + 20*y

 ! Setting solver parameters
  setparam("myxp_verbose", true)      ! Display solver log
  setparam("myxp_maxtime", 10)        ! Set a time limit

 ! Solve the problem (includes matrix generation)
  maximize(MyObj)

 ! Retrieve a solver parameter
  writeln("Solver status: ", getparam("myxp_lpstatus"))
 ! Access solution information
  printsol

 ! Turn poblem into a MIP
  x is_integer; y is_integer

 ! Solve the modified problem
  maximize(MyObj)
  printsol

 ! **** Define and solve a (sub)problem ****
  with pb do
    3*x + 2*y <= 350
    x + 3*y <= 250
    maximize(5*x + 20*y)
    printsol
  end-do

end-model

The underlying module implementation (C program) is discussed in detail in Chapter Implementing an LP/MIP solver interface of the Mosel NI User Guide, including some extensions such as the implementation of a solution callback and a matrix export subroutine. The complete source can be retrieved from http://examples.xpress.fico.com. The C program needs to be compiled into the library myxprs.dso using the makefile that is provided along with the Mosel module examples. The resulting DSO file needs to be on the search path for DSO (e.g. by including its location in the environment variable MOSEL_DSO) and the solver libraries have to be accessible and suitably licensed in order to run the example model.