GLPK/API usage

From Wikibooks, open books for an open world
Jump to: navigation, search

This page provides examples of API usage for commonly encountered tasks.

Prescribed starts[edit]

A solution can be read from file using the following API routines:

  • glp_read_sol - read basic solution
  • glp_read_ipt - read interior point solution
  • glp_read_mip - read MIP solution

The solution read with glp_read_sol can serve as initial basis for solving a LP using the simplex algorithm (this is what glpsol command line parameter --ini does).

For mixed integer problems an integer feasible (but possibly suboptimal) starting solution can be specified using the GLPK API (a feature not available through GLPSOL).


To do:

  • the following callback instructions need checking and/or refining
  • some better resolved example code could also be useful

The callback hook from the integer solver is used for this purpose. First write a custom callback routine and register this with the MIP solver using the MIP control parameter structure glp_iocp and field cb_func.

Then call glp_ios_reason from your callback routine and if GLP_IHEUR is returned, set and solve your desired starting solution as follows:

  • preset the integer variables in tree->mip
  • default initialize the parameter struct parm by calling: glp_init_smcp(&parm)
  • reset the parameter struct field thus: parm.meth = GLP_DUALP
  • solve the LP: int ret = glp_simplex(tree->mip, &parm)
  • pass all variables by calling function glp_ios_heur_sol

This early-2013 [Help-glpk] thread here and here contains relevant information and the following code snippet (while noting that the use of void* is not good style):

// caution: this code is buggy and will be
// updated as soon as a solution is found
void callback(glp_tree* tree, void* info)
  if (glp_ios_reason(tree) == GLP_IHEUR && glp_ios_curr_node(tree) == 1)
      glp_ios_heur_sol(tree, solution);

where solution is a const double array which contains the primal values of all the variables in locations x[1],...,x[n] where n is the number of columns in the original problem object.

Karush-Kuhn-Tucker optimality conditions[edit]

GLPK provides the lpx_check_ktt routine to check the KKT optimality conditions for a basic solution. The function signature is:

void lpx_check_kkt(glp_prob *lp,    // GLPK problem object pointer 
                   int scaled,      // 0 = use unscaled, non-zero = use internally scaled
                   LPXKKT *kkt)     // LPXKKT C struct pointer (meaning address)

The parameter scaled can be set to 0 to check the original unscaled problem or non-zero to check the internally scaled problem. A typical call might be:

LPXKKT kkt;                      // declaration of a suitable C struct
lpx_check_kkt(lp, 1, &kkt);      // fill call, note the address-of operator

This call fills the struct kkt, which can then be interrogated.