GIPALS32.DLL readme

GIPALS32 Library

GIPALS32 is a linear programming library provided as a Win32 Dynamic Link Library (DLL). The library is a further development of our successful linear programming environment GIPALS. The maximum size of linear program depends of the license type and can be either Unlimited (Large license) or 15,000 (Normal license) variables and constraints.

GIPALS32 provides an easy and powerful way to solve any kind of liner programs within the user's application. The library can be used in different programming tools such as Microsoft Visual C++, Microsoft Visual Basic, Borland Delphi and others.

There are several examples are provided along with the headers files to demonstrate and simplify the usage of the library.

Specifying a Linear Program
Calculation
Getting Results
Solver Settings

Specifying a Linear Program.

There are two possible ways to work with the library:

First way is a linear program construction by the provided set of functions, such as specification of the variables and constraints count and assignment of the constraint matrix elements. This way provides a complete flexibility of the linear programming modeling.
The second way is importing linear program from a text file in MPS format. It can be done by one function LoadFromMPS. This way can be suitable if there is a set of prepared MPS files that need to be calculated.

The typical scenario to construct a linear program is the following:

Specify the type of the calculation - either minimization or maximization of the objective function. SetObjectiveType(1) indicates minimization, SetObjectiveType(0) indicates maximization.
Specify the decision variables and the objective function either by the pair of functions SetVariableCount and SetVariable or by the function AddVariable. The function AddVariable can add only 5000 variables. If the number of variables more than 5000 then the function SetVariableCount must be used to setup the number of variables and then the function SetVariable to setup each variable from 0 to variable count - 1. Please note the variable index is zero-based.
Specify the constraint types and right sides either by the pair of functions SetConstraintCount and SetConstraint or by the function AddConstraint. The function AddConstraint can add only 5000 constraints. If the number of constraints more than 5000 then the function SetConstraintCount must be used to setup the number of constraints and then the function SetConstraint to setup each constraint from 0 to constraint count - 1. Please note the constraint index is zero-based.
Specify the constraint matrix by the function SetConstraintElement (relatively slow way) or by the pair of functions DirectSelectConstraint and DirectAddConstraintElement (fast way).

Note. Functions SetVariableCount, SetVariable, AddVariable, SetConstraintCount, SetConstraint, AddConstraint, SetConstraintElement, DirectSelectConstraint and DirectAddConstraintElement return 0 if the number of variables or constraints exceeded the maximum.

Calculation.

After a linear program (LP) is specified or imported from MPS file it is ready for the calculation. There is the function Calculate that performs the following steps:

Preprocessing. Reduces the LP's size and improves the performance.
Scaling. Improves numerical stability of the calculation.
Dense columns extracting. Detects and removes dense columns from the constraints matrix to preserve sparsity and therefore improve the performance.
Calculation. Performs iterative procedure until the specified tolerances are reached. During this stage the several additional methods are engaged to improve the quality of the result and reduce number of iterations.

The Calculate function returns the status of the calculation: positive results indicate successful run, negative - unsuccessful. Zero indicates the calculation has not been provided yet. The only parameter specified for the function is a pointer on a callback function which can control the calculation by returning an appropriate value (res_Abort, res_None or res_Stop). The parameters for callback function are number of the current iteration, primal and dual objective function values and infeasibilities and optimality.

Getting Results.

After the calculation is done the results are available. Two functions are provided to get the primal and dual objective functions: PrimalObjective and DualObjective.

The function PrimalValue returns the values of the decision (primal) variables X. The function DualValue returns the values of the dual variables (each dual variable associated with one constraint).

Solver Settings..

There are several routines to setup the behavior of the calculation engine. They are provided for advanced users and can be used to account some features of the specified linear program.

The routine SetPreprocessor turns the preprocessing On/Off by the combination of the preprocessor flags. The call SetPreprocessor(prep_Disable) turns preprocessing off.

The routine SetTolerances specifies the primal, dual and optimality tolerances for the calculation.

The routine SetMaxIterationCount setups the limit of the iterations (default is 100).

The routine SetScaling takes the parameter 0 or 1 to disable (0) or enable (1) the LP scaling. By default scaling is enabled.

The routine SetRefinement takes the parameter 0 or 1 to disable (0) or enable (1) the iterative refinement of the solution at each iteration. By default refinement is enabled.

The routine SetGondzio takes the parameter 0 or 1 to disable (0) or enable (1) to use high order Gondzio correction to improve the solution quality and reduce the iterations number.