Manipulating the Matrix

In general, the basic usage of the FICO Xpress Optimizer described in the previous chapters will be sufficient for most users' requirements. Using the Optimizer in this way simply means load the problem, solve the problem, get the results and finish.

In some cases, however, it is required that the problem is first solved, then modified, and solved again. We may want to do this, for example, if a problem was found to be infeasible. In this case, to find a feasible subset of constraints we iteratively remove some constraints and re–solve the problem. Another example is when a user wants to 'generate' columns using the optimal duals of a 'restricted' LP problem. In this case we will first need to load a problem and then we will need to add columns to this problem after it has been solved.

For library users, FICO Xpress provides a suite of functions providing read and modify access to the matrix.

Reading the Matrix

The Optimizer provides a suite of routines for read access to the optimization problem including access to the objective coefficients, constraint right hand sides, decision variable bounds and the matrix coefficients.

It is important to note that the information returned by these functions will depend on whether or not the problem has been run through an optimization algorithm or if the problem is currently being solved using an optimization algorithm, in which case the user will be calling the access routines from a callback (see section Using the Callbacks for details about callbacks). Note that the dependency on when the access routine is called is mainly due to the way "presolve" methods are applied to modify the problem. How the presolve methods affect what the user accesses through the read routines is discussed in section Working with Presolve.

The user can access the names of the problem's constraints, or 'rows', as well as the decision variables, or 'columns', using the XPRSgetnames routine.

The linear coefficients of the problem constraints can be read using XPRSgetrows. Note that for the cases where the user requires access to the linear matrix coefficients in the column–wise sense the Optimizer includes the XPRSgetcols function. The type of the constraint, the right hand side and the right hand side range are accessed using the functions XPRSgetrowtype, XPRSgetrhs and XPRSgetrhsrange, respectively.

The coefficients of the objective function can be accessed using the XPRSgetobj routine, for the linear coefficients, and XPRSgetqobj for the quadratic objective function coefficients. The type of a column (or decision variable) and its upper and lower bounds can be accessed using the routines XPRSgetcoltype, XPRSgetub and XPRSgetlb, respectively.

The quadratic coefficients in constraints can be accessed either in matrix form, using the XPRSgetqrowqmatrix routine, or as a list of quadratic coefficients with the XPRSgetqrowqmatrixtriplets.

Note that the reference section in Chapter Console and Library Functions of this manual provides details on the usage of these functions.

Modifying the Matrix

The Optimizer provides a set of routines for manipulating the problem data. These include a set of routines for adding and deleting problem constraints ('rows') and decision variables ('columns'). A set of routines is also provided for changing individual coefficients of the problem and for changing the types of decision variables in the problem.

Rows and columns can be added to a problem together with their linear coefficients using XPRSaddrows and XPRSaddcols, respectively. Rows and columns can be deleted using XPRSdelrows and XPRSdelcols, respectively.

The Optimizer provides a suite of routines for modifying the data for existing rows and columns. The linear matrix coefficients can be modified using XPRSchgcoef (or use XPRSchgmcoef if a batch of coefficients are to be changed). Row and column types can be changed using the routines XPRSchgrowtype and XPRSchgcoltype, respectively. Right hand sides and their ranges may be changed with XPRSchgrhs and XPRSchgrhsrange. The linear objective function coefficients may be changed with XPRSchgobj while the quadratic objective function coefficients are changed using XPRSchgqobj (or use XPRSchgmqobj if a batch of coefficients are to be changed). Likewise, quadratic coefficients in constraints are changed with XPRSchgqrowcoeff.

Examples of the usage of all the above functions and their syntax may be found in the reference section of this manual in Chapter Console and Library Functions.

Finally, it is important to note that it is not possible to modify a matrix when it has been 'presolved' and has not been subsequently 'postsolved'. The following section Working with Presolve discusses some important points concerning reading and modifying a problem that is "presolved".