The DLPFI Language

Abstract

Once a model has been formulated along lines described in the previous chapter, it must be specified to the D _L P decision support system using the D _L P format interface language, henceforth called DLPFI (pronounced ‘Delphi’). Examples of this low-level language have been presented in Chapters 1 and 2. We now give a more detailed and complete description of DLPFI, which has been designed with the following objectives and usage in mind:

• 1. It provides the means for making a precise statement of a D _L P model to the D _L P decision support system, in much the same way that standard MPS input format provides the means for precisely specifying a linear programming model to a Mathematical Programming System, for example, MINOS (Murtagh and Saunders [1983]). MPS input format is very rigid—character and numeric data must be provided within specific column ranges in each input record. It is, in essence, a set of conventions for specifying the non-zero elements of a large, sparse LP matrix. In constrast, much of the DLPFI input data is format-free, and local and global constraints and objectives can all be specified with considerable flexibility. Each DLPFI data record has a specific meaning in relation to a D _L P model. Thus DLPFI can be viewed as a low-level, simple, very easy-to-learn language for D _L P modeling.

• 2. In conformity with a principle used within most modern modeling systems, DLPFI provides a mechanism for separating D _L P modeling and optimization phases. Standard MPS input format plays an identical role vis-à-vis the linear programming model/algorithm interface.

• 3. DLPFI can serve as a target language for user-supplied front-end routines that are written in a high-level programming or modeling language and can accept data in a more convenient, application-specific way. In standard linear programming, such a set of routines is called a ‘matrix generator’. Here, ‘D _L P model generator’ would be the corresponding analog.

• 4. DLPFI is designed to be extensible, in order to facilitate the implementation of D _L P model/algorithm enhancements discussed in Chapter 10.