Our goal is the integration of accurate scientific computation and intelligent object-oriented techniques to obtain powerful tools that can be used for the successful development of scientific expert systems, i.e. systems which provide expertise in the choice of algorithms within a particular problem domain and generate reliable, robust and maximally accurate results of full integrity. The need for such scientific expert systems is increasing significantly [1,4,11,12,15,20,21]. Through this study we want to identify some general principles which must be taken into account both concerning the computational aspects, such as automatic validation of the results (see Section 1), and aspects of programming methodology, especially with respect to classification and inheritance (see Section 2). The usefulness and power of such intelligent object-oriented scientific expert systems will be tested on a prototype which will be developed for the problem domain of multivariate rational data fitting [8]. A more detailed description of this prototype is given in Section 3.
