Intelligent software systems have often been highly specialised for symbolic computation, but weaker on numeric computation. This is one reason why the combination of symbolic constraint solving and mathematical programming are proving to be so interesting both in theory and practice.
One recalcitrant problem for numeric computation is the problem of numeric instability. Under certain, unlikely, circumstances, tiny rounding errors in the basic mathematical routines can unexpectedly cause serious errors in the final result. The difficulty is that these errors are hard to predict, and no practical way has been found to predict them.
A way to contain this instability is to reason on numeric intervals, instead of numbers, ensuring that at each calculation the interval is rounded out so as to ensure the actual solution lies inside it. Unfortunately these intervals tend to grow too wide to be useful. Recently, however, using intervals as primitive constraints in a constraint programming framework some excellent results have been obtained for well-known mathematical benchmarks. These results compete with the best mathematical programming approaches, in particular when the input intervals are quite wide.
Intervals appear in a multitude of different contexts as a way of approximating values. In particular they are used in database indexing, in constraint propagation, and for specifying uncertainty.
The author predicts that interval constraints will play a crucial role in spatial and temporal databases and in the handling of uncertainty.