Allowing functions with more descriptive power as an interval bounds is the obvious way to generalize interval arithmetic.
For any particular function there is an interval arithmetic number system with an abstract model . I will assume that the function has k ``coefficients'', and n ``parameters''. Each interval bound would then be specified by k extended real numbers, and would vary over an n-dimensional domain:
An interval model of an m-ary function g is valid if has the inclusion property. The model has the inclusion property if
The interval extension of g is also defined as before:
Different choices of f lead to differing complexity in the implementation of the operator models (such as , , and ) and the demotion operators and . The choice of f will affect how well the intervals can track the underlying stream of real computations as well as how useful the computed results will be.
Jeff Tupper | March 1996 |