Consider the task to give a parametrization of the surface
When we compute with floating points, the computation is quick (the whole example took less than one second with Maple). But the amplification of the accumulated rounding errors made the result meaningless.
It is impossible to compute this example with symbolic representations of real algebraic numbers. The reason is best explained by the following comment of a colleague answering to a posting to the Maple list:
This polynomial is the product of a linear factor and an irreducible degree 6 factor. The degree 6 factor has Galois group , so you're asking maple to construct a degree extension (as a degree 2 extension of a degree 3 extension etc.) This is a huge task, so I'd consider it normal that this takes so long. I can imagine that the answer wouldn't even fit in your computer.
Nils Bruin, email from Mar 19, 1999
Using exact real numbers, the computing time is expected to be bigger than the time for floating point computation, but much smaller than the time for the symbolic computation with algebraic numbers. The result is a parametrization with ``exact real'' coefficients. Since evaluation of rational functions is easy in exact real computation, this parametrization can be used to produce points with arbitrary small distance to the given surface.