Stable Maintenance of Two-Dimensional Point Set Triangulations

Can geometric algorithms be implemented using floating point arithmetic? A geometric algorithm uses a sequence of primitive tests on continuous data to produce a combinatorial output. Computing a primitive exactly may not be feasible because of the large precision required for the calculation. Computing the primitive approximately, say with floating point arithmetic, may invalidate the correctness of the algorithm, since the primitive may give the wrong answer for some inputs.