The exact part struck me the most. I was expecting some technique that tries to figure out the maximal error during the calculations and provides enough space f

The exact part struck me the most. I was expecting some technique that tries to figure out the maximal error during the calculations and provides enough space for an exact floating point presentation in advance to avoid the usual problem of continually increasing memory demands that you usually get with exact numbers. Such techniques can be used while calculating delaunay triangulations and maybe they are applicable here as well.