Real numbers and computers
- Norbert Müller
- University of Trier
| Date: |
2015/08/25 (Tue) 15:10 to 16:40 |
| Place: |
JAIST, Collaboration room 6 (I-57g) |
| Group: |
Logic Unit |
Sometimes, people think that computers could compute correctly. Unfortunately, this is not true in general: Overflows, underflows, rounding errors and truncation errors can lead to grossly wrong results, even if the used algorithm is implemented with care. This can often be traced down to use of "double precision numbers" as a replacement for the "real numbers".
In the talk we present how a computer can compute more precisely or even "exact", just using basic concepts of object-oriented programming. Additionally, we present some of the basic ideas that are used to get an efficient implementation of exact real arithmetic.
