Petr Tichy - Computations and rounding errors
Date: 19/03/2014 12:20
Place: Seminární místnost KNM
Nowadays, computations are performed on computers that use finite precision arithmetic. Each number is represented by a given number of bits. To express the results of an operation, a computer has to approximate the result again by the given number of bits. The error associated with this approximation is called the rounding (or roundoff) error. In the talk we will show examples of several popular algorithms of numerical linear algebra, where the rounding errors can play a crucial role. It is not always necessary to try to suppress the influence of rounding errors (e.g., by reorthogonalization). However, it is always important to understand, how the rounding errors can influence the behavior of the algorithm (delay of convergence, ultimate level of accuracy), and to what extent the computed result is reliable.