Modular arithmetic for linear algebra computations in the real field (Q1264434)

Techniques from computer algebra and numerical computation are combined in order to decrease the precision needed during certain computations involving real numbers. A bound on the final result is used to determine a modulus \(m\), such that all intermediate computations can be performed modulo \(m\), still obtaining the result within the prescribed tolerance. In this way the number of digits of the numbers involved is decreased and thus the complexity improved. The method is applied to sums, scalar products, iterative refinement for a linear system and to the solution of differential equations.