The relative tolerance being applied to the norm of the whole residuum vector.
The relative error e is computed as
e = \frac{||r||}{||r_0||+\epsilon},
where r and r_0 are the residuum and the residuum from computed in the first iteration, respectively, and \epsilon is the machine epsilon guaranteeing that the denominator is always nonzero.
double
Used in no end-to-end test cases.