Here we solve
where and are real. Well known techniques (larger magnitude root
computed by quadratic formula, smaller magnitude root computed by applying Vieta's rule:
) are used to ensure numerical accuracy of the roots. Additionally the
code is equipped to deal with extremely large coefficients, such that computational
overflow will not occur. A coefficient rescaling technique is applied if during evaluation
of the quadratic formula either the term or the discriminant
become larger than the largest positive number on the machine.