(35S) - Quadratic root finder -- high performance

[Douglas De Boer]

Jeff,

The bug in my program is that in ax^2 + bx +c = 0, the b must be non-negative in Middlebrook's technique, which I ignored. There is no loss of generality because if b < 0 then just multiply the whole equation through by -1. That does not change the roots. Taking this into account improves the results considerably on my program.

But there are more issues, which I cannot exactly put my finger on at this moment. Middlebook's technique still seems to suffer from an inaccurate calculation of the discriminant in some cases. I'll keep thinking about it. And I think the next thing I'd like to do is try some of the other programs myself, including the "Cadillac" program.

Thanks for all the enlightening help.

Regards,

--DDB