Is the error due to a limitation in float numbers precission?; some kind of rounding?.

If this is the case, and if your compiler supports this option, maybe you should use double floats, at least for the intermediate computations.

Regards,

Alfredo.