The file mbe9bfloat.pdf has been revised.
The section on the use of bfallroots is new with
a simple example which clearly illustrates the
need the assign values to <both> fpprec and ratepsilon
which "match" in order to take advantage of the
arbitrary precision behavior.
I am grateful to Michel Talon for alerting me to the
bfallroots behavior.
Perhaps the present behavior should be considered
a bug rather than a feature.
Ted Woollett
http://www.csulb.edu/~woollett