exp() (and related exp functions) and pow() give rise to LOS exceptions. I
can't think of any other way of calculating these functions. The results I get
agree with the results from a SUN 4 and with Abramowitz and Stegun 
(to 15 places). If anyone knows of another algorithn please let me know.

There may still be problems with some functions for pathological arguments.
In some cases I should break the range of the argument up and use
approximations for large and/or small arguments. In fact, the ESIX supplied
exp() is just as bad (good?!?) as mine :-)
