additional errors happen in the last step.  The trick is to avoid
division by 0.1**n -- multiply by 10.0**n instead.