higher bound for a
float point error function is
one half ___