higher bound for a
float point error function at
in compress ___