To calculate a relative error my book says to divide the error bound by the rounded mid-value. Later a question has answers where the relative error must have been calculated from dividing the accurate mid-value by the maximum absolute error. There is often a difference of 0.05% or so between these methods. I don't know which method to use.