It sometimes happens that you have reason to believe that a large portion of the difference, s_j - s, decrease as j increases at some fixed rate, c, with 1 > c > 0, while the rest of this difference decreases even faster., giving us |s_j -s| c |s_{j -1} -s|.

The method described in the next section allows us to get rid of the portion of  s_j -s that decreases by a factor of c, which improves the convergence rate of the sequence. By employing this method once or twice you can obtain new predictors, t_j or u_j that are much closer to s than s_j is.