A Convergence Acceleration Method based on a good Estimation of the Absolute value of the Error