Matlab's 'double' does not round off after four decimals as far as the actual computations are concerned. It is always accurate to about 16 significant decimal digits. It's only the display of its results that is given to four digits if you are using format short. It is important to understand the difference.
Try this experiment.
format short
x = pi^2
fprintf('x = %19.16f\n',x)
The second display is accurate to 16 places which means that this was the accuracy of the original computation. The 'short' format only displayed four of the significant digits.
