Hello,
The problem in numerics is the neutralization of the errors for example plus (+)error and minus (-)error. For that reason it is difficult to locate a convergent process. Some numerical algorithms avoid these omissions like leat squares methods which handle the square of the errors in the minimization function. In the other applications like solving processes truncating errors and roundoff errors can be analyzed successfully.