** Next:** Sparse linear solvers
** Up:** Advanced options
** Previous:** Equations decomposition
** Contents**

###

Convergence Criteria

There are three different methods of declaring a convergence:

*absolute tolerance*, relative tolerance and running for a specified number of iterations. In the absolute tolerance mode algorithm stops when an absolute difference in any of the variables is lower than the threshold.
*relative tolerance* In this method I compute the
difference between the new approximation and the old one for each of the
variables and divide it by the previous(old) value of the variable (in case it
is not equal to 0).
*number of iterations* Here we let the algorithm run for the specified number of steps no matter what
absolute or relative error was obtained before that and return the approximation on the last step as the result of the computation.

Dominik Wojtczak
2006-10-31