How do I deal with a divergent solution?

Divergent solutions usually imply that there is no equilibrium state associated with the model.  When analyzing tunnels or slopes, it usually implies collapse.  In order to get the model to converge, you usually have to change the design by adding support.  In some cases, you need to increase the number of iterations to allow the system to converge.  If increasing the number of iterations does not improve convergence, then the model is unstable.  Do not increase the tolerance!  This just creates a situation of false convergence and an invalid solution.  The tolerance should never be increased, but it should be decreased in order to ensure that the model really does converge to an equilibrium state.  The model is stable if you lower the tolerance and still get convergence.