Sometimes subsequent
approximations to the iterative matrix solution result in very small changes
in head with each iteration (dashed line in figure below) and we would like
to speed the process up, other times the approximation causes large changes
that repeatedly overshoot the answer in opposite directions (dotted line in
figure below), so we want to slow the process down. Note that if the head changes
are quite small, this may result in premature convergence, which is likely to
produce mass imbalances.
Often the code provides the user the opportunity to
specify an acceleration or relaxation (two terms for the same type of variable)
parameter to speed up or slow down the convergence process.
If progress toward convergence is proceeding too slowly, use a factor larger
than one. This will result in the head changes being increased by this factor
at each iteration.
If progress toward convergence is over shooting the answer (in this case you
will notice head changes of alternate sign with each iteration), use a factor
less than one. This will result in the head changes being decreased by this
factor at each iteration.
Generally the head change you are noting for each time step is not at one node.
Typically the maximum head change at any node is used for testing convergence
and this is likely to vary in location. A good code will print the value used
to test convergence at each iteration and will report the node at which the
maximum occurred.
If your model is not converging efficiently, it can be useful to note where
the maximum head change is occurring. Sometimes evaluation of your input will
reveal that you have a "typo", for example you may have typed the exponent on
a hydraulic conductivity incorrectly resulting in a large contrast of conductivity
between adjacent cells, or you may have added an extra "0" to a head for one
cell causing a huge influx at one cell.