Reducing Long Run Times

If the model is now convergent but still takes unreasonably long to converge either because the time steps are too small or it requires too many iterations per time step then there is another parameter which may be useful.  Under Analyis/Solution Parameters/Relaxation Parameters switch the  ‘Relaxation Application Flag:’ back to “1, Heat Transfer Mode Group” and change the Advection Mode Multiplier
from 1.0 to a number between 0.5 and 0.7.  This multiplier is applied to the computed Relaxation Parameter for Advection Nodes.  This is  a number between 1.0 and a Maximum of 1.999.  The resultant Relaxation Parameter is then multiplied by this factor to yield a Relaxation Parameter near or below 1.0.  A relaxation parameter less than 1.0 will under-relax rather than over-relax the advection
nodes.  Often under-relaxation is required for advection nodes in “stiff” problems.

In order to assess convergence it is important to allow the model to move sufficiently across the time line to allow startup transients to settle out.  In many cases, even for test runs, patience is required to be able to assess model runtime performance along the time line.  If your initial temperature distribution is unrealistic and arbitrary it may take a significant dip in delta-t, time step, to get the model to catch and begin to proceed down the timeline.  Once it does the time step will usually begin to double significantly thereby increasing performance and decreasing analysis time.

Abrupt step changes in thermal loads can have an effect on the size of the time-step. To alleviate this, a ramped load is recommended, with rounded ‘corners’ to allow the solver/time step to adjust on a continuous line.

Another thing to try is to use the Weakly Non-linear Solver or the Direct Solver.