When to stop a simulation
 

Hello All;
I'm new to this forum.
I'm using Star-CCM to simulate a tube having a hot spot in the middle of it. water flow inside this tube as a coolant.
I'm asking about the stopping criteria.
Residuals seem to saturate after about 300 iterations, and I can achieve temperatures of hot water and the hot spot on the tube.
When I increase to iterations (with very saturated residuals still) to say 5000, temperature of the hot spot on the tube seem to increase noticeably.

So, how to judge the adequacy of the number of iterations, and which one is correct : the results with few (300) iterations or that with 5000 iterations

Thanks

Hi

In a steady state case like this I do look at the convergence of results I am interested in, in this case mass flow, temperature of outflow, heat transfer coeffcient and heat flux of the hot spot.
As the residuals are normalized on the first five (or ten) iterations the value for judging is limited.

In a transient case the residuals should return to a similar value at the last inner iteration.

Best regards,
Matthias

You can judge more easily on residuals if you turn the normalization of them of so you can see them in SI units.

I find doing a separate report for a maximum and minimum (Temperature) Scalar witin the flow region, then creating a monitor and plot (which can combine the two in one graph) very useful.

Also a summation of all mass flows at boundaries is good, as should be near enough zero when converged. But to note that this type of report does very quickly reach 'convergence' for simple flows.

I think Mathias' idea is better, you should judge the convergence of a simulation, if its not converging according to conventional criterion through the use of data you desire to obtain by the means of the simulation.

You can set up a monitor for the temperature(vs iteration) and then setup an asymptotic limit with the value equal to the acceptable deviation for you.

I definitely would not recommend turning off the normalization becuase more often than not it leads humungous changes in the magnitude of the residuals and you are lost as to what should be used as the cutoff for the unnormalized residuals.