Influence of initial conditions in SteadyState
 

If i am right,the initial condition should not have any influence if our solution is converged.
In my Conjugate Heat Transfer simulation,After about 10 iteration i get the residual about 10^-5 and this residual isn't changed anymore... .
Therefor i expect that the results aren't changed by continuing iteration.But by each iteration(in spite of no changes in residual) my results are changed.
Also the initial condition has huge influence on my results(in spite of having the same residual about 10^-5)

Do you know why?
Thank you in advance.

You are correct that in a valid steady state simulation that you should end up with the same solution regardless of initial conditions. Your problem is your interpretation of what is being presented to you as "Residuals". In commercial CFD software, residuals are of absolutely no use in determining convergence. They may help in determining reasons for divergence, but that is about it.

The residual values provided by Star-CCM+ are normalized to 1 in the first iteration*. So, if you start your simulation with initial conditions wildly different than the steady state solution, the residuals will rapidly fall and appear to level off long before the simulation has actually converged. This is likely the behavior you are seeing. At the other end of the spectrum, if you use initial conditions that are very near the steady state solution, the residuals will never fall very far, and you may get a valid converged solution with residuals above 1e-1. This is why it is a very bad idea to have a convergence rule based on residual values.

To properly gauge convergence, you need to monitor other values. For a CHT simulation such as yours, I would monitor the heat transfer rates at each of your solid-fluid and solid-solid interfaces, the temperature of the flow at the outlet (assuming you have an outlet), and possibly the temperatures of a few individual solid cells that you expect to converge last. Once all of those values have stopped changing, or at least begin osculating about a constant value, then your simulation is converged.

* This is a terrible design. It is only done so that the residual plots look pretty for the manager that CD-adapco is selling the software to. For all the fantastic meshing and solving algorithms in Star-CCM+, their design decisions do a fantastic job of obstructing natural workflow, inhibiting flexibility, and somehow simultaneously confusing users and treating them as if they are idiots.

kyle,
the residuals are normalised to the highest residual within the first 5 iterations. And I assume this is not done to better sell the software but for better judgement about a simulation. When a simulation converges very slowly, residuals will show a completely wrong picture without normalisation.