I am solving heat finned tube heat exchanger. At Residual monitor is default setings for Convergence Tolerance. All values is set to 0,001 exep energy to 1e-06.
If velocity of air at input is 2 m/s (x direction), temperature is 288 K and in exchanger velocity is between 0 and 5 m/s is that mean that critical point is where velocity is 0 (residual is 0,001)? What influence on results will be if I set covergence tolerance to 0,01?
In ANSYS guide is stated
"Generally, a decrease in residuals by three orders of magnitude indicates at least qualitative convergence".
On which magnitude? 2 m/s?
y and z direction air velocity is small and situation is quite different!
Pease, can you give comment?
Dear Mladen Bošnjaković,
I would not advise setting the residual tolerance to 0.01 because the solution might still be evolving in the domain.
The statement: "Generally, a decrease in residuals by three orders of magnitude indicates at least qualitative convergence".....means that the difference in the values of variables from one iteration to the next is ~0.001. In other words, a three orders of magnitude reduction of that DIFFERENCE has occurred since the simulation was started. The residual is generally evaluated as a RMS (root-mean-square) over all cells in the domain.
A few years ago, whilst attending a conference I had heard a professor comment during one of the presentations that a residual convergence of 0.001 in FLUENT indicates 10% numerical error. Before hearing that statement, I too used to set 0.001 as the default residual tolerance in my simulations (because converging to 0.001 in FLUENT is, comparatively, easy).
However, the results do improve (in terms of accuracy) if you reduce the tolerance to 1e-05 or 1e-06. Having said that, it is far more difficult to achieve this level of convergence, especially for continuity or turbulence quantities such as k and epsilon. Attaining greater accuracy might necessitate re-designing of the mesh.
As such, there are no set rules as to what constitutes an optimum residual tolerance in FLUENT....I am not aware of dedicated papers that investigate the effect of (increasing or decreasing the) residual tolerance on results of a particular simulation.
The optimum residual tolerance would obviously be problem dependent and also be dependent on the numerical framework used for solution. As an example of the latter, it is generally a case that stricter residual convergence (~1e-06) can be attained in unsteady simulations in FLUENT but not in steady ones (temporal terms not considered). I hope this helps. Best regards.
Dear, Shaswat
Thank to answer. As I can see from my results greatest problemm is epsilon.
It slowly convergate to 10-3. All other parameters easily reach 1e-4 or even 1e-5. Is there any proposals how to speed convergence of epsilon.
Ok,
I will bring my past experience. To get faster convergence one possibility is to make iterations in block (forexample first block 100 iterations, second 70 etc. - which depend of particular problem). For each additional block value URF can be reduced. Often, the biggest impact has URF of momentum (and pressure if it is problem a convergence of continuity).
You may focus on the result you are looking for, set it up to not converge, when the variable you are monitoring stabilize on a value during at least 200 iterations you have the response you are looking for.
Dear All,
I have fluent ansys solver i have a 3D laminar flow model, pressure based and steady state ,but when i do a hybrid initialization i had a warning :convergence tolerance of 1.000000e-06 not reached during hybrid initialization.
I hope if some one here can help me .
Regards,
Noor,
Dear Noor,
What version do you use for Ansys? Are you using student version?
I encountered this problem, try to check the mesh quality In Fluent. Meshing it is a common cause. Also, double check orthogonal quality.
Hope it helps.
Regards,
Claudia
(In nonlinear problems, what is error tolerance for convergence? This should be stated in the paper.) This is a comment by a reviewer, what is it mean.
- Badges
- Science topic
- Similar topics
- Journal Impact Factor