# numerical computation

 User Name Remember Me Password
 Register Blogs Members List Search Today's Posts Mark Forums Read

 February 14, 2013, 10:30 numerical computation #1 Member   xxxxx Join Date: Feb 2013 Posts: 34 Rep Power: 5 Please, can someone answer to some dubts? 1- when computing a simulation, should one let it run without modify relaxation factors or is it possible/advised to modify them? It happens that changing a relaxation factor completely changes the convergence. But is it right to do so? 2- how can i understand when the simulation is converged? Sometimes it happens that although the residuals are low, the results have no physical sense. On the other hand, sometimes although the residuals are high macroscopic quantities ,pressure drop for example, do not change much through time steps and results seem to be quite good. Is there any way to be quite sure that simulation has reached convergence?

 February 14, 2013, 11:14 #2 Senior Member     Philipp Join Date: Jun 2011 Location: Germany Posts: 1,297 Rep Power: 19 1) If you change the relaxation factors mostly two things can happen: a) the simulation converges faster b) the simulation diverges. I normally don't touch them, except if my simulation doesn't converge anyway. To your question: I don't see any problem in trying, but most of the time Fluent has pretty robust values as default. 2) Best thing is to find some relevant value that you can monitor during calculation. Let's say you are doing some RANS calculation, maybe a lift / drag coefficient of some (important) part inside your domain could be your monitoring value. What you actually take depends on your problem. Integral values are often a good and easy choice. __________________ The skeleton ran out of shampoo in the shower.

 February 14, 2013, 17:23 #3 Member   xxxxx Join Date: Feb 2013 Posts: 34 Rep Power: 5 Thank you for your answer. In effect what you replied is what i usually do but sometimes i'm not sure that it is right. For example, in some simulation i have to use very low values of relaxation factor to reach convergence, even lower than 0.1 . Is it correct? I would like to ask you another question. Now i'm running a simulation, steady state incompressible flow in a quite complex geometry, with k-epsilon turbulence model and 1st order scheme. The mesh is unstructured, 240000 elements. It reached convergence (all residuals below 10^-5), y+ average value is around 30 (a bit low, i know, i would try to modify the mesh). The point is that the head loss i got is much lower than expected. I've tried to change k-epsilon inlet values and verified that head loss did not change. I tried to change the mesh to verify mesh dependence, but until now all other meshes i tried crashed before convergence. I also tried to use Spalart-Allmaras model to see if something change, but the results i got were more or less the same of k-epsilon and still head loss under predicted. Is there any way to understand if solution is good without having any experimental value to compare? I mean, i know what magnitude head loss should be, but i don't have local experimental values (i.e. velocity profile, pressure, etc..) to compare with CFD. How can i understand if solution is correct?

 February 15, 2013, 04:18 #4 Senior Member     Philipp Join Date: Jun 2011 Location: Germany Posts: 1,297 Rep Power: 19 I don't know any good answer. But someone (like a mentor) told this: sometimes local values are not helpful. When it comes to k-e RANS modeling, global values can be predicted extremely good, but if you have a look at velocity profiles on a local scale you see completely wrong results. That's why I wouldn't see local values as a guarantee for getting good agreement also in the global values. Anyway, in your case even the global values are bad... __________________ The skeleton ran out of shampoo in the shower.

March 3, 2013, 13:15
#5
New Member

Orkun Temel
Join Date: Feb 2012
Posts: 13
Rep Power: 6
Quote:
 Originally Posted by xxxx Please, can someone answer to some dubts? 1- when computing a simulation, should one let it run without modify relaxation factors or is it possible/advised to modify them? It happens that changing a relaxation factor completely changes the convergence. But is it right to do so? 2- how can i understand when the simulation is converged? Sometimes it happens that although the residuals are low, the results have no physical sense. On the other hand, sometimes although the residuals are high macroscopic quantities ,pressure drop for example, do not change much through time steps and results seem to be quite good. Is there any way to be quite sure that simulation has reached convergence?
dear friend

as you probably know relaxation factor is the ratio of the current calculated variable and the previous one, so basicly if you increase the relaxation factor your solution should converge faster (i don't mean the residuals) but generally it is convenient to keep relaxation factor relatively low in order to provide a consistent (non-divergent) solution.

it is certainly normal that when you decrease the relaxation factors residuals would remarkably decrease too, it is directly related with the how residual is computed, referring to the spalding's and patankar's journal paper at 1971 (A CALCULATION PROCEDURE FOR HEAT, MASS AND MOMENTUM TRANSFER IN THREE-DIMENSIONAL PARABOLIC FLOWS) residual for continuity let it be mass_imbalance = finite representation of contuinity equation.

if relaxation factor is too low, as you always proceed the numerical calculation almost with the previous calculated variable. your global mass imbalance will decrease but it does not mean that your finite equations have converged to the pde that you're solving.

what convergence really means is the fact that the dependent variables that you are solving each by using a transport equation(u,T,species etc) or a correction method (pressure), is not changing with respect to proceeding iterations. this is how you can judge that your solution has converged.

for your case, if it is too hard to judge convergence by tracing the dependent variables, you may consider to use the flux of a variable, to illustrate energy flux as (rho*u*cp*T) or standart deviation of a variable on a face.

hope this helps.
good luck

 Thread Tools Display Modes Linear Mode

 Posting Rules You may not post new threads You may not post replies You may not post attachments You may not edit your posts BB code is On Smilies are On [IMG] code is On HTML code is OffTrackbacks are On Pingbacks are On Refbacks are On Forum Rules

 Similar Threads Thread Thread Starter Forum Replies Last Post BalanceChen ANSYS 2 July 7, 2011 10:26 sonsiest Main CFD Forum 0 May 23, 2011 15:37 Sergio Rossi Main CFD Forum 3 August 6, 2008 08:02 ado Main CFD Forum 3 October 12, 2000 08:20 Eleuterio TORO Main CFD Forum 0 December 18, 1998 13:41

All times are GMT -4. The time now is 05:16.

 Contact Us - CFD Online - Top