CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > CFX

Under-relaxation for steady state simulation in CFX

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Display Modes
Old   March 24, 2011, 10:20
Default Under-relaxation for steady state simulation in CFX
  #1
Senior Member
 
Join Date: Oct 2010
Location: Zurich
Posts: 176
Rep Power: 6
Chander is on a distinguished road
In Solver control, one can change the Timescale factor when Timescale control is set to 'Auto Timescale' option.

For many of my steady state simulations, I observe that the simulation shows lack of convergence (residuals stabilizing above convergence criteria) with the default value of Timescale factor (which is 1).

However, when I reduce this factor, simulation converges. In some cases I had to reduce this factor to 0.1 to get convergence.

Although CFX manual describes that one can specify physical timescale instead of auto timescale to get convergence, simply reducing the timescale factor for Auto timescale option seems more straightforward and better.

Is it okay to do so? Also, does reducing this timescale factor to say 0.1 mean choosing an under-relaxation of 0.1?
Chander is offline   Reply With Quote

Old   March 24, 2011, 16:20
Default
  #2
Senior Member
 
Join Date: Apr 2009
Posts: 511
Rep Power: 12
stumpy is on a distinguished road
If you make the timestep small enough then it's similar to using a very small relaxation factor. Basically you are freezing the solution, so the residuals might converge but it doesn't mean the solution is any good. I'd recommend you figure out why it didn't converge in the first place.
stumpy is offline   Reply With Quote

Old   March 25, 2011, 05:28
Default
  #3
Senior Member
 
Join Date: Oct 2010
Location: Zurich
Posts: 176
Rep Power: 6
Chander is on a distinguished road
@stumpy
Yes you are very right that there there may be some other issues and I am trying to find out those. Especially because the CFX manual says that the timescale chosen by the solver is on the conservative side.

But isn't under-relaxation a valid way to slow down and hence help convergence for non-linear equations? If the solution does converge with under-relaxation, would it not mean that the residuals (difference between partial differential equation and its discretized counterpart) have reduced to below the convergence criteria? I am not able to understand why the solution may be incorrect. It would be of great help if you could explain it in more detail. Thanks !
Chander is offline   Reply With Quote

Old   March 27, 2011, 18:12
Default
  #4
Super Moderator
 
Glenn Horrocks
Join Date: Mar 2009
Location: Sydney, Australia
Posts: 10,646
Rep Power: 84
ghorrocks has a spectacular aura aboutghorrocks has a spectacular aura aboutghorrocks has a spectacular aura about
I do not understand your post. As stumpy says, CFX uses the time step size instead of under relaxation as the main method to stabilise the equations.

So yes, under-relaxation is a "valid way to .. help convergence", but CFX uses time step size.

A converged solution is a converged solution, regardless of how you got there.

No idea what you mean about "the solution may be incorrect".
ghorrocks is offline   Reply With Quote

Old   April 1, 2011, 05:47
Default
  #5
Senior Member
 
Join Date: Oct 2010
Location: Zurich
Posts: 176
Rep Power: 6
Chander is on a distinguished road
Quote:
Originally Posted by ghorrocks View Post
So yes, under-relaxation is a "valid way to .. help convergence", but CFX uses time step size.

A converged solution is a converged solution, regardless of how you got there.

No idea what you mean about "the solution may be incorrect".
Glenn,

Ok, so reducing the timescale multiplication factor reduces the time step taken by CFX solver . I said 'solution may be incorrect' as stumpy replied that reducing the timescale factor may not be the right approach. But you commented that converged solution is a converged solution whatever may be the approach. So should I understand that that its ok to reduce the timescale factor by as much as one order of magnitude to get convergence? or even more?
CFX support had once cautioned me against reducing the timescale factor so much and said that I should check if something else is not right in my simulation setup. I am not sure why they cautioned so.
Chander is offline   Reply With Quote

Old   May 1, 2014, 00:30
Default query
  #6
New Member
 
Manpreet
Join Date: Jan 2014
Posts: 14
Rep Power: 3
manpreet is on a distinguished road
Hello Guys,
I need a help.
Could anyone tell me physical significance of time scale factor / physical timescale. In my project. whenever I use physical timescale 1E-5 or 6 then I get smooth curve for mass ,momentum etc equations and if I use 1E-4.then its large variation in curve. Wt does it mean?? I did not get it. Through tutorials I got one formula that is for advection dominated flow;
Timescale = delta t = L /U = length scale/ velocity scale

I did not get its meaning exactly.

Thanks
Manpreet Singh

manpreet_singh_er@yahoo.co.in
manpreet is offline   Reply With Quote

Old   May 1, 2014, 09:09
Default
  #7
Super Moderator
 
Glenn Horrocks
Join Date: Mar 2009
Location: Sydney, Australia
Posts: 10,646
Rep Power: 84
ghorrocks has a spectacular aura aboutghorrocks has a spectacular aura aboutghorrocks has a spectacular aura about
The time scale in a steady state simulation determines how fast you are advancing the solution. A very small time scale means very slow advancement and slow convergence. A large time scale means fast advancement and faster convergence - as long as you are not too fast because then it goes unstable and diverges.
ghorrocks is offline   Reply With Quote

Old   May 1, 2014, 12:44
Default
  #8
Member
 
Kevin Hoopes
Join Date: Oct 2010
Posts: 40
Rep Power: 7
khoopes is on a distinguished road
Quote:
Originally Posted by ghorrocks View Post
The time scale in a steady state simulation determines how fast you are advancing the solution. A very small time scale means very slow advancement and slow convergence. A large time scale means fast advancement and faster convergence - as long as you are not too fast because then it goes unstable and diverges.
This is exactly right. I would also suggest not judging convergence based on residuals alone. Create monitor points in CFX-Pre for things that interest you, drag, pressure drop, velocity and then monitor their evolution as you run the simulation. Ideally you should choose monitors based on the goal of the simulation as this is the data you will extract when its complete anyway. Using monitor points like this will tell you a lot about whether you are reaching steady state.
khoopes is offline   Reply With Quote

Reply

Thread Tools
Display Modes

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 Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
relaxation factor for coupling in FSI simulation taedeneo CFX 3 September 11, 2009 07:29
how to implement an user defined equation of state in CFX 10.0 IvanCFD CFX 1 August 16, 2009 08:28
chart for transiente simulation - CFX Post rogbrito CFX 0 June 22, 2009 12:12
material property in steady state CHT simulation NVSD BABU CFX 3 February 25, 2009 01:58
Two-Phase Buoyant Flow Issue Miguel Baritto CFX 4 August 31, 2006 12:02


All times are GMT -4. The time now is 00:44.