CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   When to switch from steady-state to transient (

Nick R April 6, 2011 23:55

When to switch from steady-state to transient
Hi all,

In a transitional simulation for a thick airfoil at a low Reynolds at a 4deg AOA number I find that residuals start to oscillate past a certain point. My guess is transient effects are coming into effect. My question is : Should the transient solver be initiated "before" the oscillations start or at some arbitrary point after?
My second question is with regard to the Gamma Theta model and the grid requirement, is the 75 nodes advised by the manual upstream of the transition point or downstream ?
Thanks for your attention.

ghorrocks April 7, 2011 07:34

If all you care about is the steady state (or time averaged if transient) result then it usually does not matter how you get there. Starting with a unconverged steady state run is a lot better than a crude initial guess and will mean your transient run settles down much faster.

I was not aware of the 75 node recommendation. I would generalise and say do a mesh sensitivity check anyway and find out for yourself.

Nick R April 7, 2011 11:55

Thanks for your comments. Also I have a question about what to expect with the transient model. Let's say the mass residual starts to oscilate when it's at 10^-3 in the steady solver. Are the residuals supposed to converge during each time step in the transient solver? I mean are they supposed to go from 10^-3 to say 10^-5 at each smalll time step?

ghorrocks April 7, 2011 18:10

Residuals are calculated quite differently between steady and transient runs. The tolerances are not necessarily transferrable.

And yes, you need every transient time step to converge for a good solution. But if all you care about is the final solution (sorry Hitler) then a few dud timesteps is OK as long as the final result is well converged.

Nick R April 7, 2011 21:13

:) I see. I'll give it a shot. Thanks.

icemaniac178 December 15, 2011 00:52

Dear Nick,
i am doing the almost the same analysis as u. simulating a wing at low reynolds number condition. i`im using the sst turbulence modelling but my results cannot converge in my steady state. i suspect maybe there is a transient phenomenon in my case here. but i don`t have any proper guidence on the transient state. can u give me some advice on timestep selection?

FYI information my analysis is 3D wing. Anybody who have experience on this transient analysis are also appreciated to share and discuss.

ghorrocks December 15, 2011 17:17

There is some advice on time step selection here:

Nick R December 15, 2011 20:41

I would suggest starting with a very small timestep and watching the convergence behavior for a few time steps. If convergence target is achieved then try increasing the time step; otherwise you need to switch to even a smaller timestep.

icemaniac178 December 15, 2011 20:45

thanx mr ghorrocks,
i have already implemented the method listed in the link.
my monitor points are not showing any significant amaunt of fluctuation value, but my momentum and mass residual is still flunctuated and not well below 1e-4.
what is the acceptable percentage difference of this monitor point flunctuation value?
i suspect my low reynold number region (1e5) might introduce the transition region. therefore, i used the SST with theta gamma model to solve it. but unconverged solution keep on bothering me especially at higher AoA (>15).FYI, this is a FSI study on low reynolds number wing.

ghorrocks December 16, 2011 16:12

If you are in a transitional turbulence region then it is highly likely you have laminar separation bubbles and the like, they are frequently transient. In my experience I often have to run the transitional turbulence model transient because no steady solution exists.

Nick - In these cases you increase the timestep size, not decrease it. Once the initial convergence is done you increase time step size so the turbulence model can better capture small transient features.

Nick R December 16, 2011 20:57

Ghorrocks- I wrote once convergence is achieved try increasing the timestep otherwise (if not achieved) decrease it. I think this is accurate in a transient solution.

ghorrocks December 17, 2011 06:36

For a steady state simulation the usual approach is to use some small timesteps initially to get it started, then increase the timesteps as convergence starts to occur. You can run quite large time steps to accelerate convergence in many cases.

For transient runs the usual approach is to adjust the time step size to get 3-5 coeff loops per time step. This is easiest done using adaptive time stepping.

In general (either steady or transient) if the simulation is diverging then one of the first things to try is decreasing the time step to recover it. But you only do this if there are signs of divergence.

Far December 18, 2011 17:05


My second question is with regard to the Gamma Theta model and the grid requirement, is the 75 nodes advised by the manual upstream of the transition point or downstream ?
On the complete surface. If you read that report again you will see that authors have advised to have enough nodes in the transitional region to accurately resolve the flow, otherwise transition model may give you wrong prediction or no predictions at all. Also see this post

Here I am reproducing the exact words of that report by Langraty
The solution differed significantly from the grid independent one only for the case of 25 streamwise nodes where there was only one cell in the transitional region. Nevertheless the grid independent solution appears to occur when there is approximately 75 100 streamwise grid points from the leading edge to the trailing edge (i.e. one chord length). As well, separation induced transition occurs over a very short length and for cases where this is important a fine grid is necessary.

In short make sensitivity analysis as suggested by Glenn again and again.

Few more lines from that report:
One point to note is that for sharp leading edges often transition can occur at the leading edge due to a small leading edge separation bubble. If the grid is too course, the rapid transition caused by the separation bubble is not captured. A good example of this is the Zierke compressor. Contours of velocity (top) and turbulence intensity (Tu, bottom) for the Zierke compressor are shown in Figure A.5. On the suction side transition occurs at the leading edge due to a small leading edge separation bubble. On the pressure side transition occurs at about mid-chord. The effect of stream-wise grid resolution on resolving the leading edge laminar separation and subsequent transition on the suction side is shown in Figure A.6. Clearly, if there are not a large enough number of streamwise nodes, the model cannot resolve the rapid transition and a laminar boundary layer on the suction side is the result.

Note that if separation induced transition is present additional grid points in the streamwise direction are most likely needed.

All times are GMT -4. The time now is 10:18.