CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   DNS always returns to laminar initial solution (

kat September 6, 2007 06:00

DNS always returns to laminar initial solution

I'm trying to do a DNS calculation of a periodic channel flow. I start from a laminar velocity profile with a random component of 10 % added to it.

The code uses a staggered grid, central discretization in space, Adams-Bashforth in time and SOR for the Poisson equation.

I would like to achieve a turbulent velocity profile but the code always goes to the laminar one.

What could I do to prevent this?



tom September 6, 2007 09:06

Re: DNS always returns to laminar initial solution
Instead of disturbances with a short wave length try disturbances with a longer wave length. Some random cosine motions for example. Short wave length disturbances are killed by viscous forces.

Furthermore, is the comp.domain large enough and the resolution high enough?

kat September 6, 2007 09:38

Re: DNS always returns to laminar initial solution
Hi Tom

Thanks for your answer.

I use the "standard" channel configuration of Moser and Kim with at least 128 grid cells in each direction (denser at the wall).

Do you mean disturbances with a higher amplitude and/or longer wave lengths?

Thanks again


Joe September 6, 2007 10:04

Re: DNS always returns to laminar initial solution
What is the order of your scheme? Also, make sure the dissipation in your code is not too large and damping out the unsteadiness.

jojo September 6, 2007 10:13

Re: DNS always returns to laminar initial solution
Try and increase the Reynolds number. If the turbulence cannot naturally develop, your disturbance will also be killed. The disturbance does not create turbulence, it helps get it earlier. It saves your time but does not change the physics. If you are in the subcritical regime, no way for turbulence.

kat September 6, 2007 10:26

Re: DNS always returns to laminar initial solution
The scheme is second order central (only). Since the final goal is to use the code also "engineering" problems with arbitrary geometries (after implementing an LES model), it's not really practical to use higher order model, isn't it?

What could be reasons for too high dissipation? How could I test it?



kat September 6, 2007 10:29

Re: DNS always returns to laminar initial solution
Indeed, the same problem occurs when I start from a zero velocity field. It only takes more time to get to the laminar solution.

I will try your idea of increasing Re. (Now Re_turb=180 / Re about 5600).



pankaj saha September 6, 2007 17:43

Re: DNS always returns to laminar initial solution
hi, kat.... i am also working on DNS....and also facing the smae problem.....will you please disscuss with me through mail. so that we can worked together. i am also using staggered grid.....with MAc algorithm..

my mail please mail that we can discuss about the issue.

i have also read many papers on it..


tom September 7, 2007 01:53

Re: DNS always returns to laminar initial solution

The resolution and the Reynolds number seem to be OK. You can try a higher Reynolds number but then you have to increase the resolution as well and probably you still have the same problem (but you just spend more computation time to find it out). I had a channel of size 5H x 3H x H (streanwise, spanwise, wall-normal, H is the channel height). Resolution about 200 in the streamwise and about 128 in the other directions and Re_tau=180. Second-order finite volume worked just fine.

Try in particular disturbances with a longer wave length and perhaps you can increase the amplitude a bit. If the disturbances are too strong your code will blow up probably but otherwise it doens't matter. Just check carefully what happens with the disturbances initially. If they disappear quickly try something else.

phil September 8, 2007 03:13

Re: DNS always returns to laminar initial solution

If you use second order it's possible that the Kim & Moin resolution are unsuffisant in your case. Kim & Moin used spectral like method.

I think you must increase your spatial resolution and increase the level of disturbance in entry. Can you explain your method to add some noise in your laminar flow ? Is it a constant add-on or only at the start of the simulation ?

Sorry if i'm not very clear. Since fiew year P. Orlandi publish lot of result with a second order schemes in a channel flow. Orlandi try to proove that second order scheme can simulate this flow (channel flow) at the same resolution than Kim and Moin.

pankaj saha September 9, 2007 09:20

Re: DNS always returns to laminar initial solution

tom ----

I am also facing the same problem as kat.i am mentioning some question.

1. for initial condition , what mean velocity profile , i could use. can i use log-law profile and linear profile based on friction velocity as mean velocity profile. for that case , i have to non-dimensionalise the N-S equation's velocity by friction velocity. otherwise how could i convert the friction velocity to centralline velocity. 2. How could i superimpose the disturbence on mean velocity profile. could you please explain , how to implement on code.


All times are GMT -4. The time now is 12:47.