I'm a freshman in DNS and want to do the direct simulation of plane turbulence channel flow. The plane turbulence channel flow has been studied widely and have many available results, So I choose this case as my start.(the flow is 3-D,the periodic boundary condtions are imposed at the streamwise and spanwise, the no-slip boundary condition is imposed at the upper and lower wall) In the simulation, I found I can't given the proper initial condition. From the literature, some authors says they start their simulation from the laminar velocity profile with random noise. So I write the following code as the intial condition of computation.

ran=1.0e-8; srand(2); //create a seed of random-number for (i=2;i<L;i++) {

for (j=1;j<=M;j++)

{

for (k=2;k<N;k++)

{

u[i][j][k]=ran*rand()/double(RAND_MAX); v[i][j][k]=ran*rand()/double(RAND_MAX); w[i][j][k]=1.0-uy[j]*uy[j]+ran*rand()/double

(RAND_MAX); ///the main flow profile

p[i][j][k]=0.0;

dp[i][j][k]=0.0;

}

} } In the computation, if I large the variable ran, then the pressure may not convergence. If I choose ran=1.0e-8 or even smaller, as the advance of time, these random noise may decrease step by step and the main flow w[i][j][k] didn't changeat all.

So who can help walk out this quandary?? Thanks a lot!!

Hi,

It is better to introduce some large scale disturbances. Random noise with a very short wave length damps out rather quickly. Therefore, I just take some random sine/cosine motions with different wave lngths. Details do not matter so much and it does not have to be diverge free, at least if your code is working correctly. The initial fluctuations should be about 5 or 10% of the mean flow. If your code can't handle that then you should have a critical look at your algorithm.

Success,

Tom