# first order converged--second order diverged

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

 November 26, 2009, 13:44 first order converged--second order diverged #1 New Member   Shengyi Wang Join Date: Mar 2009 Posts: 22 Rep Power: 13 Dear all, In my case, I use first order upwind scheme and get a very well converged solution; however, when I switch to second order upwind scheme, totally diverge!!! Any one has some suggestion to resolve this problem? Thx in advance. Sy

 November 27, 2009, 01:35 #2 Senior Member     p ding Join Date: Mar 2009 Posts: 380 Rep Power: 14 Would you please tell me the implementation method of the second order upwind scheme? and have you try to change the relaxation parameter

November 27, 2009, 10:48
#3
New Member

Shengyi Wang
Join Date: Mar 2009
Posts: 22
Rep Power: 13
Quote:
 Originally Posted by ztdep Would you please tell me the implementation method of the second order upwind scheme? and have you try to change the relaxation parameter
Hi Ztdep,

I am sorry I do not quite understand what you mean by the implementation. Could you please give a more detailed explanation?

Underralaxation factor has been tuned down, but it doesn't work.

I am using sst k-w with transition model, y+ is less than 1. The blade is actually a NACA0012 blade with an angle of attack of 8 degree.

R, Shengyi

 December 23, 2009, 06:57 RE #4 New Member   Ioan Join Date: Dec 2009 Posts: 5 Rep Power: 13 Dear gmwsy, To get a stable second-order simulation you should do the following trick: start the simulation for the first let say 100 timesteps with a first-order scheme and after switch to the second-order one. The problem behind is that for high-order schemes is very hard to start from "bad" initial solution (the initial solution is illconditioned)...

December 23, 2009, 13:45
#5
New Member

Shengyi Wang
Join Date: Mar 2009
Posts: 22
Rep Power: 13
Quote:
 Originally Posted by tele Dear gmwsy, To get a stable second-order simulation you should do the following trick: start the simulation for the first let say 100 timesteps with a first-order scheme and after switch to the second-order one. The problem behind is that for high-order schemes is very hard to start from "bad" initial solution (the initial solution is illconditioned)...
Dear Tele,

Thank you for your reply. But what i am doing is actually a steady case...so there does not exist time steps or initial solution.

December 25, 2009, 02:34
#6
New Member

Zhao
Join Date: Dec 2009
Location: China
Posts: 4
Rep Power: 13
Quote:
 Originally Posted by gmwsy Dear Tele, Thank you for your reply. But what i am doing is actually a steady case...so there does not exist time steps or initial solution.
I agree with tele's suggestion. For steady case, do some iteration with low order scheme, then swith to high order scheme.

December 25, 2009, 09:39
#7
New Member

Shengyi Wang
Join Date: Mar 2009
Posts: 22
Rep Power: 13
Quote:
 Originally Posted by mfren I agree with tele's suggestion. For steady case, do some iteration with low order scheme, then swith to high order scheme.
Thank you! But this is just what I have done. The iteration diverged when switched to 2nd order from a converged solution using 1st order...

 December 28, 2009, 04:50 #8 New Member   Join Date: Apr 2009 Posts: 1 Rep Power: 0 Dear gmwsy I would suggest to carefully check your mesh. Second order usually relies on an enlarged stencil, and will be much more sensitive to the mesh quality. Try to find where the convergence problem comes from, and readjust your mesh in the area. You may also fix it using some classical mesh quality criteria (determinant, skewness, etc.). BR koo

December 30, 2009, 05:37
#9
New Member

Shengyi Wang
Join Date: Mar 2009
Posts: 22
Rep Power: 13
Quote:
 Originally Posted by koo Dear gmwsy I would suggest to carefully check your mesh. Second order usually relies on an enlarged stencil, and will be much more sensitive to the mesh quality. Try to find where the convergence problem comes from, and readjust your mesh in the area. You may also fix it using some classical mesh quality criteria (determinant, skewness, etc.). BR koo
Dear Koo,

Thank you for your reply. I have checked the mesh and obviously the problem occurs at the boundary layer. Because I am not using the wall function and thus y+ is around 1, so the aspect ratio of the cells is very big. But this is usual in my area. I do not know how to change the mesh.

Best wishes to the new year,
Shengyi

January 1, 2010, 18:50
#10
Senior Member

Join Date: Jul 2009
Posts: 260
Rep Power: 14
Quote:
 Originally Posted by gmwsy Dear Koo, Thank you for your reply. I have checked the mesh and obviously the problem occurs at the boundary layer. Because I am not using the wall function and thus y+ is around 1, so the aspect ratio of the cells is very big. But this is usual in my area. I do not know how to change the mesh. Best wishes to the new year, Shengyi
Hi Shengyi,

What meshing program are you using? What solver? NB FLUENT has a very useful mesh adaption criterion specially for boundary layers_: Boundary layer adaption

January 1, 2010, 19:08
#11
New Member

Shengyi Wang
Join Date: Mar 2009
Posts: 22
Rep Power: 13
Quote:
 Originally Posted by kingjewel1 Hi Shengyi, What meshing program are you using? What solver? NB FLUENT has a very useful mesh adaption criterion specially for boundary layers_: Boundary layer adaption
Dear Kingjewel1:

Thanks for reply. I know about the adaption function in Fluent, but it might not help a lot. I recently found a potential solution for this problem: using a k-e model to get a solution and then switch to k-w.