I have a problem with second order accurate solver

mym1411 · March 13, 2010, 17:35

Hi
I am trying to simulate a Mach 6 laminar flow around ahemispherical cylinder body. The problem is steady and I start with first-order accurate discretization. The convergence is perfect and the contours are smooth and OK as long as the solver is first-order. After some inerations, I switch to second-order ccuracy for better results. The forces remain unchanged, but the residulas jump to from 10e-5 to say 10e-1, and the contours get less smooth. Then happens even if I started with second order from the beginning. I tried Roe and AUSM schemes, but they rae the same.
Would you kindly help me with this issue?
Thanks in advance
Mahmoud

DoHander · March 13, 2010, 18:59

It is normal to have a jump in residual when you switch from first order to second order, but after a few iterations the residual should go down.

If you want to start from start with a second order solution, normally this will converge slower then your first order (especially if you have a transonic/supersonic speed) ... If after say 3-5000 iteration you see no improvements in residuals you should use a lower CFL number (if you use the density based solver) or smaller relaxation factors. If none of the above works ... change your mesh.

You could try to give more details: what kind of mesh are you using, turbulence model, y+, solver and so on ...

Do

mym1411 · March 14, 2010, 04:31

Hi Do
Thanks a lot for your reply.
the case is laminar, so, no y+ value. the solver is density-based, the mesh is structured with very small cell size at the walls and in the shock region (1e-4m). I start with very samll CFL value and increase it gradually. The solution converes to high residuals with second-order (from the start or fed from a first-order) The resiuals do not improve even with more iterations

What do you think?

March 13, 2010, 17:35	I have a problem with second order accurate solver	#1
mym1411 New Member mahmoud Join Date: Mar 2010 Posts: 3 Rep Power: 16	Hi I am trying to simulate a Mach 6 laminar flow around ahemispherical cylinder body. The problem is steady and I start with first-order accurate discretization. The convergence is perfect and the contours are smooth and OK as long as the solver is first-order. After some inerations, I switch to second-order ccuracy for better results. The forces remain unchanged, but the residulas jump to from 10e-5 to say 10e-1, and the contours get less smooth. Then happens even if I started with second order from the beginning. I tried Roe and AUSM schemes, but they rae the same. Would you kindly help me with this issue? Thanks in advance Mahmoud

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
problem with developing new LES solver	Edison_Ge	OpenFOAM	2	June 18, 2009 02:26
CFX 11 x64 solver? problem	Attesz	CFX	6	June 7, 2009 09:37
Through-Flow solver, Leading-Trailing edge problem	Menander	Main CFD Forum	0	May 14, 2008 19:14
Problem going to second order	andrewburns	OpenFOAM Running, Solving & CFD	1	February 11, 2008 18:28
Coupled solver problem!	Alain	FLUENT	0	February 28, 2007 07:19

March 13, 2010, 18:59		#2
DoHander Senior Member Join Date: Nov 2009 Posts: 411 Rep Power: 20	It is normal to have a jump in residual when you switch from first order to second order, but after a few iterations the residual should go down. If you want to start from start with a second order solution, normally this will converge slower then your first order (especially if you have a transonic/supersonic speed) ... If after say 3-5000 iteration you see no improvements in residuals you should use a lower CFL number (if you use the density based solver) or smaller relaxation factors. If none of the above works ... change your mesh. You could try to give more details: what kind of mesh are you using, turbulence model, y+, solver and so on ... Do

March 14, 2010, 04:31		#3
mym1411 New Member mahmoud Join Date: Mar 2010 Posts: 3 Rep Power: 16	Hi Do Thanks a lot for your reply. the case is laminar, so, no y+ value. the solver is density-based, the mesh is structured with very small cell size at the walls and in the shock region (1e-4m). I start with very samll CFL value and increase it gradually. The solution converes to high residuals with second-order (from the start or fed from a first-order) The resiuals do not improve even with more iterations What do you think?