Roe upwind scheme

simone1593 · February 24, 2020, 09:03

Hi everyone, I am trying to computing some aerodynamic coefficients (drag, lift, coefficient of pressure) using the Roe upwind scheme, but it doesn't converge. I tried also to use the multigrid method, but I had no success. Are there any options that might I have missed in my .cfg file?
wingD.txt

aero_amit · February 24, 2020, 12:12

Can you share the cfg file. Multigrid with Roe does not work well in SU2 (usually JST goes well) .

simone1593 · February 25, 2020, 15:55

Hi aero_amit, I have shared the cfg file yet (it is the txt file named wingD). I must use Roe scheme for "academic issues", is there a way to reach convergence?

pcg · February 26, 2020, 06:09

In my experience the multigrid works very well in hex-dominant grids, provided you are not using too many cores relative to problem size (see pull request 861 on GitHub)
My advice (based on RANS cases):
Use only pre-smoothing (1, 1, 2, 3), 2-3 levels should be enough.
0.7 damping coefficients.
Limiters tend to make the case flip flop, Venkatakrishnan-Wang works best but try using no limiter (your case is subsonic). If you really want limiters, increasing the Venkat coefficient may help.
Fixed CFL in the range 15-25, going higher sometimes makes convergence harder.
Increase the entropy fix factor until you get convergence, then worry about its influence on the solution (once you have a converged solution for initialization it is easier to converge with lower fix factors).
FGMRES + LU-SGS, no more than 10 linear iterations and tolerance of 0.05 or so should be more than enough, especially on inviscid grids.
I stay away from the low-Mach schemes / options, you can probably do without them too at Mach 0.2.

simone1593 · February 26, 2020, 09:14

Hi pcg, my case is an Euler one, with a low Mach (0.2). Is your advice still valid for my situation?
I have another (maybe stupid) question: I have ever studied finite difference methods right now, and for upwind schemes the CFL condition is less or equal than 1. I noticed that either in SU2 tutorial or in the other threads the suggestions take in consideration CFL numbers of 10/20/30 and so on...are the upwind schemes (like Roe's scheme, for example) still stable at such a high CFL number?

I have started using SU2 and finite volume method since a very little time, so I ignore a lot of things! :S

pcg · February 27, 2020, 03:24

I would say it is worth a shot.

The CFL limit depends on the (pseudo)time integration method, for explicit Euler (the integration, not the equations) it is very low like you say, Runge-Kutta methods allow slightly higher values, and implicit Euler is unconditionally stable (but the implicit part is based on an approximate linearization of the residual so there are practical limits).

simone1593 · February 27, 2020, 05:07

That's right, but there is a thing that I can't understand: I have two grids, a coarse one and a finer one; if I try to do the computation on the coarser grid it converges, but if I use the same options on the finer one, it does not converge. Which could be the reason?

February 24, 2020, 09:03	Roe upwind scheme	#1
simone1593 New Member SALERNO Join Date: Jan 2020 Posts: 4 Rep Power: 6	Hi everyone, I am trying to computing some aerodynamic coefficients (drag, lift, coefficient of pressure) using the Roe upwind scheme, but it doesn't converge. I tried also to use the multigrid method, but I had no success. Are there any options that might I have missed in my .cfg file? wingD.txt

February 26, 2020, 06:09		#4
pcg Senior Member Pedro Gomes Join Date: Dec 2017 Posts: 466 Rep Power: 13	In my experience the multigrid works very well in hex-dominant grids, provided you are not using too many cores relative to problem size (see pull request 861 on GitHub) My advice (based on RANS cases): Use only pre-smoothing (1, 1, 2, 3), 2-3 levels should be enough. 0.7 damping coefficients. Limiters tend to make the case flip flop, Venkatakrishnan-Wang works best but try using no limiter (your case is subsonic). If you really want limiters, increasing the Venkat coefficient may help. Fixed CFL in the range 15-25, going higher sometimes makes convergence harder. Increase the entropy fix factor until you get convergence, then worry about its influence on the solution (once you have a converged solution for initialization it is easier to converge with lower fix factors). FGMRES + LU-SGS, no more than 10 linear iterations and tolerance of 0.05 or so should be more than enough, especially on inviscid grids. I stay away from the low-Mach schemes / options, you can probably do without them too at Mach 0.2. giovanni.medici and ugurtan666 like this.

February 27, 2020, 03:24		#6
pcg Senior Member Pedro Gomes Join Date: Dec 2017 Posts: 466 Rep Power: 13	I would say it is worth a shot. The CFL limit depends on the (pseudo)time integration method, for explicit Euler (the integration, not the equations) it is very low like you say, Runge-Kutta methods allow slightly higher values, and implicit Euler is unconditionally stable (but the implicit part is based on an approximate linearization of the residual so there are practical limits). giovanni.medici and ugurtan666 like this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
bounded Gauss upwind Scheme	deepinheart	OpenFOAM Running, Solving & CFD	1	February 23, 2015 05:57
Implementation of 2nd order upwind scheme	jaason	OpenFOAM Running, Solving & CFD	4	February 6, 2015 17:40
2nd order upwind vs 2nd order upwind!!!	Far	Main CFD Forum	7	March 14, 2013 12:29
Use of upwind scheme for interpolation of u/v	quarkz	Main CFD Forum	6	August 30, 2011 04:10
2nd order upwind scheme (Fluent and CFX)	Far	FLUENT	0	May 22, 2011 01:50

February 24, 2020, 12:12		#2
aero_amit Member Amit Join Date: May 2013 Posts: 85 Rep Power: 13	Can you share the cfg file. Multigrid with Roe does not work well in SU2 (usually JST goes well) .

February 25, 2020, 15:55		#3
simone1593 New Member SALERNO Join Date: Jan 2020 Posts: 4 Rep Power: 6	Hi aero_amit, I have shared the cfg file yet (it is the txt file named wingD). I must use Roe scheme for "academic issues", is there a way to reach convergence?

February 26, 2020, 09:14		#5
simone1593 New Member SALERNO Join Date: Jan 2020 Posts: 4 Rep Power: 6	Hi pcg, my case is an Euler one, with a low Mach (0.2). Is your advice still valid for my situation? I have another (maybe stupid) question: I have ever studied finite difference methods right now, and for upwind schemes the CFL condition is less or equal than 1. I noticed that either in SU2 tutorial or in the other threads the suggestions take in consideration CFL numbers of 10/20/30 and so on...are the upwind schemes (like Roe's scheme, for example) still stable at such a high CFL number? I have started using SU2 and finite volume method since a very little time, so I ignore a lot of things! :S

February 27, 2020, 05:07		#7
simone1593 New Member SALERNO Join Date: Jan 2020 Posts: 4 Rep Power: 6	That's right, but there is a thing that I can't understand: I have two grids, a coarse one and a finer one; if I try to do the computation on the coarser grid it converges, but if I use the same options on the finer one, it does not converge. Which could be the reason?