SST converges faster than SA

tomaspzapico · August 29, 2024, 06:31

Hello,

I have been running a simulation on the NACA 0012 airfoil at Re=6e6 and M=0.3, with first order for the turbulence equations and JST discretisation scheme. Even though SA is a 1-equation model and SST is 2-equations, SST is converging faster (both in iterations and time) than SA, which should not be happening. Does anyone know what is causing this?

bigfootedrockmidget · August 30, 2024, 08:00

SA does not necessarily converge faster in terms of iterations compared to SST. It really depends on the case.
SST has one more transport equation, so if every transport equation costs an equal amount of time, then you would expect SA to be slightly faster in terms of computing time per iteration. I would expect this to be true, because the SA equation is not twice as expensive to compute compared to the k-w equations. But there are case dependent computations, for instance SA needs to compute some wall distance, which might make it more expensive than SST under certain circumstances.

TKatt · August 31, 2024, 17:54

You could check the eddy viscosity for example, I could imagine if SST leads to a larger increase in overall viscosity, that might aid convergence.

If it converges faster in Iterations, the faster in overall time is not really that surprising in most scenarios. Following what Mr bigfootedrockmidget, the single iteration might not even be cheaper at all.

If you want to check that, you should look at time per single iteration. Put WALL_TIME in your SCREEN_OUTPUT and look at that. To level the playing field also make sure to have the linear solver doing similar work by e.g. setting LINEAR_SOLVER_ERROR= 1E-16.

August 29, 2024, 06:31	SST converges faster than SA	#1
tomaspzapico New Member Tomas Perez Zapico Join Date: Mar 2023 Posts: 17 Rep Power: 3	Hello, I have been running a simulation on the NACA 0012 airfoil at Re=6e6 and M=0.3, with first order for the turbulence equations and JST discretisation scheme. Even though SA is a 1-equation model and SST is 2-equations, SST is converging faster (both in iterations and time) than SA, which should not be happening. Does anyone know what is causing this?

August 30, 2024, 08:00		#2
bigfootedrockmidget Senior Member bigfoot Join Date: Dec 2011 Location: Netherlands Posts: 676 Rep Power: 21	SA does not necessarily converge faster in terms of iterations compared to SST. It really depends on the case. SST has one more transport equation, so if every transport equation costs an equal amount of time, then you would expect SA to be slightly faster in terms of computing time per iteration. I would expect this to be true, because the SA equation is not twice as expensive to compute compared to the k-w equations. But there are case dependent computations, for instance SA needs to compute some wall distance, which might make it more expensive than SST under certain circumstances. Sakun likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
One more time: wall functions - SST	CellZone	OpenFOAM Pre-Processing	22	June 11, 2021 14:27
Turbulence Model switch RKE to K-omega SST	Astan	OpenFOAM Running, Solving & CFD	3	April 6, 2019 04:54
General question about SST turbulence model	Richard Renaud	CFD Freelancers	1	December 19, 2015 05:13
wall treatment k-w SST model in Fluent	behest	FLUENT	0	December 26, 2014 09:14
Laminar-to-turbulent transition modelling using SST and SST Gamma Theta models	ttu004	CFX	3	June 13, 2014 07:43

August 31, 2024, 17:54		#3
TKatt Member na Join Date: Jul 2018 Posts: 90 Rep Power: 8	You could check the eddy viscosity for example, I could imagine if SST leads to a larger increase in overall viscosity, that might aid convergence. If it converges faster in Iterations, the faster in overall time is not really that surprising in most scenarios. Following what Mr bigfootedrockmidget, the single iteration might not even be cheaper at all. If you want to check that, you should look at time per single iteration. Put WALL_TIME in your SCREEN_OUTPUT and look at that. To level the playing field also make sure to have the linear solver doing similar work by e.g. setting LINEAR_SOLVER_ERROR= 1E-16.