number of iterations per time step

mortazavi · August 5, 2016, 08:13

hi,
my transient simulation of flow over a 3D hydrofoil converges in each time step after 1-2 iterations (the courant number is around 0.3).
Do you think I have to increase the time step or it is ok? (convergence target 1e-5)

ghorrocks · August 6, 2016, 02:00

Yes, you can probably increase the time step size. Why not just use adaptive time stepping homing in on 3-5 coeff loops per iteration and then it will sort it out for itself?

Sasquatch · August 6, 2016, 06:16

If it converges within 1-2 iteration loops the time step is probably sufficiently small, however, the computational time might be unnecessarily big.

On the other side, if too many inner loops are required, the time step is probably too big.

Keeping coeff. loops between 10-15 is sometimes recommended but I have to admit that my experience on this front is still small.

Regards,

ghorrocks · August 6, 2016, 06:33

The general recommendation is 3-5 coeff loops per iteration. The recommendation for 10-15 applies only for simulations with complex multiphase physics, especially where the phase equations couple strongly to the momentum/mass equations. I have hardly ever used this either, the majority of multiphase stuff I do is free surface modelling and that works best with 3-5 coeff loops as well.

mortazavi · August 6, 2016, 11:22

what about the courant number then? It is not around 1 anymore if i increase the timestep.

mortazavi · August 6, 2016, 11:25

Does it affect the accuracy of final solution if i increase the time step in the middle of my simulation? (edit run in progress)

ghorrocks · August 7, 2016, 06:52

CFX is an implicit solver. It does not have a direct link between Courant Number and stability like explicit solvers do. The time step requirement is set more by numerical accuracy considerations, which means you adjust the time step to get the accuracy you require - and this means there is no fixed Courant number requirement.

Some CFX simulations can be done with very large Courant number (eg flows tending towards steady state), yet some require Courant numbers much less than 1 (eg surface tension driven flows).

So I do not recommend use of Courant number to assess time step size as it does not tell you much. A mush better thing to do is either a time step sensitivity study to determine the time step you need for the accuracy you need; or adaptive time stepping homing in on 3-5 coeff loops per iteration.

Large changes in time step can be problematic. Also note that if you are using second order time stepping you will probably loose most of the second order components on the time step you change size on. This is the same as large changes in mesh size are bad for stability and accuracy.

August 5, 2016, 08:13	number of iterations per time step	#1
mortazavi Member SMN Join Date: Jun 2009 Location: CANADA Posts: 71 Rep Power: 16	hi, my transient simulation of flow over a 3D hydrofoil converges in each time step after 1-2 iterations (the courant number is around 0.3). Do you think I have to increase the time step or it is ok? (convergence target 1e-5)

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Decomposing meshes	Tobi	OpenFOAM Pre-Processing	22	February 24, 2023 09:23
time step continuity problem in VAWT simulation	lpz_michele	OpenFOAM Running, Solving & CFD	5	February 22, 2018 19:50
Extrusion with OpenFoam problem No. Iterations 0	Lord Kelvin	OpenFOAM Running, Solving & CFD	8	March 28, 2016 11:08
SLTS+rhoPisoFoam: what is rDeltaT???	nileshjrane	OpenFOAM Running, Solving & CFD	4	February 25, 2013 04:13
Could anybody help me see this error and give help	liugx212	OpenFOAM Running, Solving & CFD	3	January 4, 2006 18:07

August 6, 2016, 02:00		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	Yes, you can probably increase the time step size. Why not just use adaptive time stepping homing in on 3-5 coeff loops per iteration and then it will sort it out for itself?

August 6, 2016, 06:16		#3
Sasquatch Member Join Date: Jul 2016 Posts: 33 Rep Power: 9	If it converges within 1-2 iteration loops the time step is probably sufficiently small, however, the computational time might be unnecessarily big. On the other side, if too many inner loops are required, the time step is probably too big. Keeping coeff. loops between 10-15 is sometimes recommended but I have to admit that my experience on this front is still small. Regards,

August 6, 2016, 06:33		#4
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	The general recommendation is 3-5 coeff loops per iteration. The recommendation for 10-15 applies only for simulations with complex multiphase physics, especially where the phase equations couple strongly to the momentum/mass equations. I have hardly ever used this either, the majority of multiphase stuff I do is free surface modelling and that works best with 3-5 coeff loops as well.

August 6, 2016, 11:22		#5
mortazavi Member SMN Join Date: Jun 2009 Location: CANADA Posts: 71 Rep Power: 16	what about the courant number then? It is not around 1 anymore if i increase the timestep.

August 6, 2016, 11:25		#6
mortazavi Member SMN Join Date: Jun 2009 Location: CANADA Posts: 71 Rep Power: 16	Does it affect the accuracy of final solution if i increase the time step in the middle of my simulation? (edit run in progress)

August 7, 2016, 06:52		#7
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	CFX is an implicit solver. It does not have a direct link between Courant Number and stability like explicit solvers do. The time step requirement is set more by numerical accuracy considerations, which means you adjust the time step to get the accuracy you require - and this means there is no fixed Courant number requirement. Some CFX simulations can be done with very large Courant number (eg flows tending towards steady state), yet some require Courant numbers much less than 1 (eg surface tension driven flows). So I do not recommend use of Courant number to assess time step size as it does not tell you much. A mush better thing to do is either a time step sensitivity study to determine the time step you need for the accuracy you need; or adaptive time stepping homing in on 3-5 coeff loops per iteration. Large changes in time step can be problematic. Also note that if you are using second order time stepping you will probably loose most of the second order components on the time step you change size on. This is the same as large changes in mesh size are bad for stability and accuracy.