CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   Adaptive timestep (

siw November 12, 2012 05:21

Adaptive timestep
2 Attachment(s)

Glenn posted some information about using the adaptive timestep for a transient simulation when the timestep is not known beforehand ( I need to do the same.

However, in the other topic it is not clear to me were the mentioned coefficient loops are input (analysis type or solver control - see images) since I've tried and they don't like being the same. So in which should the recommend 3-5 be input?

Also, what's the reason for assessing the influence of convergence tolerance? I always set my simulations to a residual target = 1E-06 or lower and let the solver keep iterating until it reaches whatever values it gets too with enough iterations to observe either constant residual values (steady flow) or oscillating residual values (unsteady flow). Then there's no need to assess this parameter.


evcelica November 12, 2012 19:53

Analysis type is where you would set up your adaptive time-steps. Solver control is where you would set your convergence criteria and min/max number of time-steps per loop. The timestep will be adapted to try to meet your convergence criteria in 3-5 timesteps, but will never go over 10, or under 1, (or whatever you have set in solver control)

1E-6 might not be possible in some simulations, or may be terribly inefficient for some transient runs. But it sounds like you are talking about a steady state simulation in your last sentence?

siw November 13, 2012 03:19

5 Attachment(s)

I applied those data from Glenn's post as I thought he was detailing and it has greatly improved all the convergences, compared to the pre-cursor steady-state run (200 iterations) which I used for establishing an initial flowfield. When looking at the *.out file on the last few iterations (which would done for only 3 coeff loops) the solver found at timestep size (2.1074E-08 s). Is it now better practice to continue over the next few thousand iterations using this constant time step?

I was suprised not to see oscillations in the monitor points though. The simulation is a transonic aerodynamic flow behind some bluff bodies, so I'm expecting unsteady behaviour. Next I'm going to keep everything the same but use SAS-SST rather than URANS-SST.

My last sentance was just about why would anyone assess the convergence tolerance when all I do is set it really low.

evcelica November 13, 2012 12:28

You set it "really low" which is causing your timestep to have to get extremely small to meet that convergence criteria. If you were to set your convergence criteria higher you could solve it in much much less time. Then you can change your convergence criteria which would also change your timestep size and you would be checking for convergence and timestep independance.
Your idea of just "setting it really low" is just innefficient.
Also 10^-6 RMS may not be good enough for some types of flows, or may be unattainable for others, meaning your adaptive timestepping would keep on decreasing and decreasing and roundoff errors start to effect the solution.

I would leave the timestepping on adaptive if you are going to run more iterations, there will be times in unsteady flow where a larger timestep can be used and other times when a smaller timestep has to be used; let the solver adapt itself to the flow.

All times are GMT -4. The time now is 19:31.