Hi, Previously I've asked the question of extracting CFL number for a CFX solution. However it was explained to me that this is not required as CFX is a fully implicit solver ? How does this then relate to transient simulations ? As I understand it for LES simulations the timestep should be small enough for a CFL number of 1. Am I correct in assuming this and if so, how should I set this up usiing CFX ? In these cases is the CFL number still relevant ?

My other question is regarding mesh dependency tests. For LES or DES simulations when conducting a mesh dependency test, should you always do a time step dependency test (again this brings you back to the CFL number question.) ?

Hope my questions are not as confused as I am feeling.

Bob

Hi Bob,

Yes, you should always do time step size dependancy checks. You will probably find that there is a weak correlation with mesh size and time step size, which will be caused by the finer mesh resolving more flow features.

In my experience the CFL number only becomes relevant in implicit CFD codes when you are simulating sonic and supersonic flows. Back when I was using CFX4 to simulate transient shock wave movements I remember being limited to a maximum CFL of about 0.5. There will probably be a similar limit in CFX5.

However, for subsonic flow you can run timesteps much larger than CFL=1 for implicit solvers like CFX, and the CFL number is no longer of any significance.

Regards, Glenn

Hi Glenn, cheers for the reply. How should I best choose my time step size then ? If CFX does not follow the CFL number concept, then is it a case of purely undertaking timestep sensativity test, based upon a sensible initial guess for time step size based upon Re and Mesh size ? Bob

Dear Bob,

the Courant Number is a useful way to analyze CFD simulations and schemes and often is used in 2 regards:

1. Stability

2. Accuracy

The CFL condition was derrived by Courant, Friedrichs and Lewey in 1928 and deals with the stability of an explicit scheme.

Although they are connected issues there is often some confusion. For example any advice which indicates that one should run a fully implicit scheme at a Courant number greater than 1 for a transient calculation is only looking at the stability issue. Accurate results ie 1% or less numerical error will never be obtained even with second order (in time and space) schemes for transient runs with Cr>1.

Explicit schemes will become unstable above some Courant number, usually 1.0

Since accurate simulations are important for LES, people are often running at Cr < 1 anyhow and sometimes prefer the economy of an explicit scheme. There are lots of combinations of which terms to treat explicitly and which to treat implicitly.

CFX-TASCflow will give a warning if you are running at Cr>1 see below. I am not sure if CFX-5 gives the same warning?

There are lots of good references on these issues. See Ferziger and Peric "Computational Methods for Fluid Dynamics"

Hirsch: "Numerical Computation of Internal and External Flows"

Regards,

BAK_FLOW ================================================== ============================ TIME STEP =11764 ( 1) SIMULATION TIME = 3.87E-02 CPU = 2.45E+07 (4.48E+01) Wall clock time of day: Tue Nov 26 11:11:14 2002 Seconds since midnight: 40274.000 | | | ****** Notice ****** | | The CFL number is bigger than 1 at 1.296 % of the nodes | | (maximum CFL number is 7.147 at node [19,36,24]:MAIN). | | To increase the accuracy of the transient computation, | | please reduce the time step size. | | |

Hi Bob,

For a starting point, take the timescale of the fastest mechanism you would expect in the model (maybe the rotation speed of a vortex, oscillation frequency, or transport time of the fluid through an important section of the flow), and divide it by 100. This means you will be resolving that flow feature with 100 timesteps.

As this is only a first guess, from there do a timestep sensitivity check to ensure your timestep accurately resolves the flow. For complex flows it will need to be finer, for simple flows you may be able to run coarser timesteps. It is problem dependant so this will be up to you.

If you are doing LES then this will be tricky (read "interesting") as the finest flow features are determined by the resolution of your mesh, so your timestep will have significant dependance on mesh size.

BAK_FLOW makes the point that LES models often use explicit schemes as they are simpler, quicker and you often have to run the fine timesteps they require anyway with LES. If you are trying to do "serious" LES then CFX might not be the best code to use, you should consider an explicit code. For flows which are only weakly turbulent, or where the largest turbulent structures are reasonably large compared to the flow, then CFX's implicit scheme should work well.

Regards, Glenn

Dear Glenn and Bak_Flow, That has really cleared up some of the questions/issues I had. Thanks for the help Bob