CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   physical time step (

Alex August 14, 2002 13:29

physical time step
hi everybody, i want to know how choose a physical time step for my run, what factors i must take


Astrid August 14, 2002 15:55

Re: physical time step
Devide a characteristic length by a characteristic velocity. This should give you a rough estimate of the time step you can use. If you run into FINMES, devide the time step by 2.

You can also use Autotimestep. Then CFX5 will calculate a time step automatically.


Dave August 15, 2002 09:20

Re: physical time step
Astrid, do you have any preference as to using automatic, physical or automatic wich a fixed maximum or even local time stepping ?? Are there times when its better to use one method over another ?? Dave

Astrid August 16, 2002 02:02

Re: physical time step
Based on my experience with CFX 5.5 but mainly CFX 5.4.1, the following:

- Local timestepping is sometimes handy for startup. But as the timestep used is not provided by CFX, you do not know what you are doing exactly.

- For simple flow problems you can use autotime stepping. This gives some flexibility and may lead to a solution within a short period of time as time steps can increase very fast. The same applies for auto timestepping with a fixed maximum.

- I use physical timestepping for more complicated flow problems (e.g., M > 1). Then it might be required to fix the time step, if you want a solution anyway.

Hope this helps you, Astrid

Dave August 16, 2002 05:54

Re: physical time step
Astrid, cheers for the pointers on time stepping. Dave

Robin August 16, 2002 10:44

Re: physical time step
Hi Dave,

A local timestep factor may be used if you have a really hard time getting the solution started. In my experience, I have rarely required to use it as it is more often the initial guess that causes problems. If you use a local timestep factor, you should always switch to a physical timestep and converge the solution before interpreting the results.

For most problems, you should be able to get away with a timestep as large as the characteristic time of your problem. For an advection driven problem, this is the time it takes for fluid to flow through the domain. If a diffusion process takes the longest, then you will want to use this time. See the documentation for more information.

Looking at the automatic timestep value is a good indicator, as it calculates a series of characteristic times and makes the timestep 1/5th of the shortest. This is a very conservative estimate. The automatic timestep should behave very stable, but may take longer to get to a solution.

In practice, you should take as large a timestep as you can get away with. Small timesteps introduce a lot of relaxation and can help you deal with a bad initial guess, but once the solution starts converging, you should increase the timestep to as large a value as the solution will allow. Try increasing it by a factor of 5 or 10 and running for 10 iterations before increasing it again. This will get you to a solution the fastest. Eventually you will get a good feel for what you can get away with.

For most problems, you should arrive to a solution converged to 5e-5 MAX residual within 200 iterations. If your solution requires more timesteps, but convergence is OK, then increasing the timestep will help. If the residuals flatten out, or begin to bounce, you may have a problem. Contact technical support for some good advice.

Best regards, Robin

All times are GMT -4. The time now is 22:42.