[meangreen]

I really need some help with a problem I have run into. I am running a eulerian multiphase simulation using the k-epsilon turbulence model and an unsteady (implicit) solver. I am using fluent and i am including dense phase stresses for the solids phase. I am modeling the interaction of a jet of gas a bed of particles.
Is it possible for a time step to become too small and cause instability? I have found as I decrease the time step, my solution continues to change and I can not figure out what the problem is. I have used different grid sizes and I even see the same problem running with second order implicit discritization.

my time step has gone down to 1x10^-7s and I am still seeing time step sensitivity, and I dont want to go below this because it doesn't really make any physical sense nor does it make sense from a computational cost point of view.

Has anyone found that with a time step too small there are stability problems? If so please explain.

Are there any residuals (or balances) I can look at to monitor the stability or maybe to find what the root cause of this time step problem may be?

ANY INPUT WOULD BE HELPFUL.

Thanks,
Casey

here is a link to some plots that show the change in the scour hole depth as the time step in changed.
http://picasaweb.google.com/Casey.La...eat=directlink

[hadesmajesty]

I think a possilble reason is the use of second order implicit method. You may try to use a higher order explicit method and keep the CFL number small enough.

[Ahmed]

Let us say that the solver has to maintain a Courant number Less or equal 1 throughout the whole solution domain, look for the smallest cell size in your domain and calculate the time step accordingly
Good Luck

[meangreen]

Why would the implicitness be the problem? I thought stability problems were avoided by using implicit schemes. Also, when I calculate the Courant number it is very low.

[Rich]

Remember that, as you march forward into the solution, influences travel along the characteristics of the flowfield. From two adjacent points or two adjacent cells, characteristic lines will propagate into the future and intersect each other after a certain length of time. It's at that point in space and time that we want to make our next calculation, for that new point's domain of dependence includes exactly the two preceding points. Instead, we make our timestep slightly smaller. And that means that, traveling backwards along the characteristics, our new point's domain of dependence now falls slightly *inside* of the two preceding points. And that means that the solution at our new point will have some error in it--because we're now calculating based on values from two prededing points, but the domain of dependence for this point no longer quite includes those points. If we continue making the timestep smaller and smaller, the error grows. So for greatest accuracy, it's important to use the largest timestep possible, without exceeding the timestep dictated by the characteristics. For your implicit solution, you'll know that you've exceeded that point when the solution thrashes back and forth between two different values.

[meangreen]

Rich,

thanks for the insight. Would you be willing to open an email dialogue with me? I have a few questions that may be easier to answer with the quick response of email.
Do you have the same error if you start out with a smaller time step? SO I am seeing this time step dependence when I start the simulation with time steps of DT1Eneg6, DT1Eneg7, DT1Eneg7.
below is a link to an album of plots that show the time dependence.

http://picasaweb.google.com/Casey.La...eat=directlink

THANKS SO MUCH FOR YOUR HELP SO FAR!!!!

[UGA]

I meet the same question when using e-e model. there is a oscillation for fraction along the vertical boundary, especially if the gradient of fraction/velocity is large. I still do not know how to address it.