smaller timestep leads not to converge
 

Dear all,

I got a transient run with constant timestep 0.001s, smoothly done by CFX (ver 13.0).
Other I input BCs:
Laminar, isothermal.
inlet flowrate around 0.003 kg/s; outlet is about 10000Pa
For first two iteration , the courant number is about 200s, then reduced to the level of 50s.(I was noticed the CFX is fully implicit so greater courant number is not issue once converged. )
I monitor the outflow and inlet pressure, everything seems alright for this run.

Then I tried smaller timestep 0.0001s but the it couldn't get convergence for the first iteration.
The solver manager shows the linear solution failed in equations of U-Mom, V-Mom, and W-Mom; then fatal error, Floating point exception: Overflow.
I haven't checked the mesh yet.
I ticked the CFX-Pre->Solver control->Basic Setting->Timestep Initialisation-> Upper courant Number, and gave value to 10.
The first iteration just flow through and running and running.

I am wondering if this is right. Or there's another way around, I could use 0.0001s without divergence.

THANKS

hey,

really not sure about that but i think it could be a problem if the time step isn´t sufficient for letting the flow pass the very first volume with given velocity.

neewbie

You should set the outlet to 0Pa pressure and use the reference pressure to give the correct absolute pressure. Using a large outlet pressure causes large round-off errors which will get worse when you decrease the time step.

Also running double precision numerics might help.

I am doing FSI problem, so expect having the 'real' pressure.
I almost have the run finished I will check any diffrerence between two cases(0.001s and 0.0001s)

Thanks!

I am no expert on FSI but I am sure it is smart enough to be able to handle a reference pressure. This is really basic stuff. And very important for exactly the reason you have found.

I I were you, I would use a steady state first.
I would define the reference pressure and I would defined the outlet pressure as a relative pressure, as a Glenn said.
Besides, I would start as a Steady state with and Auto time scale = 1 (conservative), and let the program to make some iteration, as a result you will get an "accurate" time step.
After you del with steady jump to transient....
Good luck!!