too small time-step interFoam solver
i'm using interFoam solver working on a multiphase problem.
My target is to impose a Froude number up to 1-1.5.
I setup the adjustTimeStep as ON in the controlDict file in order to respect the CFL condition due to a Courant number of 1.0. This is very restrictive on the time-step, that results of order 1e-05 (or 1e-06) for Froude number = 0.57 .
Is anyone able to use interFoam with a larger time step (i.e. 1e-03) using a similar Froude number?
If yes (I hope!), is this a problem related to my numerical setup, or I have to impose unsteady "smooth" boundary conditions on velocity (ramp, cosine...ect...) ?
Thanks for your help,
thanks for the quick reply.
I know about CFL condition and Courant formulation. I agree with you about a roughly "a priori" estimation of time step, and I've tried to do it for my case. The result of my estimation agree more or less with the dt used by the solver. And this is good.
But I know that commercial solvers (as Fluent) are able to run CFD simulation up to Courant 5, 7, and in same cases 10 in order to increase time step. Of course this involves a reduction of solution accuracy, but my interest is to understand if OpenFOAM is able to run CFD simulation with Courant larger than 1.0, maybe after having modified some numerical parameters in fvSchemes and fvSolutions.
Hope this can be clearly explained. What is your idea about it?
In principle the CFL criterion is only necessary for explicit time-marching schemes. So I guess that if you set up your discretization schemes (in fvSchemes) such that you solve all equations fully implicit that you can increase your timestep.
I am pretty sure this is also how it is possible in e.g. Fluent to have convergence with CFL>1, because with explicit schemes a CFL>1 will not only give you inaccurate answers, it will give you a diverging result.
But if you don't care too much about the accuracy of your results I think it will be easier and quicker to just lower the number of grid cells that you have.
OpenFOAM used implicit schemes. Otherwise your time step would be even smaller as it is now.
One reason why to need CFL<1 is the piso algorithms used in the solver. If you include internal iterations with relaxation in the solver, you can run at higher CFL numbers (see rhoPimpleFoam). Or, as a faster method, use time discretization schemes with local time stepping (CoEuler …). This method can be used if you are interested in the steady state solution
|All times are GMT -4. The time now is 23:45.|