search

erdc / air-water-vv

Verification and validation tests for computational models of air/water flow
MIT License
5 stars 14 forks source link

The CFL condition and setting runCFL #15

Closed cekees closed 9 years ago

cekees commented 9 years ago

@ezve and @adimako found that "During simulations with more refined meshes, it was observed that in some cases, a Courant number restriction of 0.33 resulted in very low time-steps that excessively increased the computational time. We would like more information about the stability (CFL or VSL) restrictions of the numerical method"

cekees commented 9 years ago

We're solving the multiphase flow equations sequentially 1) Navier-Stokes 2) VOF 3) Level Set 4) Redistance Level Set 5) Correct Mass

While each step is theoretically unconditionally stable and uses and implicit time discretization, the splitting in the steps introduces "splitting error" that can lead to poor results. In practice we find that running with a CFL less than one is necessary for accuracy.

That said, I found that the time step is severely restricted in the 2D Ubbink problem after the wave hits the obstacle and shoots over the top of it. I'm going to look at the velocities more closely and see if they are actually generated by high air-phase velocities. We may need a better turbulence model for the air phase.

cekees commented 9 years ago

After doing a lot of testing, I think runCFL=0.9 seems to be a good choice for most of the problems.