[Sylvain]

Hi everyone
I’m currently working on benchmarking a pollutant dispersion setup for OpenFOAM. I use the isolated building benchmark used by Tominaga & Stathopoulos (1997 2018). To do so, I use “scalarTransportFoam” (the classic one and a homebrewed one for the turbulent dispersion). The velocity field is frozen and calculated before, using simpleFoam.

While trying to get better results, I realized that the solution was very sensitive to the time step (see figures attached). This was done with the classic scalarTransportFoam



I replicated the issue with the pitzDaily test case provided in the tutorials (here openfoam8). I just turned down the molecular diffusion to 1e-5m2/s. I did some snapshots of the concentration field at t=0.2s, when the pollutant starts to spread inside the recirculation zone.

Turns out when the time step is too low, the solution “freezes”. I tried to increase the writing precision for both the time and other variables. But it didn’t change anything. I also discovered that for those low courant number, the initial condition for T (for example set at 1e-5 instead of 0) has a great impact on the converged solution.

I did not expect to have such issues for low Courant number, quite the contrary. I thought that the lower is the time step, the more precise is the simulation. But it appears it is the contrary!!!

Is this normal? Has anyone ever encountered this problem? I would be interested in any explanation of this behaviour.

Thanks ahead for any insight about this.

Sylvain

[Sylvain]

So, it turns out that the final solution is also very sensitive to the initial condition for T. So you can get anything you want, depending on the time step and the initial condition.

But… because there is a but… I found out that if you lower down the tolerance parameter of the solver in fvSolution, things are getting much better, and are no longer depending on neither the time step nor the initial condition. More, the results are nicely fitting the benchmark.

The tolerance parameter was 1e-06 in the tutorial. I lowered it down to 1e-8. I guess that the relative high value of molecular diffusivity chosen in the tutorial allows to keep the value of 1e-6 without problem.

solvers
{
    T
    {
        solver          PBiCGStab;
        preconditioner  DILU;
        tolerance       1e-08;
        relTol          0;
    }
}

I hope it might be helpful to anyone who might encounter this issue.