[knuckles]

I've been using OpenFOAM to simulate combustion. I began by simulating with the combustion model inactive, and then turned it on. Around 20 time steps after activating combustion, this happened:

DILUPBiCG: Solving for H2O, Initial residual = 2.122748001606821e-05, Final residual = 4.609232998037878e+21, No Iterations 1001

This didn't cause a full failure of the code, but it did make the results extremely non-physical (for example giving a peak H2O mass fraction of 1e16). For the time step in question,

Courant Number mean: 0.001837135708655162 max: 0.2926824599786872 deltaT = 6.018954193305121e-07

...and the relevant settings in fvSolution are:

solvers
{
    Yi 
    { 
        solver          PBiCG; 
        preconditioner  DILU; 
        tolerance       1e-09;         
        relTol          0;     
    } 
}

Because I have combustion, the linear system that I'm trying to solve includes an explicit chemical source term, but this should be, at most, wdot_H2O/rho = 1000 [1/s]. For the time step above, this would mean that the term added to the linear system should be no more than 6e-4*rho. The mass fraction of H2O before the time step is in the 0 to 0.1 range.

At first I thought that my chemistry model had malfunctioned and given non-physically large reaction rates, but then if this were the case then I would also expect the initial residual of H2O to be extremely large. This suggests to me that the problem is numerical.

I'm wondering how to remedy this situation; specifically:

1. Why might this happen? Are there particular types of linear systems which are known to experience this kind of divergence when solved using DILUPBiCG?
2. What parameters should I be tweaking before re-running: the linear system solver? the stopping tolerances? the Courant number?

[mkraposhin]

Hi,

Try to use smoothSolver

Or BiCGStab , which is present in foam-extend

[knuckles]

Thanks mkraposhin, this was helpful. I think that the real problem turned out to be my inlet conditions, however: I have a turbulentInlet BC and the magnitude of the fluctuations was far too large (over 0.1). My simulation has not diverged in this way since I've reduced the fluctuations (to more like 0.01).