[wildabyss]

Using pressure-based solver (coupled) and setting the under-relaxation factors to:
Momentum: 0.5
Pressure: 0.1

I get Divergence Detected in AMG Solver even while all residues were steadily decreasing (at around 20 to 60 iterations). This happens regardless of whether I use first-order or second-order discretizations.

Is this a sign that mesh quality is bad?

[LuckyTran]

This is all assuming that the simulation is setup properly. Double-check to make sure all your boundary conditions and solver options are appropriately selected.

Divergence in the AMG solver is similar to any other divergence problem. It is difficult to say if the mesh is bad or not. Divergence can easily occur even when residuals are dropping; when certain flow quantities are tending towards very huge magnitudes, it can cause the residuals to drop (because of relative magnitudes) and give the false impression that the solution is converging (based on residuals) when in fact it is diverging.

Since the divergence occurred after a number of iterations, it is most likely a flow initialization problem. You initial guess is likely not "good" enough. This is more likely to be the case, since even a bad mesh will produce converged results (although inaccurate) if the initial guess is good enough.

Try to simplify the problem to obtain a loose solution to give it a better initial guess. Try using the segregated solver for a number of iterations first. Try using simpler turbulence models and boundary conditions. Once you have a decent solution, then switch over to the options that you would like to use.

More importantly, it will be a bit of debugging effort, but try to figure out what is actually diverging in the solver by examining solution values.