[rae]

I have a convective flow driven by the buoyancy of air, which is heated by a local energy source (approx. 1.1MW/m^3) whose size is defined by an additional variable within a much larger yet single domain.
The minimum temperature within the domain and opening boundaries is 298K. In my transient analysis the temperature plummets in a nonsensical manner to as low as 60K within a small region where inflow (about 4 m/s) is significant to the energy source volume. The attached image shows domain detail (bottom sloping line is adiabatic boundary and vertical line marks the energy source volume to its right)
I suspect the density drop (to about 0.2kg/m^3) due to temperature within the energy source is responsible for a similar density drop of adjoining elements outside the source and subsequently silly temperatures to some elements with significant inflow to the energy source???
I use Sutherlands formula for dynamic viscosity and thermal conductivity. I have not used tables and do not see where/how to clip the temperature to 298K where it should be and do not know if that is the best approach.
Alternatively perhaps the problem is a parallel partition problem ..., or then again, I have seen the tutorial for a steady state problem where gravity was introduced incrementally which seems like a great idea but I see no convenient way to do that during the iterations of each transient timestep????
Help out there would be most welcome. Thanks.

[ghorrocks]

It sounds like you better do some basic bug finding:
* Run it serial and parallel.
* Run it single precision, double precision.
* Run it to a tighter convergence, and a looser convergence.
* Run on a finer mesh and coarser mesh.
* Run it using constant properties versus Sutherland's
* Turn any other tricky physics on and off

Do any of these things affect it

[rae]

Thanks ghorrocks.
Further info: SAS model with B-CDS advection scheme being used and have diverge issue too -- last time the omega equation jumped to Rate 99.99 the step before failure on hydrodynamic equations.
Progress to date:
(1) Still running in serial mode, to one timestep beyond previous point of solver divergence so far, and temperatures still go very low.
(2) Running in double since before.
(3) Equation rms residuals bounce around 3E-4 to 2E-5 in 3 or 4 iterations, slowing the time step by factor 10 changes negligibly residuals, solver divergence or low temperature excursion.
(4) Expect smaller mesh will reduce but not remove temperature drop so still looking for other improvements.
(5) Change from Sutherland's to Material 'Air Ideal Gas' and 'Air @ 25C' (i.e. Bussinesq T dependence) hasn't changed issues.
(6) Running Expert: max continuity loops = 2, since before as found to help earlier (curiously RMS > max residual for 2nd P-mass loop occasionally); tef numerics option = 1, has allowed simulation to progress without solver divergence; and highres energy = 3, also allows simulation to progress. Still gets too cold for my liking.