Temperature drop after porous zone

MN

A system I'm modeling has a catalyzed highly exothermic reaction inside a tube. The catalyst activity/kinetics and porosity are known/estimated, with the porosity around 0.3-0.4. thus, flow entering the system is heated, for example from 400K to 1000K. Because of the flow rates and porosity (with calculated values for the viscous resistance), there is a significant pressure drop up to 10atm across the catalyst bed.

Modeling the reaction in the bed is not a problem, but I'm finding that unless I drastically underrelax the energy equation, FLUENT predicts drastic temperature drops after the porous zone, or even within the porous zone after all reactants are consumed; so while the temeperature will get to 1000K at some point in the bed, the temperature will fall back to around 600-700K. This seems to happen with the compressible ideal gas law; it doesn't seem to happen with the incompressible law for the component mixture, but again with the high pressures we are seeing, incompressible isn't a valid assumption for our case.

Again, I can cut the energy equation under-relaxation to around 0.7 and the temperature will remain where it ought to be in the bed, but the residual for energy is quite high compared with other variables, and some of the post-bed sections remain at rather high temperatures after the bed (since I'm patching higher temperatures to 'spark' the reaction). I would assume that using this relaxation parameter would eventually get the solution to converge correctly, but the time it would take is excessive. I have played with sneaking the relaxation parameter up towards 1 slowly, but there seems to be cases where setting it just too high causes the solution to collapse to a cold-flow or the cold-flow-after-reaction case.

This is using the segerated solver. I've tried using the coupled solver, but have not had any success getting any sort of appreciable convergence in that (the rate with solution iteration to move the BC into the main portion of the bed is extremely and unusably slow.)

I am using a 3ddp solver, since I have read that this is better for any mass-heat coupled problem.

Any suggestions on ways to avoid the cold flow effect that I'm seeing here, or other ways to coax the solver to avoid it?