Comparing Energy of different Polymorphs with DMOL3 - k-point dilemma!

Hello,

I have a fundamental question regarding the energetics of polymophs. I want to predict the crystal structures of some simple organics systems. The idea is simple: to predict the low-energy structures with Polymorph module and to further optimize the selected ones with disperssion corrected DFT using DMOL3.

With Polymorph there is no problem, since the computations are performed in real space and the energy seems to be expressed in kcal/mol per asymmetric unit. But what with DMOL3? Imagine that you have 50 different crystals with cell size varying from polymorph to polymorph which are also built by different ammount of molecules (Z = 2, 4, 8). Is the energy expressed in kcal per unit cell?

What about the k-point sampling? Every structure is suggested to be calculated with different k-point mesh (as suggested by MS Visualizer). What we can do - use the fine criterion is the best what we can do? The energy will not converge variationally and the k-point sampling error error will not be constant and will differ from one structure to another. We cannot use any common value for the k-point mesh. Using only gamma sampling for a small real space with hydrogen bonding inside (expected to result in the rapid changes of electron density) definitely seems not to be enough. So what would you suggest? How could you compare the energetics? Is the free energy less affected by the k-point sampling error? Is it more appropriate to compare the free energy what about the unit)? Is it forbidden to compare the energetics? How to assort the polymorphs and hence which structure is believed to be the lowest one?

I would really appreciate your comments!

All the best,

Casper