Geometry Optimization of Crystals

[handsomecat]

I'm stuck with a problem of finding the minimum energy of an inorganic crystal. Crystal is of monoclinic structure with 18 parameters to specify its internal coordinates and 2 atomic species. Am using a classical interatomic potential found in literature.

The minimization involves both the 18 parameters and the cell vectors.

Have tried conjugate gradient method but algorithm never gives me a satisfactory minimum - the L2-norm of the forces is far from zero.

The best results I got was with the Nelder-Mead simplex method. Even then, the L2-norm of the forces at the termination of the algorithm was not close to zero.

Any pointers? I was thinking of trying Damped MD next ...

 

[Rajini]

Try using Gaussian software...use proper method and basis set...all u need is the cartesian coordinates (not the internal coordinates which u get directly from x-ray crystallographic data)

 

[Modey3]

Hello handsomecat,

Are you sure the potentials you are using are applicable to your specific system. Interatomic pair potentials tend to be system and model specific. For instance, the transferability of a Fe-C potential for diffusion modeling isn't transferable to modeling to structure of cementite (Fe3C). The main reason being that the potentials aren't just pair dependent, but also depend on the chemical environment of that pair. Maybe based on these potentials the L2 structure isn't stable. Use a ab-initio code to avoid the pitfalls of pair-potentials.

Modey3

 

[handsomecat]

Hi Modey3, thanks for your comments.

I have already performed ab-initio calculations and have given up on using MD for that purpose :) I am certainly aware of the shortcomings of using MD interatomic potentials (something that has given me much grief in the past 4-5 years!)

However, as a matter of interest, I never stopped wondering about how energy minimization of a unit cell wrt cell vectors and internal coordinates is done simultaneously. Nevertheless, I've found an article and all readers can refer to this article:

Bernard G. Prommer et al. Journal of Computational Physics 131, 233- 240.