Strategy for using hiphive for disordered systems

Hi, I am trying to use hiphive for geting force constants of a disordered system out of a rutile prototype structure.
The ground state structure of this system (found via cluster expansion) has 24 atoms (e.g. 4 formula units of the rutile primtive cell). In this case, should the prototype structure for the cluster space be the 24-atoms cell or the rutile prototype structure?

I have tried to use the 24-atom cell as the prototype for ClusterSpace. With a cut off radius of 5,4,3 for two/three/four bodies, there are already about 13,000 parameters in the model, which seems to make fitting the force constant potnetial to be a bit slow already?

Thanks in advance.

If you want to make an FCP for the disordered structure then you should use the discorded structure as prototype (with relaxed ions such that forces are close to zero for the prototype structure).

With a cut off radius of 5,4,3 for two/three/four bodies, there are already about 13,000 parameters in the model,

What is your end goal with obtaining the force-constants? e.g. if you only want harmonic phonons there is no need to include 3rd and 4th order terms.

Yes fitting FCPs is quite slow when including higher orders for low symmetry structures. But ideally you only need to fit the model once. (Note if the StructureContainer takes a long time to build you can save it to file and then read it from file when needed)

Thanks for the reply! The end goal to estimate the lattice thermal conductivity, probably using phono3py.

e.g. if you only want harmonic phonons there is no need to include 3rd and 4th order terms.

Just to doulbe check - do you mean that I can fit using pairs, triplets?

e.g.

cutoffs = [5.0, 4.0]
cs = ClusterSpace(prim, cutoffs)

Yes if you’re going to use phono3py, then there is no need to include fourth-order terms so a cutoff of [5.0, 4.0] would be enough.

1 Like