Maybe, you can get advice on the following question.
I have a new crystal structure for intermetallic compound of the Y-Ni-Si system determined by XRPD down to R(I)=2.72%. The conventional unit cell contains 152 atoms, space group is I–43m. Y, Ni and Si atoms are completely ordered in this structure, except three crystallographic sites with atomic statistical mixtures (Ni,Si):
0.796(10)Ni + 0.204(10)Si in 8(c), 0.850(9)Ni + 0.150(9)Si in 8(c) and 0.693(16)Ni + 0.307(16)Si in 2(a).
i.e., the approximate ratio of Ni/Si is 0.8/0.2
I cannot use supercell in spite of limited computational resources (MPI, 28 processors) in this case.
After primcell=”true” there are 76 atoms in reduced cell, with 9 (Ni,Si) atoms in fact.
If these atoms are interpreted as Ni (species file Ni.xml) (because Ni is dominant in these statistical mixtures), the calculations do not convergence after ~100 SCF runs ((especially for change in total energy and neither for 2*2*2, nor for 8*8*8 k-point mesh).
What is the best probable solution for this?
1. Only add chgexs=”–24.122” to compensate the electron deficiency, with original Ni.xml species-file for these 9 atoms?
If so, than the total nuclear charge will be –1953 and total electronic charge will be 1928.878, with total excess charge -24.122. Is it good?
2. For these 9 atoms: create a new species-file ns.xml from the original Ni.xml by modified z, mass, r(mt), and occ (in the atomic state) according to the approximate ratio of Ni/Si 0.8/0.2, and without using the chgexs in the <groundstate>.
But i think that the modification of occ for electrons is not quite good for such high Ni-content in mixture.
3. add chgexs=”–24.122” and use ns.xml with modified z, mass, r(mt) according to the Ni/Si ratio, but with unmodified occ in the atomic state (remain as for Ni.xml)?