I am interested in the Fe-Si system and having problems in calculating cluster expansion: the cross-validation score is about 2 eV after calculating ~200 clusters and has no tendency to decrease. I use VASP, PAWPBE potentials, magnetic iron with or without setting the initial magnetic moment to 2.2 for Fe (multiplied by the number of Fe atoms) and 0 for Si, shape/volume and ion relaxation, the initial lattice parameter of 2.8 Angs., and the lattice vectors corresponding to the bcc lattice structure.
I should also report about problems with several clusters, e.g. Fe2-Si and Fe-Si2: the energies of these clusters have been calculated but the ‘vasp.out.relax’ contains huge number of messages: `hit a member that was already found in another star’.
It looks like your fit is being "contaminated" by large erroneous numbers.
You can spot them graphically by running mapsrep
If only a few points show up, also try
ls */error
to see if many structure are being ignored due to errors.
BTW, the vasp error you report can often be eliminated by reducing SYMPREC to 1e-5 or so.
Another possibility is that some structures relax too much and can no longer be considered on the same lattice. You can run
checkrelax
and do touch nn/error for any relaxation number above 0.08 or so (nn is the structure number).
There also exists a cleaner solution to the overrelaxation problem: https://dx.doi.org/10.1038/ncomms8559
that you can also try if needed.
What could be happening is that the B2 phase is often very wide (it allows for a lot of point defects): so the DO_3 phase is "surrounded" by the B2 phase even though the "center" of B2 should be at 1/2.
I would suggest running emc2 on a line of constant T and look for any jumps in composition or peak in heat capacity/suseptibility. This would mark phase transitions that phb ignored (because it was not asked to follow them ) .
It is possible that your DO3 phase is "too stable" because vibrational effects were neglected. But that is higher-order effect - not your first priority.
Thank again for your help with cluster expansion for Fe-Si. However, I am still struggling with reproducing Fe-rich part of the Fi-Si phase diagram. What is really confusing to me is that the experimental phase diagram contains region A2+B2 to the left of the DO3 phase. [attachment deleted] I calculated the cluster expansion (figure attached), which has four ground states in the Fe-rich region; and performed Monte Carlo (phb) calculations of the phase boundary between phases 0 (Fe) and 1 (Fe3-Si), figure attached, which I suppose correspond to A2 and DO3 phases, respectively. [attachment deleted] The results do not indicate presence of B2 phase between phases 0 and 1. The question is: is it possible at all to analyze A2-B2 equilibrium in Fe-Si.
I have one more question: there is a reference to the IMR-CVM code
"[19] M. H. Sluiter. Imr-cvm code, 2000. https://www-lab.imr.tohoku.ac.jp/"
in your ATAT manual, which I want to try. Unfortunately, the website does not exist. Do you happen to know about this (or other) CVM codes available?
Hy,
You can spot them graphically by running mapsrep
If only a few points show up, also try
ls */error
to see if many structure are being ignored due to errors.