Some questions regarding maps/emc2

Hi Axel,

I had a couple of quick questions regarding maps and emc2.

  1. I’m doing a maps runs on the bcc form of MgSc. However, two problems arise. The first is that while the majority of the structures fall nicely within the convex hull, a handful of structures fall far below it in terms of energy. Looking at the vasp output shows the warning "hit a member that was already found in another star". For some of the structures I can "fix" the problem (return the energy to a more reasonable value) by moving the atomic positions by small amounts (either manually or via cellcvrt -jc or -ja). Do you have any intuition why this phenomena might occur, and is there a better way to fix it. I’ve also found that setting ISYM=0 can fix the problem, but I don’t know if this has other negative consequences.

  2. Related to the last question, even after doing the "atomic jiggling" I find that the checkrelax scores for my structures are not ideal. In the manual it says anything above a relax score of 0.1 should not be included, but roughly 1/3 of my MgSc bcc structures have checkrelax scores higher than 0.1 (most fall in the range 0.1-0.4). The highest scores belong to those structures which I "fixed" (as discussed above). I would just remove all these offending structures, but I worry because I feel it will remove much necessary information from the cluster expansion, especially when the offending structures only have ~3 atoms. Do you have any thoughts about the best way to proceed with this problem?

  3. Finally, relating to emc2: I understand almost all of the rationale behind the -mu0 and -mu1 tag values. However, I am a little confused what it means to have -mu0 or -mu1 be negative (when the -abs tag is NOT set). Could you please explain?

Thank you.

  1. This is really more a question about vasp! The issue is in vasp’s symmetry detection engine and turning off symmetry should do the trick (it will just make the calculations slower). You can also play with vasp’s symprec value.

  2. That could be due to the fact that some structures are mechanical unstable. We recently found a nice solution to this, see: https://dx.doi.org/10.1038/ncomms8559

  3. Negative mu would just linearly extrapolate the mapping from "scaled mu" to real mu that was obtained between the first 2 ground states.

Sorry for the delayed answer!

Thanks very much for the reply! In the interim since my last post, I’ve tested the robustrelax method, as well as several other fixes, and it did work pretty well.

Best,
Adam