# Hexagonal lattice volume EOS fit

I have a little problem concerning hexagonal volume fitting using VASP. I have followed these steps:

1. Relax the structure from a given volume using ISIF =4

2. Copy CONTCAR to POSCAR and relax it again

3. Run with ISMEAR = -5 and without relaxation

4. Repeat for different volumes (+10% +5% 0% -5% -10%)

5. Use your script for EOS fitting.

It worked successfully with cubic systems. But for a hexagonal system (e.g. Carbon), I got the figure below. I don't know why I always get wrong results for the first volume point. Even with 7 volume points the first point always wrong.

• What point in this figure is the first point? When you say its "wrong", what is your desired margin of error?
– Tyberius
May 23 '20 at 19:24
• +1. Seems like a good question. I agree with Tyberius that the problem could be a little bit better described. As someone who doesn't work as much in this field, I'm not sure what the "correct" behavior should look like. Should it be monotonic (meaning the 3 points with smallest volume, are "wrong" here), or should it be somewhat flat (meaning that the point with largest volume here, is wrong)? May 23 '20 at 20:04
• I think the problem is that the variations are to high. Normally the cell variations are around 5-7%. Higher than that can be considered as deformations. (I just edited the firs comment).
– Camps
May 23 '20 at 21:21
• How are you varying the volume? And by hexagonal carbon you mean graphite? Does the first point also have a low energy when you increase the volume slightly (e.g. make the first point 36 Å$^3$)? Finally, does still occur with a different hexagonal crystal (e.g. metallic Ti, $\beta$-quartz)? May 23 '20 at 23:58
• The expected behavior should be parabolic for energy vs volume, which is what the OP is trying to say. Although I don't know the goal of this. Anyway, I would use ISIF = 2 to obtain energies. Another thing you could try is change a/b lattice constant sequentially instead of the volume if you can - i have had better success with it. If nothing works, maybe check if any of pymatgen's functions would help.
– gogo
May 24 '20 at 1:58

Normally this should follow a nice equation of state (Rose, Murnigham etc.). I'd say the curve looks okay: $$35 \mathrm{A}^3$$ seems to be the questionable outlier.