Conventional vs. Primitive unit cells give different results

[wmaddox]

Hello,

I have successfully calculated optimal lattice parameters for (hypothetical) FCC germanium. However, my results when using the conventional unit cell are markedly different from those of a primitive unit cell (difference = 0.73 Bohr).

acell1 = 3 * 8.1664499280 Bohr (conventional)
acell2 = 3 * 8.8987188913 Bohr (primitive)

I checked this behavior with diamond cubic germanium and found the values to be reasonable, differing only by 3.4e-5 Bohr:

acell1 = 3 * 10.921623101 Bohr (conventional)
acell2 = 3 * 10.921589483 Bohr (primitive)

For these calculations I used the JTH PAW PBE pseudo found at http://www.abinit.org/downloads/PAW2/JTH-TABLE/index.html
I also observed the same behavior with the FHI PBE pseudo.

Please find attached the input scripts. I would benefit greatly from any advice or tips you may have.

Thanks,

- Will Maddox

[sponce]

Dear Will,

Could you also provide us with both output file (they should be of small size)?

The acell might be different at the end of the output file but you also need to take into account the rprim that you associate with the acell (the acell by itself it meaningless).

If you want to compare comparable quantities you should compare the rprimd variable (acell*rprim). If they are different then there is a problem somewhere in the input file.

Regards,

Samuel Poncé.

[wmaddox]

Dear Samuel,

I initially tried to add the output files but they were too large to upload (~450 kB).

Since the rprims weren't modified by abinit during the code during relaxation, a direct comparison of acell1 to acell2 is acceptable. I should have mentioned that in my original post.
Nonetheless, here are the rprims from the end of the output files:
rprim1
1.0000000000E+00 0.0000000000E+00 0.0000000000E+00
0.0000000000E+00 1.0000000000E+00 0.0000000000E+00
0.0000000000E+00 0.0000000000E+00 1.0000000000E+00
rprim2
0.0000000000E+00 5.0000000000E-01 5.0000000000E-01
5.0000000000E-01 0.0000000000E+00 5.0000000000E-01
5.0000000000E-01 5.0000000000E-01 0.0000000000E+00

Will

[wmaddox]

UPDATE:

I have investigated this further with an element that is known to have FCC crystal structure. I chose copper. The behavior was the same!?!?
Then I realized that copper is a metal, NOT a semiconductor like dc-Ge. Duh!

So I treat fcc-Ge as a metal by adding the line:
occopt 4
and commenting out,
# diemac 100.0
in my "Ge_fcc.in" script (see original post).

The results are MUCH better (difference = 2.0e-3 Bohr):

acell1 = 3 * 8.1003878771 Bohr (conventional)
acell2 = 3 * 8.1023139570 Bohr (primitive)

The error is still a bit high, so I will investigate it further. I will post again when I have determined why (unless someone else solves it before I do).

-Will

[Fm3m]

Are you generating the identical kmeshes, i.e. are the kpoints in the first (primitive) BZ in the same points? If not, perhaps your difference is one of numerical precision and convergence.