Density of states in low-symmetry large supercell

[tzuchen_liu2024]

Dear all,

We are working on the density of states for a series of low-symmetry large supercells (> 60 atoms in a primitive cell).

INCAR setting:
ISTART = 1
ICHARG = 11
ISPIN = 2
ENCUT = 520
LORBIT = 11
ISMEAR = -5
LMAXMIN = 4 (d-elements included)
EMAX = -12
EMIN = 18
NEDOS = AAA (from 301 to 30000)

All structures have well-converged relaxation and static run (SCF), before the non-SCF run for DOS.
We did the convergence test on the NEDOS value from 301 to 30000 and found these complicated systems DOS converged to very spiky plots with wiggles everywhere.
We tried the same parameter on simpler structures and found DOS plots to look normal and agree with the literature.
While it is possible we got this due to the nature of the low-symmetry supercell, we are not confident that such DOS are physically reasonable.

Also, when we did a quick check on integrating the reported DOS, the resulting values were far from those reported integrated density of states in DOSCAR.
(We tried a. the formula given in VASP wiki b. trapezoidal c. Simpson integral, they all result in similar numbers that are very far from reported IDOS )

While the VASP wiki gives the DOS formula: DOS is determined as the difference of the integrated DOS between two pins, we manually calculated the DOS value using this given formula and got some numbers totally different from the reported DOS value in DOSCAR.
(In simpler structures, the calculated numbers by using the given formula are still not exactly the same as reported in DOSCAR, but much less different)

We are curious if the reported DOS is really decided by the formula listed on the VASP wiki.
We appreciate all the help and related information.

Thanks!
Tzuchen

[martin.schlipf]

NEDOS is not a convergence parameter, it will just add more points to the mesh. The main convergence parameter for the DOS is the number of k points used.

The equation given for the DOSCAR does not apply for the tetrahedron method (ISMEAR = -5). While ISMEAR = -5 generally produces very nice DOS because there is no smearing, it may also have more spikes compared to the smearing methods (ISMEAR > -2).

Could you be more specific about the differences that you see? We cannot reproduce the issue without the input and output files.