question on the method for smoothed DOS data

[luke419]

I performed DOS analysis on my material.
ISMEAR=-5 was used without explicit assignment for SIGMA (default value of SIGMA should be used).
The geometric optimzation was done with k-mesh of 2X3X3 and I've performed DOS analysis with k-mesh of 4X6X6 using the same positional data without further geometric optimzation for 4X6X6.
(That of 6X9X9 faild to be peformed in my parllel computers system.)

I've got following DOS data by substracting Fermi energy (-4.1997) to orignal data.

-------------------------
-14.0773 0
-13.8823 0
-13.6873 0
-13.4923 0
-13.2973 0
-13.1033 0
-12.9083 0
-12.7133 0
-12.5183 0
-12.3233 0
-12.1283 0
-11.9343 0
-11.7393 0
-11.5443 0
-11.3493 0
-11.1543 0
-10.9593 28.52
-10.7653 0
-10.5703 0
-10.3753 0
-10.1803 0
-9.9853 0
-9.7903 0
-9.5953 0
-9.4013 0
-9.2063 0
-9.0113 0
-8.8163 0
-8.6213 0
-8.4263 0
-8.2323 0
-8.0373 13.27
-7.8423 42.98
-7.6473 1.473
-7.4523 96.4
-7.2573 0
-7.0633 0
-6.8683 0
-6.6733 74.06
-6.4783 0
-6.2833 0
-6.0883 0
-5.8943 0
-5.6993 0
-5.5043 33.13
-5.3093 0
-5.1143 0
-4.9193 19.44
-4.7253 16.7
-4.5303 21.92
-4.3353 48.19
-4.1403 0
-3.9453 255.7
-3.7503 0
-3.5563 0
-3.3613 69.72
-3.1663 15.56
-2.9713 9.668
-2.7763 12.79
-2.5813 9.184
-2.3863 7.965
-2.1923 12.52
-1.9973 29.33
-1.8023 31.95
-1.6073 50.31
-1.4123 37.69
-1.2173 15.84
-1.0233 40.26
-0.8283 21.99
-0.6333 9.754
-0.4383 5.834
-0.2433 26.86
-0.0483 0
0.1457 0
0.3407 0
0.5357 0
0.7307 0
0.9257 0
1.1207 0
1.3147 0
1.5097 0
1.7047 0
1.8997 0
2.0947 0
2.2897 0
2.4837 0
2.6787 0
2.8737 0
3.0687 0
3.2637 0
3.4587 0
3.6527 0
3.8477 0
4.0427 0
4.2377 0
4.4327 29.76
4.6277 63.72
4.8227 67.97
5.0167 32.04
5.2117 0
5.4067 0
5.6017 0
5.7967 0
5.9917 0
6.1857 9.608
6.3807 9.913
6.5757 21.09
6.7707 24.34
6.9657 4.667
7.1607 12.58
7.3547 15.87
7.5497 25.45
7.7447 19.91
7.9397 6.934
8.1347 23.77
8.3297 12.85
8.5237 25.32
8.7187 20.91
8.9137 8.439
9.1087 14.63
------------------------------

My questions are as follows.

1. If I perform post treatment of smoothing for data, DOS expand to beyond 0 (Efermi) continuously and the material, consequently, apears to be "conductor" rather than semiconductor or insulator though original data doesn't show DOS value right above Efermi (see the data).

Then, what kind of material should this be regarded as, conductor or imsulator (or semiconductor)?


2. In the manual, it is written that we must use ISMEAR=-5 for "insulator or conductor", but if we use gamma or 2X2X2 k-point mesh (or very big system), we may use ISMEAR=0.

Should the use of ISMEAR=1 for insulator or conductor ((or the use of ISMEAR=0 for dense k-point mesh for small system) result in siginficant error?
Will ISMEAR = -5 be even O.K. for conductor though we don't have to use it?
With what criterion should we determine method for ISMEAR commonly?


3. What is the default value of SIGMA and what does it mean in detail?
What is the common value for ISMEAR=-5?
How should we determine this value for ISMEAR=0,1,-5 respectively?

4. There is very high DOS (255.7) value at -3.9453, which is pecular compared to geoneral cases.
How should we interpret this result?
Should it have sigificant meaing with the respect to characteristics of material?


5. The shape of DOS distribution is too sharp and it seems to be somewhat sparse.
In spite of this, should it be O.K.?
If not, what should we do for it generally?
Should we increase NEDOS?

(The increase of K-points may be helpful in this respect but it seems to be not in this case, I think. Furthermore denser k mesh than 4X6X6 is difficult in my system.)


I hope you to answer as in detail as possible, please.
With best regards,


Sincerely
Luke

[admin]

a smearing of 5 eV is rather unphysical, please note that 1eV corresponds to an energy of about 10000 (tenthousand) Kelvin

[Danny]

1) If you do a post calculation smoothing of your data, you have to be carefull. The system is what it is, som this type of small modyfications around Ef should be disregarded since you introduce them yourself through your smoothing algo (use a dense NEDOS, such that you can smooth and the only influenced points abouve Ef are still close to it )

2) If you have to few kpoints (eg Gamma only gives you one) you can not construct tetrahedra(which need at least 4 points (one for each vertex), hence the description in the VASP manual.

3) Default Sigma=0.1 (cf VASP manual), it is a "thermal" smearing of your energy levels. However if you use tetrahedron integration of your k-space this parameter is not used, only for Gaussian and MP smearing integrations.

4) These are integrated values, If you use a denser mesh(NEDOS) you will see what happens to this spike.

5) NEDOS will only increase your number of gridpoints in your DOS-output. The smoothness depends on the density of your kpoint-mesh(denser=smoother) and the algorithm used in ISMEAR (-5: will be very spiky, however a Gaussian smearing of MP smearing with will smear the levels somewhat also smoothening the DOS. Mind that in theory these last two options should go to the Tetrahedron results for Sigma--> 0.

try to use a small smearing (eg SIGMA=0.02) and see what it does to the DOS.
Also set your NEDOS to a larger value (eg to have an energy resolution of 5meV instead of 195mev as you have now.)
=> set following parameters in INCAR
EMIN = -14.0 !minimum energy in your dos is now -14 eV
EMAX = 10.0 !maximum energy in your dos is now +10 eV
NEDOS = 4800 ! (10+14eV)/0.005eV=4800 5meV resolution

Note that with ISMEAR=-5 this will look very spiky, but don't wory, either smooth the data with a smoothing algo afterward or use SIGMA and Gaussian of MP smearing and you will see things clear up, while giving you a good DOS.

Cheers
Danny