Does SCF calculation have to be done before calculating DOS?

Question

The standard procedure to calculate DOS or band structure is:

1. Structure optimization,
2. SCF calculation,
3. non-SCF calculation.

According to the vasp wiki, it is because:

Mind: Since the charge density written to the file CHGCAR is not the self-consistent charge density for the positions on the CONTCAR file, do not perform a bandstructure calculation (ICHARG=11) directly after a dynamic simulation (IBRION=0).

But according to the result I got from the structure optimization:

CHGCAR:

unknown system                          
   1.00000000000000     
    13.950000    0.000000    0.000000
     0.000000   13.950000    0.000000
     0.000000    0.000000    8.928000
   O    Ti
   108    54
Direct
  0.268106  0.065717  0.167082
  0.065369  0.268163  0.166767
  0.101149  0.101162  0.000275
  0.231588  0.232049  0.000228

CONTCAR:

O Ti                                   
   1.00000000000000     
    13.9500000000000011    0.0000000000000000    0.0000000000000000
     0.0000000000000000   13.9500000000000011    0.0000000000000000
     0.0000000000000000    0.0000000000000000    8.9280000000000008
   O    Ti
   108    54
Direct
  0.2681055481518182  0.0657169184815687  0.1670824834593189
  0.0653691262260372  0.2681632907730134  0.1667673218695771
  0.1011490079420389  0.1011622291646118  0.0002747383173194
  0.2315880276338604  0.2320489480787421  0.0002283670824430

POSCAR:

O Ti 
 1.0000000000000000
    13.9500000000000011    0.0000000000000000    0.0000000000000000
     0.0000000000000000   13.9500000000000011    0.0000000000000000
     0.0000000000000000    0.0000000000000000    8.9280000000000008
 O   Ti 
 108  54
Cartesian
  3.7365895255885855  0.9134104744114493  1.4879999999999789
  0.9134104744114493  3.7365895255885855  1.4879999999999789
  1.4115895255885178  1.4115895255885178  0.0000000000000000
  3.2384104744114151  3.2384104744114151  0.0000000000000000

Apparently, the CHGCAR file uses the same atom coordinate as CONTCAR.

Does this mean that the VASP wiki is wrong and I don't have to do SCF calculation before calculating DOS or do I understand it wrong?

Answer 1

The community wiki by Tyberius is a bit misleading here, you are not doing molecular dynamics calculations (IBRION=0) but likely a geometry optimization (IBRION=1,2,3). In the case of geometry optimization, the CHGCAR is not the predicted charge density, but is instead the charge density of the last completed step.

For this reason, you should find no difference between doing an SCF calculation and just using the geometry optimization calculation. However, I do advise still doing the SCF calculation since this is a good opportunity to refine the kpoint grid further (and check the forces at this stage to ensure they are still low). You might do something like the following (numbers picked arbitrarily).

• (3, 3, 3) for Geometry
• (6, 6, 6) for Single Point
• (24, 24, 24) for NSCF DOS

Answer 2

In the passage just above what you quoted, it says:

For dynamic simulation (IBRION=0), the charge density on the file is the predicted charge density for the next step: i.e. it is compatible with the CONTCAR file, but incompatible with the last positions in the OUTCAR file.

This means while the density in the CHGCAR file is a density for the position given in the CONTCAR, it is only a predicted charge density, essentially the initial guess for the SCF at that position. This means you would still have to get the self-consistent density for that position.

Note the quote you mentioned only applies to IBRION=0, i.e. a molecular dynamics simulation. For your geometry optimization, the remainder of the previous paragraph confirms that the CHGCAR should be fine for determining a band structure:

For static calculations and relaxations (IBRION=-1,1,2) the written charge density is the self-consistent charge density for the last step and might be used e.g. for accurate band-structure calculations.