Question on total energy of free atom.

[lahaye]

Hi everyone,

For calculating the binding energy of a Pt atom to a much larger compound,
I must also calculate the energy of the free Pt atom.

I have checked that the energy difference between LREAL=True and
LREAL=False is very small (i.e. less than 1 meV, whereas the expected
binding energy is about 2 eV).

Both, in the VASP guide and In the Handson1 workshop sheets for the
Oxygen atom, it suggests to use ISMEAR=0 and a reasonable smearing
width SIGMA of 0.1 or 0.2 eV.

To my surprise, the free-atom energy E0(for sigma to zero) varies
linearly with the smearing width; almost 0.1 eV difference between
SIGMA=0.1 and 0.2. This is what I observe:

Code: Select all

 SIGMA=0.1   E0=-0.17
 SIGMA=0.2   E0=-0.24
 SIGMA=0.3   E0=-0.31
 SIGMA=0.4   E0=-0.38

Is this something to worry about? Should I use a different smearing
method instead to bypass this variation with SIGMA?

Thanks,
Rob.

[admin]

free atoms levels are SHARP levels, the smearing has to be very low (SIGMA=0.01).
(p13 of handsonI.pdf)

[lahaye]

I can go as low as SIGMA=0.05; lower values do not converge, but terminate
after 60 steps (the default maximum number of steps?), without reaching the
required accuracy.

Rob.

[Vlad]

Sigma needs to be so low that E and E(0) are close within your required level of accuracy. E and E(0) will be exactly the same when the populations of the atomic levels are all integer numbers (as they should be in an atom), and you do not have fractional populations. You can help converging results with lower sigmas if you start with WAVECAR and CHGCAR previously converged with a larger sigma, and ISTART=1 in INCAR. You can also switch to a different minimization algorithm.

Vladimir

[admin]

if the electronic structure is not cenverged in 60 steps, increase NELM in INCAR.