My Community

Dear VASP Forum,

I am currently working on energy level and band bending analysis for a semiconductor-organic molecule interface using VASP. However, I am facing an issue where the HOMO and LUMO energy levels of the organic molecule are significantly deeper than expected, resulting in poor alignment with the semiconductor's band edges. This misalignment is affecting the expected electronic interactions at the interface.

Could you please suggest possible approaches or techniques that could be employed within VASP to shift the energy levels of the organic molecule upwards, relative to the semiconductor, to improve the alignment?

There are many different techniques to change the level alignment in VASP. The most common approach would be to investigate whether different xc functionals can yield better results. This depends strongly on the physics that cause the misalignment. Typical choices would be metaGGAs (e.g. SCAN or MBJ) or hybrid functionals (e.g. HSE or PBE0).

A more expensive solution would be to compute GW or other many-body corrections. Keep in mind that this is orders of magnitude more demanding than local functionals.

Another possibility is to employ DFT+U. This adds an on-site Coulomb correction that changes the alignment of the corresponding bands. The computational cost is comparable to a local functional. The disadvantage of this approach is that the U chosen is typically an empirical parameter. There are ways to determine U ab initio but this adds more complexity.

Thank you for your suggestions, Martin.

The calculations involve a semiconductor slab (approximately 3 nm) and an organic molecule, with a vacuum present as well. Hybrid functionals do not seem to be a suitable option. DFT-1/2 provides a good band gap for the slab under quantum confinement. Is it feasible to modify the potentials of the atoms comprising the molecule to correct its band gap and achieve proper alignment with the slab?

What is DFT-1/2?

Regarding changing the potential, there is a way but that involves some coding. Assuming you have a recent version of VASP there is an ADD_TO_EXTERNAL_POTENTIAL routine in pot.F. If you call this routine at some point before with an appropriately constructed external potential, VASP should compute the correct ground state given this external potential.

We plan to make this kind of features more user friendly in the future but at the moment you need to be able to read and write Fortran code to do that.

Hi Martin,

As you might know, the DFT-1/2 method is a semi-empirical approach designed to correct the self-interaction error in local and semi-local exchange-correlation density functionals, (Phys. Rev. B 78, 125116). This method involves using modified POTCAR files for the atoms to correct the electronic bandgap. However, this method does not work for molecules.

Is there any technique like DFT-1/2 method in which we can we modify the POTCAR files mainly for molecule simulation purposes?

Thank you for providing information about the ADD_TO_EXTERNAL_POTENTIAL routine in pot.F. Could you please clarify if recompiling VASP is necessary to utilize this feature?

I am not aware of such a method in VASP but I will inquire if something similar is possible.

If you change any source file in VASP you need to recompile the code so that the executable is changed to incorporate this new feature.

Dear Martin,

Thank you for your response and for looking into this! I am particularly interested in the possibility of an empirical projector shift method for pseudopotentials in VASP, similar to what is discussed in this
https://docs.quantumatk.com/tutorials/d ... f_pps.html

Such a feature could be very useful for fine-tuning energy levels, especially in heterostructure systems. I look forward to any updates on this possibility.

Thank you for the suggestion, I will bring it up in out internal meetings.

You could also look into the MBJ functional (METAGGA=MBJ), which is designed to improve bandgaps. Perhaps it also helps with the level alignment.

Hi Martin,

I am facing convergence issues with LMBJ calculations, even when utilizing up to 512 cores. The calculations cannot converge within a 24-hour walltime, and we are unable to restart the calculation from the previous LMBJ run.

Could you advise on more efficient parallelization strategies or a solution to enable restarts for LMBJ runs?

Thank you for your assistance.

Hi,

Could you please provide the OUTCAR file of your calculation such that we can see how the energy varies during the iterations? Showing us the INCAR file may also be useful.

The LMBJ implementation uses some technical approximations that may have some non-negligible influence on the results, in particular when there are transition-metal atoms (see forum/viewtopic.php?t=19115). A better alternative may be to use MBJ with a fixed (arbitrary) value of the parameter \(c\) (see CMBJ). A reasonable range of value for \(c\) is 1.1-1.3.

Hi Fabien,

I reduced the energy cutoff from 550 eV to 450 eV and relaxed the energy convergence criteria from 1E-5 to 1E-3, which allowed me to complete the calculation within the 24-hour wall time. I have attached the relevant files for your reference.

However, I noticed that the number of iterations from 1E-1 to 1E-2 and from 1E-2 to 1E-3 was approximately 60. Could this be reduced by modifying the TIME parameter in the INCAR file? Or would other adjustments be more effective in optimizing the iteration steps?

Hi,

That is a good that you could converge the calculation within the time limit. The options that I used for calculations with LMBJ on systems with vacuum were:
ALGO = A
IMIX = 1
AMIX = 0.4
AMIN = 0.01
However, the convergence was still not great in some cases. You may give a try, but I do not expect a big improvement. In general, calculations with (L)MBJ are not fast, especially in the presence of vacuum it seems.

Thank you. I will try that.