How to remove the spiky features in layer-resolved density of states?

Dear all,

For identifying the interfacial effect of the heterogeneous metal-semiconductor contacts, I have tried to extract the layer-resolved density of states by using prtdos 3 option.
The problem is, there are large spikes in LDOS (to show the overall dos of each atom, I summed up the dos of each orbital component e.g. l =0,1,2,3,4).
Results are shown below.
I constructed the system using Au, Ti, and O to mimic the Au/TiO2 interface. However, I found that large spiky patterns randomly exist in the overall curve. For instance, there are huge deeps around ~2 eV region of the 4th Ti curve. And also, the dos of each Ti does not consistent for - 6 and 2 eV regions.
My question is,

  1. It is natural to have this kind of deeps?
  2. How can I get rid of this spiky pattern and get the smooth DOS?

Thank you for reading my post, I hope you have a nice day!

Best regards,
Jun Heo

Hi Jun,

the spikes come from the limited k point sampling in your calculation.

  • Easy fix: just smooth the DOS curves with your plotting software.
  • Next, calculate with Gaussians instead of tetrahedra (I think there is a combination of options for this in the prtdos stuff), and adjust the width to what you want.
  • Last, most accurate but most time consuming, run with a better density of k-points. The biggest spikes come from tetrahedra which are degenerate in energy, and with more k these become quite rare - the DOS should be a sum of parabola segments, so smooth with some derivative discontinuity when you hit the edge of certain tetrahedra.

As you are running a surface, adding kz perpendicular to it will be hard, or you have to boost the number of layers. In that case it will prove hard to make the DOS properly smooth…

1 Like

Thank you for your answers.
I hope you had a happy Christmas.

For the calculation, I used 416 k-points for 45 atoms (4 gold and 4 TiO2 unit cells) but it might be not enough to properly cover the system (e.g. k point options or nkgpt values… as you mentioned above). Based on your suggestion, it seems that increasing kz would be one of the solutions which might be worth trying.
Additionally, I have a question about the meaning of the “boost the number of layers”. Is that mean creating more physical layers would be helpful?

Thank you.

Best regards,
Jun Heo

So: if you have a slab geometry with vacuum there should not be any effect of adding kz points (no dispersion of electronic bands across the vacuum).

The physical equivalent is to add layers of substrate, where the different extended dispersive band states will slowly build up in real space. However this is very expensive, have to see what you can afford. First and most important thing is to add k in plane

1 Like

Thank you for your explanation!

So, that’s why you suggested smoothing.

Thank you again for your answers.
Have a nice day!