FDBM simulations with the structure model of the tip

[andr3ac]

I'm trying to perform AFM simulations using the FDBM approach with a structural model of a CuOx tip, based on the structure shown in Figure 2d of this paper: https://doi.org/10.1039/D1NR04080D.

During my initial attempts, I encountered errors because the default valence electron dictionary did not include entries for copper (Cu). To resolve this issue, I modified the "valelec_dict.py" file by adding the entry "29: 11.0," corresponding to the Cu atomic number and the ZVAL value from my POTCAR file. I'd like to confirm whether this is the correct procedure, or if I should modify something else as well.

Beyond this technical fix, I have conceptual questions regarding the use of atom clusters in FDBM. The example of pyridine ("pyridineDensOverlap") you provide uses a simple CO dimer in an empty box, presumably because the apex atom and stiffness are the primary contributors to the AFM contrast. I would be interested in using a more complete tip model similar to the one in Figure 2c of the referenced paper. I'd like to know if you have experience running simulations with such detailed models and if the code requires specific adjustments to handle a multi-atom cluster, rather than a single atom (such as oxygen) or a dimer (such as CO).

Additionally, could you explain how the cluster should be aligned within the empty cell to avoid simulation artifacts, and how the tip stiffness parameters should be handled when replacing the standard effective particle with a full structural model?

Thank you

[NikoOinonen]

I modified the "valelec_dict.py" file by adding the entry "29: 11.0," corresponding to the Cu atomic number and the ZVAL value from my POTCAR file.

Yes, this sounds correct.

I would be interested in using a more complete tip model similar to the one in Figure 2c of the referenced paper. I'd like to know if you have experience running simulations with such detailed models and if the code requires specific adjustments to handle a multi-atom cluster, rather than a single atom (such as oxygen) or a dimer (such as CO).

To my knowledge, at least with the FDBM, all existing literature only uses the CO without the metallic tip. I have thought also that it would be interesting to test with a more complete tip structure.

Additionally, could you explain how the cluster should be aligned within the empty cell to avoid simulation artifacts, and how the tip stiffness parameters should be handled when replacing the standard effective particle with a full structural model?

The code assumes that the final atom of the functionalized the tip (the O atom) is centered at the origin of the periodic cell. You also have to make sure that there is enough vacuum above the metal tip so that the tip in the periodic convolution does not hit the bottom of the periodic image of the sample. There is also some difference if you are running the simulation through the CLI or the OpenCL Python API. In the CLI, currently, the grid of the sample and the tip have to have the same step and dimension, whereas in the OpenCL version the tip grid can be interpolated to match the sample grid (although this can result in artifacts in some cases).

As for the stiffness, there is no difference to the particle-without-tip case, since the stiffness of the probe particle should be fitted as if it's bending with respect to the tip. I'm not familiar with the CuOx case specifically, I imagine the stiffness in much higher than CO. You would have to look in the literature if someone has figured out some value, or otherwise fit it to DFT simulations.

@ProkopHapala Do you have any perspective on this?

[ProkopHapala]

Hi,

I think it should be in principle possible to include the metallic cluster in the simulation, and I remember we were experimenting with that back in ~2015-2016.

The code should be perhaps capable to handle it without fundamental changes (except changing some parameters or defauls as you did), however I did not tested it with the recent version of the code so it is quite probable some issue will emerge.

I would be carefully about several things

1. the the grids in which the tip and sample are computed must have exactly the same dimensions (number of voxes, size of voxels, lattice vectrors), if the tip is high you need to compute sample in larger grid. Make sure there is enough space arround both tip and sample in all directions (unless it is periodic)
2. As Niko said - make sure the apex atom of tip is at origin (0,0,0), that is the corner of the grid
3. code assumes periodic boundary conditions in all dimensions, be sure there are no sharp discontinuitis e.g. due to dipole correction in your DFT calculation, that would cause serious problems
4. For heavy atoms like metals there may be some deep spikes in potential and density (especially when using all-atoms like FHI-aims, VASP PAW should be more OK), this may produce some artifacts
5. It is quite possible that the back-side of the metalic cluster (the edges) produce quite strong electric fields which may produce visible features in the image, while being fake (as the real tip is not cut on top)

I'm myself curious to hear from you if it works when you try.

[ondrejkrejci]

Hi @andr3ac ! Is this issue still open, or was it solved? Is there maybe something, that we should add to the documentation? Thank you!

[andr3ac]

Hi @ondrejkrejci, thanks for following up!

I have tried the FDBM simulations using the actual CuOx tip model following your previous suggestions. However, I am currently unable to reproduce the experimental contrast, specifically regarding the surface Cu atoms.

Setup

• Sample: Partially oxidized Cu(110) surface (based on https://doi.org/10.1039/D1NR04080D). (POSCAR_sample.txt)

• Tip: CuOx cluster model (based on the same reference), placed in the unit cell with the apical oxygen at (0, 0, 0). (POSCAR_tip.txt)

• Large supercell: ≈20.5 Å × ≈21.7 Å × 25 Å.

• DFT: VASP with a high cutoff (600 eV) to minimize artifacts and the same grid for both the sample and tip calculations.

• FDBM Config:

  • valelec_dict.py: Updated to include the Cu element by adding the entry "29: 11.0," (as specified above).
  • params.ini: Default parameters (from the pyridineDensOverlap example), adapted only for cell size and scan region.
  • run.sh: Default parameters (from the pyridineDensOverlap example), but I increased the stiffness (-k) to account for the stiffer CuOx tip compared to CO. This procedure works properly for the "Hartree-potential electrostatics + Lennard-Jones" simulations (based on the LOCPOT file).

While the simulated contrast for the oxygen atoms appears somewhat reasonable (round bright features), the Cu atoms (which should be clearly visible and appear dark) are not resolved at any z-distance. See here a representative (it looks very similar at different tip-sample distances) FDBM-based AFM simulation:

Interestingly, AFM simulations using only the electrostatic potential of the surface (LOCPOT + LJ) reproduce the experimental contrast pretty well:

This suggests the issue is likely specific to the FDBM implementation or, more likely, the parameters I am using for the FDBM simulations using the CuOx cluster model.
Since I haven't experimented much with tuning the parameters beyond stiffness, do you have any suggestions on which parameters I should target? Or perhaps I need to make further adaptations to the code to properly handle the CuOx cluster? Do you think the POSCAR files (attached above as txt files) I am using are fine?

Any insights would be appreciated! Thank you!

[ProkopHapala]

@andr3ac

Hi, can you post how your tip clusters look like (CuO, Cu), and more even better how the electrostatic potential of these clusters look like?

I worry about two things

1. that the cluster is large and may cause strange things by its periodic replica
2. that the dipole/quadrupole moment of that cluster can controbute to the constrast by the edges on the oposite side, which is artifact of trying to reproduce half-inifinite tip by finite cutted cluster. But there may be way how to handle this properly.
   E.g. using 1-2 layers or Cu atoms and then 1-4 Cu atoms sticling from it down. That could shield the unwanted fields from the opposite side.