Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Messages - Anders Blom

Pages: [1] 2 3 ... 361
Any tutorial which mentions NC files is just old. We use HDF5 files now for all data. So your previous step should have written an HDF5 file with the optimized structure, and you just read that same file in the next step.

Should be possible for Normconserving LCAO, yes.

It requires 2 steps.

1) Define constraints via tags. Supported tags are "constrain" and "constrain_x" (y, z) in order to constrain each atom fully or only in x/y/z
2) Use a few lines of code

Code: python
from AddOns.VASPPlugins.Scripter.VASPConstraintsWidget import extractConstraintsFromTags
from AddOns.VASPPlugins.IO import writePOSCAR
vasp_constraints = extractConstraintsFromTags(configuration)
writePOSCAR(configuration, "POSCAR", vasp_constraints)

Example config attached, with the resulting POSCAR.

Note that you can also do this directly from the Builder, using Export and choosing the VASP format.

Why do you need to do that? In general I would say it's not possible, these are not standard file formats you can just convert between.

From these files it is at least clear that the self-consistent loops proceeds very differently in the two cases for some displacements (but not all), and in the end of course this makes a big difference for the dynamical matrix.

Now, when I submit this to our bug fixing team, the first thing they will ask is if it was tested in the latest version. I personally don't have a machine with 168 cores at hand to test on, but you do, so it would be really helpful if you could rerun this with V-2023.12-SP1 (which has many new features compared to U anyway), since in theory it could have been fixed since the version you used. This should only take you 7-8 hours :-)

Second, I find your choice of processes_per_displacement=28 odd when you run on 24-core machines. This gives at the very least odd load-balancing and unnecessary network communications.

So, please check using processes_per_displacement=24 instead. One can argue that any value should work numerically, so we might still look into it, but only if it appears in V-version also.

The nice news, once we resolve this, is that the speedup using a large processes_per_displacement is substantial, basically linear!

The paper you refer to is done with DFT, but your model in the script is using an extended Huckel model. That might work, sometimes, and would save a lot of time, however as your test shows, for this system the Huckel parameters do not give the correct band structure etc for B-P, thus it cannot be used.

You might try other parameters such as the Muller or Hoffmann models, and also it will be important to perform the calculations self-consistently since there will be significant charge transfer between B and P atoms (might try that with Cerda parameters too but I doubt it will work, as the two parameter sets you are trying to mix come from fits to AlP and BN).

I think your main option will be DFT in this case.

Add the dielectric region, going all the way out to the edges of the system in C (Z), before creating the device. Then when you use Device from Bulk, the regions will automatically be copied into the electrodes.

Looks like the main.ini file was corrupted, try rename it to main-backup.ini and try again. The file is in C:\Users\username\.quantumatk or .vnl for old versions.

These are for 28 processes, right? The point will be to compare with the run using 4, so please upload those too.

Could you zip the log files too, might help to look at them before rerunning, the HDF5 files don't tell us much how the problem arises

I don't think you need a script, actually. If you open the IV curve in the GUI (NanoLab) you should be able to change the curves to show any spin component.

I think we need to look at the exact scripts and output. If you don't want to share them publicly, you can email then.

All doable, yes

The script is rather specific to the calculation done with an explicit defect. In your case, you should go back to definitions in the theory section; looking at the line which actually defines the MFP, you can evaluate all quantities for your system and extract the MFP as

(R(L)/Rc-1) = L/MFP

(all depending on energy).

Here Rc is the conductance of the perfect system, and R(L) = 1/T(L) is computed for a device where L is the length of the part that has been thermally perturbed. A better option might even be to plot R(L) as a function of L, which should be a linear relation and you can get MFP from the slope.

Pages: [1] 2 3 ... 361