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 ... 353
1
General Questions and Answers / Re: error in mpiexec.hydra
« on: September 19, 2023, 19:44 »
It's a bit "old", but more importantly we have had issues with Intel MPI for a while. I have also seen reports online where people solved issues like this by upgrading their Intel MPI version, which in your case would be easiest done by moving to 2022.12 or even 2023.09, that way you also get access to all the latest features! There is not much we can do to troubleshoot an old version, esp. if this is a bug in Intel MPI as it very well might be.

2
General Questions and Answers / Re: error in mpiexec.hydra
« on: September 19, 2023, 07:12 »
What actual version of QuantumATK is this? There have been some issue with Intel MPI in the past. With 2022.12 things improved, but ideally you should run 2023.09. Also, is there some other MPI library in your path, or generally recommended/used on this cluster?

3
General Questions and Answers / Re: Basis set for Hydrogen
« on: September 18, 2023, 22:38 »
ConfinedOrbitals and HydrogenOrbitals are really quite similar in shape and function. Both have compact support (they go to zero exactly at a certain cut-off radius), and they are both based on solid harmonics for the angular part. The only difference is how we determine the radial part, which in both cases is a numerical tabulated function in the end.

For a ConfinedOrbital we use the pseudo- and core density from the pseudopotential to calculate Hartree and exchange-correlation terms. For a HydrogenOrbital, there is no other density than the one for the single orbital that you are solving for, so the Hartree and exchange-correlation terms are zero and we are left with only a 1/r potential.

4
Yes, it is possible if the model is of the common Slater-Koster type.
An example on how to create your own model is given here:
https://docs.quantumatk.com/tutorials/slater_koster/slater_koster.html
You need a small change in order to use this in the newest versions of QuantumATK, you need to cast each offsite argument to a list, so just replace zip(...) with list(zip(...))

Once you get the model to work (which can be a bit tricky, so don't hesitate to reach out for help), you can use it for band structure and NEGF calculations.

5
General Questions and Answers / Re: error in mpiexec.hydra
« on: September 18, 2023, 21:13 »
More info, but not clue. Now please add the -v option to mpiexec.hydra also, but post the results in an attachment text file, rather, as it will be very long.

6
General Questions and Answers / Re: error in mpiexec.hydra
« on: September 15, 2023, 19:27 »
The use of localonly might be specific to Linux vs Windows. I use it all the time on Windows if I launch from the command line to avoid having to set up ssh keys or an MPI service, or give my password to each ssh process. Perhaps there is a smarter way to do it, but there is no indication this keyword is deprecated on Windows at least (https://www.intel.com/content/www/us/en/docs/mpi-library/developer-reference-windows/2021-10/global-hydra-options.html).

But you use Linux, rebacoo, so indeed you can skip it.

Did you try the -v version (mpiexec.hydra -v -np 4 ...)? Also, you can try

mpiexec.hydra -n 4 -genv I_MPI_HYDRA_DEBUG=1 -genv I_MPI_DEBUG=5 atkpython script.py

to generate a LOT of debug info that might help

7
Well, actually an isosurface by definition only has 1 color (or 2, which is why I asked about the phase). The isosurface consists of all the points in space where the quantity you are plotting are equal to a certain value. The only way it can have "many" colors is if we then color it by a different quantity. The reason we sometimes see 2 colors is that we can pick the points by absolute value and color by sign, or phase.

If you want to see the potential variation in a plane, use cut planes instead.

8
General Questions and Answers / Re: error in mpiexec.hydra
« on: September 14, 2023, 19:25 »
To troubleshoot something like this, I would try a few things

* Can you run a trivial command in parallel, like mpiexec.hydra -np 4 echo "hello".
* Is parallelization across nodes set up correctly in general on the cluster? Can you run mpiexec.hydra -np 4 -localonly echo "hello".
* Is your path set up correctly, so that mpiexec.hydra actually points to our binary? Same for atkpython
* Always use the latest version of quantumatk. We released 2023.09 just this month
* Make sure the test system is really small (start with 1 Au atom), to exclude problems like running out of memory
* Add the -v option to mpiexec, it will print tons of debug information

9
General Questions and Answers / Re: DFTB basis set
« on: September 13, 2023, 03:55 »
This is not possible because unlike DFT, the DFTB basis sets are not just for the atoms themselves, but also contain pairwise interactions. So, all elements must be contained in the same parameter set, which obviously is a strong limitation on the applicability and transferability of DFTB.

If your goal is to run "quick-and-dirty" and somewhat approximate calculations, then I would go with the SingleZetaPolarized basis set in QuantumATK's LCAO-DFT model. It will be slower than DFTB  but orders of magnitude faster than plane waves, and also a lot faster than our default LCAO PseudoDojo basis sets.

10
If you need a real physical shift of the Fermi level, this would have to be done with doping. This can either be done with actual dopants, which puts limits on how low concentration you can use, and you need to simulate a large supercell no matter what, or you can use implicit doping as we do in several examples; see
https://docs.quantumatk.com/tutorials/inas_p-i-n_junction/inas_p-i-n_junction.html#adding-p-type-and-n-type-doping-to-inas
https://docs.quantumatk.com/tutorials/nisi2-si/nisi2-si.html#dope-the-device
https://docs.quantumatk.com/manual/technicalnotes/doping_methods/doping_methods.html

11
Not currently implemented, but in the list of ideas for the future

12
It's been a long time ago since this was discussed, and it's less supported than it was perhaps in ancient version of our code... But if you check the script projection.py at https://docs.quantumatk.com/tutorials/low_level_entities/low_level_entities.html#projecting-the-transmission, you can see in the line "w, v = scipy.linalg.eigh(H_dtb, S_dtb)" how to compute the eigenvalues (w) and eigenvectors (v).

13
General Questions and Answers / Re: Job manager crashes
« on: August 24, 2023, 00:35 »
This error has been posted many times on the Forum, you can find the answer by searching for any keyword in the error message, like "loadSettings".
The solution is provided in e.g. https://forum.quantumatk.com/index.php?topic=9720.msg32520#msg32520

14
General Questions and Answers / Re: Error
« on: August 23, 2023, 07:58 »
You should just use the energies from the DOS object, i.e.
energies = dos.energies()
If you are trying to zoom in on a certain energy range, use pylab.ylim(-2,2) instead.


15
Yes, we are aware of this. Most likely it is due to different defaults in the newer versions. Good example of the important of always checking your results for convergence, and impact of model assumptions. I don't know which parameter precisely introduces this difference, but we did for instance change the default electron temperature (1000 K, which is kind of high in some cases).

Pages: [1] 2 3 ... 353