CSC calculation(Waiting for VASP lock to appear...)

[quantumable]

Hi,

I can't do the csc calculation of the tutorial. I first did the single-shot calculation which ended well when I run the CSC calculation(vasp_dmft -n 144 -i 15 -j 1 -p ~/triqs/install/bin/vasp_std csc.py), it stops on the following message "Waiting for VASP lock to appear..." and runs for hours without progressing and this with or without ICHARG = 5; NELM = 1000; NELMIN = 1000 ;IMIX=0. I have compiled vasp with MKL and with lapack in both cases I have the same result.
how to calculate csc using vasp?

PS: I have compiled vasp and triqs vasp the same configurations of the gcc compiler

[the-hampel]

Please share more information to get an answer. Is Vasp running at all? Did you see any specific Vasp output in the std out or the folder where you run the job? Any error messages?

Did you modify the Vasp source code according to the instructions listed here: https://triqs.github.io/dft_tools/unstable/guide/conv_vasp.html#remarks-on-the-vasp-version .

[quantumable]

Thank you for this suggestion.
1-) By changing stdbuf -o 0 $MPIRUN_CMD -np $NPROC "$VASP_DIR" & by stdbuf -o 0 $MPIRUN_CMD -np 1 "$VASP_DIR" & ; this only works on one node ie vasp_dmft -n 24 -i 15 -j 1 -p ~/triqs/install/bin/vasp_std csc.py but not for example vasp_dmft -n 144 -i 15 -j 1 -p ~/triqs/install/bin/vasp_std csc.py( and I still have the calculation that stops on ....Waiting for VASP lock to disappear...)

2-) After the cacul dmft I created a new folder where I copied the files of the dmft calculation and at each refresh error the files n deleting the files and copying again from the folder dmft. I use of course the script daft et dmft+dft present on your site. I do not know where the error can come from.

> ```
> Traceback (most recent call last):
>   File "/home/anaconda3/lib/python3.8/runpy.py", line 194, in _run_module_as_main
>     return _run_code(code, main_globals, None,
>   File "/home/anaconda3/lib/python3.8/runpy.py", line 87, in _run_code
>     exec(code, run_globals)
>   File "/home/triqs/install/lib/python3.8/site-packages/triqs_dft_tools/converters/plovasp/sc_dmft.py", line 257, in <module>
>     main()
>   File "/home/triqs/install/lib/python3.8/site-packages/triqs_dft_tools/converters/plovasp/sc_dmft.py", line 254, in main
>     run_all(vasp_pid, dmft_mod.dmft_cycle, cfg_file, n_iter, n_iter_dft, vasp_version)
>   File "/home/triqs/install/lib/python3.8/site-packages/triqs_dft_tools/converters/plovasp/sc_dmft.py", line 139, in run_all
>     corr_energy, dft_dc = dmft_cycle()
>   File "/home/TUTO/csc.py", line 123, in dmft_cycle
>     SK.symm_deg_gf(S.Sigma_iw,ish=0)
>   File "/home/triqs/install/lib/python3.8/site-packages/triqs_dft_tools/sumk_dft.py", line 1847, in symm_deg_gf
>     helper.from_L_G_R(v.conjugate().transpose(), gf_to_symm[key], v)
>   File "/home/triqs/install/lib/python3.8/site-packages/triqs/gf/gf.py", line 673, in from_L_G_R
>     assert L.shape[1] == G.target_shape[0], "Dimension mismatch between L and G"
> AssertionError: Dimension mismatch between L and G

What parameter is responsible for this?

[quantumable]

Dear Alex,
I have to post in the main group to know if another person has succeeded in making the cdc calculation with VASP and has an idea of how to do it?
Bests

[the-hampel]

Hi,

1. but this means that in principle your calculation seems to work. Than the problems are related to your cluster environment, and I can't help you there. For me this works as expected on our cluster. Meaning I use VASP for CSC DMFT calculation regularly. Unless you provide me a detailed error message why Vasp is not starting I cannot help you. The message that you show me ....Waiting for VASP lock to disappear... is a regular output. Not the error message.

2. Which files did you copy? Did you copy the h5 file called vasp.h5 ? You need to recreate this file, because the results are stored inside the file. I am very certain that this is the problem.

[quantumable]

Dear Alex,

1-)when I do the usual Vasp calculation mpirun -np 144 Vasp_std it works fine. but does not work on multiple cores and nodes if I use the CSC calculation mode.

2-)Indeed I deleted the .h5 file and redid the single-shot dmft calculation. and restarted the calculation. and it just doesn't work.

I did install mpi4py and installed the code via the source.

[quantumable]

Dear Alex,

1-) I solved the problem by changing the threshold parameter in the DMFT and DFT+DMFT calculation and the calculation goes to the end.

2-) however the calculation works only if I modify vasp_dmt as you suggested(stdbuf -o 0 $MPIRUN_CMD -np $NPROC "$VASP_DIR" & by stdbuf -o 0 $MPIRUN_CMD -np 1 "$VASP_DIR" &) and on only 1 node(of 24core in my case). do you have an idea why the CSC calculation does not work on several nodes?

3-) I also have the last version of vasp6.2.1 have you ever tried this version with triqs because on the file to modify main.F is not the same as that of version 5.4.4

4-)for my system for example which is composed of molecules + substrate and whose selected energy window (EWINDOW in plo.cfg) contains a lot of electrons (for example 2000 or more) the dmft calculation is not running. but if I reduce the window the calculation is running. the concern is that by reducing the window where the d-electron is present, I am afraid of not capturing the physics that it describes.

> 
> 
> PS: in the script it is also necessary to modify 
> mpi.report("Orbital %s density: %.6f"%(sig,dm[sig][0,0]))
> mpi.report('Total charge of Gloc : %.6f'%S.G_iw.total_density())
>  
> to
> 
> mpi.report("Orbital %s density: %.6f"%(sig,dm[sig][0,0].real))
> mpi.report('Total charge of Gloc : %.6f'%S.G_iw.total_density().real)
> 

[the-hampel]

No I did not try with the latest version of Vasp. In principle it should work, but as the source code slightly changed I cannot give you exact patch files.

I never tested the converter with that many electronic states in the window, and I think I am also not 100% sure why you need that. Do all of the 2000 electronic states hybridize with the target d states? If not you can start cutting them out. But again these things are physical questions, which I do not intend to help you with.

If you want to fix the Vasp CSC I really need the raw stdout and stderr files, that slurm creates to analyze any further what goes wrong. We saw that Vasp is not correctly starting in that case. I suspect that this is due to the cluster mpi configuration.