The lattice geometry optimization of SSCHA

[biglinn]

From the section 3.2.3 of paper J. Phys.: Condens. Matter 33 363001, the direct minimization is performed at a constant volume.
If applying lattice geometry optimization, it means to perform full minimization at each step of lattice geometry optimization.

I find that the object for optimization relax = sscha.Relax.SSCHA(minim, calculator, N_configs, max_pop) needs a calculator to get energies, forces and stress.
In the https://sscha.eu/Tutorials/tutorial_02_advanced_submission/ , a script espresso_calculator.py is given to substitute the ase calculator. If I use VASP to perform calculations, how could I do?
Is there some guides to help writing a script as a calculator for sscha.Relax.SSCHA like espresso_calculator.py?

[mesonepigreco]

Hi, there is a VASP thread here.
Some people tried (successfully) to use VASP with the SSCHA; you can ask them.
In general, manual submission should always work with Vasp; you could use a simple ASE calculator for automatic local submission.
Automatic submission on a cluster has not been implemented with VASP, and it only works with Espresso.

You can look at this discussion for more details on how to combine VASP with SSCHA
#161

[biglinn]

Hi, I don't aim to get automatic submission with VASP.
I find that the sscha.Relax.SSCHA(minim, calculator, N_configs, max_pop) needs a calculator. I don't know how to do if I want to use sscha.Relax.SSCHA to apply lattice geometry optimization when using VASP.

[biglinn]

If I use VASP with manual submission, how could I set relax object to apply lattice geometry optimization?

[biglinn]

According to the paper J. Phys.: Condens. Matter 33 363001, lattice geometry optimization is to perform full minimization at each relaxing step.
If I use VASP with manual submission, should I write scripts to establish this workflow?

[diegomartinez2]

Hi, someone that used VASP with SSCHA successfully here. I used a script to made the VASP calculations with Machine learning because the ASE din't have implemented the ML part of the VASP. They (ASE team) commented that they were working on it.
The script I used was https://github.com/diegomartinez2/vasp-phonopy-sscha, that is a fork of https://github.com/QuantumMaterialsModelling/vasp-phonopy-sscha with modifications as it didn't work for me as is.

[Guodonglin-cqust]

python interface.py --to_vasp 1 4 4 2

Supercell multiplier: 1.88973
basis_vectors: [[11.5669501626, 0.0, 0.0], [0.0, 4.1907407448, 0.0], [0.0, 0.0, 4.415116444]]
type_atoms: ['Se', 'Sn']
atom_occurencies: [4, 4]
unit_cell_atoms:
['Se', 'Se', 'Se', 'Se', 'Sn', 'Sn', 'Sn', 'Sn']
atomic_positions: [[0.6422985664, 0.75, 0.0262445387], [0.1422985664, 0.75, 0.4737554613], [0.8577014336, 0.25, 0.5262445387], [0.3577014336, 0.25, 0.9737554613], [0.8811385456, 0.75, 0.0977187629], [0.3811385456, 0.75, 0.4022812371], [0.6188614544, 0.25, 0.5977187629], [0.1188614544, 0.25, 0.9022812371]]
n_atoms: 8
scf_population1_1.dat
Traceback (most recent call last):
File "/home/cqwl111/vaspfile/vasp-phonopy-sscha/vasp-phonopy-sscha-main/vasp-phonopy-sscha//interface.py", line 278, in
ensemble.poscar_to_vasp(np.array([args.to_vasp[0][1]]))
File "/home/cqwl111/vaspfile/vasp-phonopy-sscha/vasp-phonopy-sscha-main/vasp-phonopy-sscha//interface.py", line 238, in poscar_to_vasp
if self.positions.iloc[index,0] == atom:
~~~~~~~~~~~~~~~~~~~^^^^^^^^^
File "/home/cqwl111/anaconda3/envs/sscha/lib/python3.11/site-packages/pandas/core/indexing.py", line 1183, in getitem
return self.obj._get_value(*key, takeable=self._takeable)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/cqwl111/anaconda3/envs/sscha/lib/python3.11/site-packages/pandas/core/frame.py", line 4212, in _get_value
return series._values[index]
~~~~~~~~~~~~~~^^^^^^^
IndexError: index 256 is out of bounds for axis 0 with size 256
(sscha) cqwl111@cqwl111:~/vaspfile/vasp-phonopy-sscha/123/pop1$

It produce only one POSCAR, namely, POSCAR_1
like this

generated by phonopy
1.0
23.133901980749023 0.0 0.0
0.0 16.762964178662624 0.0
0.0 0.0 17.66046703990777
Se Sn
256 256
Cartesian
9.889681702182374 1.2011605309461892 2.35729533555372
1.5652189463590414 3.137819703386992 2.0865027818685866
7.380164406446138 3.213037375369724 0.1296090404913471
4.102100559169548 0.9692193065962112 4.203725305168796
9.922473758193991 1.1764905400088443 6.747007535193804
1.743334949158089 2.9325169561685254 6.431619373165528
7.506961662731727 3.1036188385312866 4.577252551994153
4.235047400604649 0.8117480117405399 8.889041909486004
9.961885895740636 1.1397003853614682 11.049472380823453
1.6306647140912924 3.150415899943438 10.879831386384392
7.160307866165425 3.0917681733434663 8.844757161126017
3.948831400676625 0.9537343868359204 13.15660789240436
10.080434399161994 1.1895169999743112 15.554205974890884
1.4992229864271618 3.066785066198583 15.321662710015802
7.457898468861283 3.21561808488876 13.457495678041871
4.188346368730646 0.9788923409295498 17.366765591066844
9.90601294851038 5.368318974765683 2.4336685130949687

[diegomartinez2]

It looks like the script is trying to take a 4x4x4 instead of a 2x4x4, so is taking only the first 4. I checked the script an it takes that input as an array, so try with (2,4,4) or 2,4,4.

[Guodonglin-cqust]

I want to test the (2,4,4) or 2,4,4., but the computer system is broken. so i install the ssach code on another computer with Ubantu.

use the same method

conda create -n sscha -c conda-forge python=3.11 gfortran libblas lapack openmpi julia openmpi-mpicc pip numpy scipy spglib=2.2

conda activate sscha

pip install ase julia mpi4py

pip install cellconstructor python-sscha tdscha

when i type the "python interface.py --generate 100 1 18 300",

(sscha) gengzi@gengzi:/anaconda3/envs/pop1$ sh test.sh
Traceback (most recent call last):
File "/home/gengzi/anaconda3/envs/pop1/vasp-phonopy-sscha-main/vasp-phonopy-sscha//interface.py", line 13, in
import cellconstructor.Phonons
File "/home/gengzi/anaconda3/envs/sscha/lib/python3.11/site-packages/cellconstructor/Phonons.py", line 25, in
import cellconstructor.calculators as calculators
File "/home/gengzi/anaconda3/envs/sscha/lib/python3.11/site-packages/cellconstructor/calculators.py", line 8, in
import ase, ase.io
File "/home/gengzi/anaconda3/envs/sscha/lib/python3.11/site-packages/ase/init.py", line 8, in
import ase.parallel # noqa
^^^^^^^^^^^^^^^^^^^
File "/home/gengzi/anaconda3/envs/sscha/lib/python3.11/site-packages/ase/parallel.py", line 227, in
world = MPI4PY()
^^^^^^^^
File "/home/gengzi/anaconda3/envs/sscha/lib/python3.11/site-packages/ase/parallel.py", line 129, in init
from mpi4py import MPI
ImportError: /home/gengzi/anaconda3/envs/sscha/lib/python3.11/site-packages/mpi4py/MPI.cpython-311-x86_64-linux-gnu.so: undefined symbol: MPI_Neighbor_alltoallv_c
(sscha) gengzi@gengzi:/anaconda3/envs/pop1$

for mpi4py/MPI.cpython-311-x86_64-linux-gnu.so: undefined symbol: MPI_Neighbor_alltoallv_c, i do not know how to fix it, could you tell me.

[Guodonglin-cqust]

It looks like the script is trying to take a 4x4x4 instead of a 2x4x4, so is taking only the first 4. I checked the script an it takes that input as an array, so try with (2,4,4) or 2,4,4.

(base) cqwl111@cqwl111:$ cd /home/cqwl111/vaspfile/vasp-phonopy-sscha/123/pop1/
(base) cqwl111@cqwl111:/vaspfile/vasp-phonopy-sscha/123/pop1$ conda activate sscha
(sscha) cqwl111@cqwl111:~/vaspfile/vasp-phonopy-sscha/123/pop1$ sh test.sh
Supercell multiplier: 1.88973
basis_vectors: [[11.5669501626, 0.0, 0.0], [0.0, 4.1907407448, 0.0], [0.0, 0.0, 4.415116444]]
type_atoms: ['Se', 'Sn']
atom_occurencies: [4, 4]
unit_cell_atoms:
['Se', 'Se', 'Se', 'Se', 'Sn', 'Sn', 'Sn', 'Sn']
atomic_positions: [[0.6422985664, 0.75, 0.0262445387], [0.1422985664, 0.75, 0.4737554613], [0.8577014336, 0.25, 0.5262445387], [0.3577014336, 0.25, 0.9737554613], [0.8811385456, 0.75, 0.0977187629], [0.3811385456, 0.75, 0.4022812371], [0.6188614544, 0.25, 0.5977187629], [0.1188614544, 0.25, 0.9022812371]]
n_atoms: 8
scf_population1_1.dat
Traceback (most recent call last):
File "/home/cqwl111/vaspfile/vasp-phonopy-sscha/vasp-phonopy-sscha-main/vasp-phonopy-sscha//interface.py", line 278, in
ensemble.poscar_to_vasp(np.array([args.to_vasp[0][1]]))
File "/home/cqwl111/vaspfile/vasp-phonopy-sscha/vasp-phonopy-sscha-main/vasp-phonopy-sscha//interface.py", line 231, in poscar_to_vasp
f.write(f"{index*int(qpoints[0])*int(qpoints[0])*int(qpoints[0])} ")
^^^^^^^^^^^^^^^
ValueError: invalid literal for int() with base 10: '2,4,4'

(sscha) cqwl111@cqwl111:/vaspfile/vasp-phonopy-sscha/123/pop1$ sh test.sh
test.sh: line 32: syntax error near unexpected token (' test.sh: line 32: python $QE_BIN/interface.py --to_vasp 1 (2,4,4)'
(sscha) cqwl111@cqwl111:/vaspfile/vasp-phonopy-sscha/123/pop1$

[Guodonglin-cqust]

It looks like the script is trying to take a 4x4x4 instead of a 2x4x4, so is taking only the first 4. I checked the script an it takes that input as an array, so try with (2,4,4) or 2,4,4.

python interface.py --to_vasp 1 [2,4,4]

generated by phonopy
1.0
23.133901980749023 0.0 0.0
0.0 16.762964178662624 0.0
0.0 0.0 17.66046703990777
Se Sn
32 32
Cartesian
9.889681702182374 1.2011605309461892 2.35729533555372
1.5652189463590414 3.137819703386992 2.0865027818685866

the results are wrong, it should be 128(Se), 128(Sn )

[mesonepigreco]

This is likely an error with setuptools. It should be fixed to a lower version, most likely <= 64.
You can downgrade setuptools with

pip install --force-reinstall setuptools==64

Then try again with the command that failed.

[Guodonglin-cqust]

Based on your suggestion, I have successfully installed it. Thanks.