代码拉取完成，页面将自动刷新
The current code is the implementation of the xDeepH (extended Deep Hamiltonian) method described in the paper Deep-learning electronic-structure calculation of magnetic superstructures (accepted by Nature Computational Science).
The current xDeepH repository is based on DeepH-E3 and will be integrated into DeepH-pack in the future.
To train the xDeepH model, Python 3.9 interpreter and following packages are required：
In Linux, you can quickly achieve the requirements by running
# install miniconda with python 3.9
wget https://repo.anaconda.com/miniconda/Miniconda3-py39_4.10.3-Linux-x86_64.sh
bash Miniconda3-py39_4.10.3-Linux-x86_64.sh
# install packages by conda
conda install numpy
conda install scipy
conda install pytorch==1.9.1 ${pytorch_config}
conda install pytorch-geometric=1.7.2 -c rusty1s -c conda-forge
conda install pymatgen -c conda-forge
# install packages by pip
pip install e3nn==0.3.5
pip install h5py
pip install tensorboard
pip install pathos
pip install psutil
with ${pytorch_config} replaced by your own configuration. You can find how to set it in the official website of PyTorch.
The usage of xDeepH is similar to that of DeepH-pack (https://github.com/mzjb/DeepH-pack).
Download the processed dataset from Zenodo and unzip it to the directory data/monolayer_CrI3/.
Alternatively, you can also use OpenMX to perform
the spin-constrained DFT calculation, and then convert the output of DFT codes to the format
that can be directly read by xDeepH using the deeph-preprocess command of DeepH-pack
(https://github.com/mzjb/DeepH-pack/#prepare-the-dataset) to obtain a dataset. You should set
parse_magnetic_moment to True and local_coordinate to False in the config file of deeph-preprocess
(https://github.com/mzjb/DeepH-pack/blob/1130ad648d8b02be96e9de0144f8f8cff886a178/deeph/preprocess/preprocess_default.ini#L19).
Based on the downloaded data, you can train an xDeepH model. Run the following command:
python -u xdeeph-train.py examples/CrI3_monolayer.ini
The config CrI3_monolayer.ini includes a description of each input variables.
Once you have a trained model, you can use that model to predict the Hamiltonian of some material structures
and get hamiltonians_pred.h5.
Like DeepH-pack, you should first calculate the overlap for the desired
materials using 'overlap only' OpenMX (https://github.com/mzjb/overlap-only-OpenMX), and then convert
it to our format using the deeph-inference command of DeepH-pack
(https://github.com/mzjb/DeepH-pack#inference-with-your-model). Please note that the deeph-inference
command is only used for converting the overlap format and not for performing model inference.
You should set task to [1] in the config file of deeph-inference
(https://github.com/mzjb/DeepH-pack/blob/1130ad648d8b02be96e9de0144f8f8cff886a178/deeph/inference/inference_default.ini#L6).
After obtaining the converted overlaps.h5 file, you should create a magmom.txt file to specify the magnetic structure of the material you want to predict.
This is a text file where each line corresponds to an atom. The first column indicates whether the atom
is magnetic, with 1 for yes and 0 for no. The second column indicates the magnitude of the magnetic moment.
The third column indicates the polar angle of the magnetic moment in spherical coordinates.
The fourth column indicates the azimuthal angle of the magnetic moment in spherical coordinates,
with angles in degrees. Here is an example of a magmom.txt file:
1.0e+00 3.46e+00 3.23e+01 -5.62e+01
1.0e+00 3.45e+00 4.39e+01 7.07e+01
0.0e+00 7.75e-02 9.43e+01 1.11e+02
0.0e+00 1.43e-02 1.58e+02 8.73e+01
0.0e+00 4.68e-02 1.32e+02 -3.16e+01
0.0e+00 1.41e-01 1.48e+02 3.06e+01
0.0e+00 7.55e-02 8.40e+01 1.11e+02
0.0e+00 9.92e-03 1.20e+02 8.54e+01
Inference of xDeepH can be done by the command:
python -u xdeeph-eval.py examples/CrI3_monolayer_inference.ini
This config also includes a description of each input variables.
The band structure of the material can be calculated from the predicted Hamiltonian. The calculation is also based on the sparse calculation script from DeepH-pack (https://github.com/mzjb/DeepH-pack/blob/main/deeph/inference/sparse_calc.jl). The required Julia environment is also identical to that of DeepH-pack (https://github.com/mzjb/DeepH-pack#julia).
Run the following command:
julia sparse_calc.jl -i ${predicted_hamiltonian_path} -o ${output_path} --config config.json
where config.json is the config for the sparse calculation.
The description of parameters in this config can be found in
https://deeph-pack.deepmodeling.com/en/latest/keyword/inference.html#json-configuration-file.
The predicted hamiltonians_pred.h5 and files generated by deeph-inference
(overlap.h5, info.json, rlat.dat, orbital_types.dat, site_positions.dat)
must be placed in the ${predicted_hamiltonian_path} directory.
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
马建仓 AI 助手