Skip to content

SIAB

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History
 
 

SIAB

README.md

SIAB Package Description

Systematically

Improvable

Atomic orbital

Basis generator based on spillage formula

HOW TO USE SIAB

The optimization can choose one of the three minimization methods:

  • Simulated Annealing (SA),
  • PyTorch Gradient (PTG),
  • PyTorch Gradient with dpsi (PTG_dpsi).

The executable files for the three methods are:

  • ./SimulatedAnnealing/source/SIA_s.exe,
  • ./PyTorchGradient/source/main.py,
  • ../opt_orb_pytorch_dpsi/main.py,

respectively.

1. Write input file

Firstly, write the input file, such as ORBITAL_INPUT_DZP in example-directories, for script Generate_Orbital_AllInOne.sh. All three approachs work with the same bash script and use the same input file. Please use absolute path for each file/directory in input file.

2. Set up dependence environment

Secondly, we set up the dependence environment for ABACUS and SIAB, such as:

$ module load hpcx/2.9.0/hpcx-intel-2019.update5 mkl/2019.update5 elpa/2019.05.002/hpcx-intel-2019.update5

Especially for SIAB with PyTorch Gradient approach, we need pytorch v1.1.0.

How to install pytorch:

Take the HanHai20@USTC system for example:

$ module load gcc/7.5.0min      #optional, maybe unnecessary.
$ module load anaconda3_nompi
$ module list
Currently Loaded Modulefiles:
  1) elpa/2019.05.002/hpcx-intel-2019.update5   4) hpcx/2.9.0/hpcx-intel-2019.update5         7) libxc/4.3.4/hpcx-intel-2019.update5
  2) gcc/7.5.0min                               5) mkl/2019.update5
  3) intel/2019.update5                         6) anaconda3_nompi
$ python3 -V
Python 3.7.4

$ conda create -n pytorch110 python=3.7
$ source activate pytorch110    #or: conda activate pytorch110
$ conda install pytorch torchvision torchaudio cpuonly -c pytorch
$ source deactivate             #or: conda deactivate

$ source activate pytorch110  #or: conda activate pytorch110
$ pip3 install --user scipy numpy
$ pip3 install --user torch_optimizer

3. Run generation

Finally, cd into an example folder, and run command like this:

$ ../Generate_Orbital_AllInOne.sh ORBITAL_INPUT_DZP
 or
$ bsub -q idle -n 8 -oo running.log ../Generate_Orbital_AllInOne.sh ORBITAL_INPUT_DZP