Skip to content

🔧 Workflow Scripts

Workflow scripts automate repetitive preparation steps for DFT, MD, and NEP development.

Script Location: Scripts/workflow/

Interactive Entry

gpumdkit.sh

Choose:

3) Workflow

The workflow menu looks like:

+---------------------------------------------------------+
|                      WORKFLOW TOOLS                     |
+---------------------------------------------------------+
| 301) SCF batch pretreatment                             |
| 302) MD sample batch pretreatment (gpumd)               |
| 303) MD sample batch pretreatment (lmp)                 |
+---------------------------------------------------------+
| 000) Return to the main menu                            |
+---------------------------------------------------------+
Input the function number:

Available entries:

Menu Workflow Purpose
301 SCF batch pretreatment Prepare single-point DFT calculations
302 MD sample batch pretreatment (GPUMD) Prepare GPUMD sampling calculations
303 MD sample batch pretreatment (LAMMPS) Prepare LAMMPS sampling calculations

SCF Batch Pretreatment

Use this when you have many structures and want to prepare single-point DFT calculations.

VASP

Interactive mode:

gpumdkit.sh
# choose 3 -> 301 -> VASP

Before running, put structure files in the current directory. The script checks files in this order:

  1. if one or more .vasp files exist, it processes .vasp files;
  2. if both .vasp and .xyz files exist, it prints a notice and only processes .vasp files;
  3. if no .vasp file exists and one .xyz file exists, it converts that extxyz file to POSCAR files;
  4. if no .vasp file exists and multiple .xyz files exist, it asks you to choose one.

Typical input directory before running:

current directory/
├── POSCAR_1.vasp
├── POSCAR_2.vasp
└── POSCAR_3.vasp

or:

current directory/
└── sampled.xyz

After running, the script creates:

current directory/
├── struct_fp/
│   ├── POSCAR_1.vasp
│   ├── POSCAR_2.vasp
│   └── ...
├── fp/
├── <prefix>_1/
├── <prefix>_2/
└── presub.sh

You must place INCAR, POTCAR, and KPOINTS in the generated fp/ directory after the script creates it. The calculation folders link to files in fp/.

The prefix you input during the workflow becomes the calculation folder prefix. When out2xyz is later used to extract results, the folder name is stored as config_type in the extxyz file, which makes it easier to trace where each structure came from.

CP2K

CP2K SCF batch pretreatment is available from the workflow menu and can also be called directly:

gpumdkit.sh
# choose 3 -> 301 -> CP2K

python Scripts/workflow/scf_batch_pretreatment_cp2k.py train.xyz template.inp calc

The interactive wrapper asks for:

<extxyz_file> <template.inp> <prefix_name>

Example:

dump.xyz template.inp H2O

The template should be configured to read coordinates from pos.xyz. A reference template is available at:

Scripts/workflow/cp2k_template.inp

The script creates <prefix>_<index>/ directories, writes input.inp, and writes pos.xyz for each structure.

MD Sampling Pretreatment

GPUMD

Interactive mode:

gpumdkit.sh
# choose 3 -> 302

Typical preparation: Before running, put .vasp files or one extxyz file in the current directory. The detection order is the same as the VASP SCF workflow: .vasp files are preferred; if multiple .xyz files are found without .vasp, the script asks you to choose one.

Typical input directory:

current directory/
├── POSCAR_1.vasp
└── POSCAR_2.vasp

or:

current directory/
└── dump.xyz

After running, the script creates:

current directory/
├── struct_md/
│   ├── model_1.xyz
│   ├── model_2.xyz
│   └── ...
├── md/
├── sample_1/
├── sample_2/
└── presub.sh

You must place nep.txt and run_*.in files in the generated md/ directory. The script links run_1.in to sample_1/run.in, run_2.in to sample_2/run.in, and so on.

LAMMPS

Interactive mode:

gpumdkit.sh
# choose 3 -> 303

Typical preparation: Before running, put .vasp files or one extxyz file in the current directory. .vasp files are preferred when both formats exist.

After running, the script creates LAMMPS data files in struct_md/, sample folders, and an md/ directory:

current directory/
├── struct_md/
│   ├── lammps_1.data
│   ├── lammps_2.data
│   └── ...
├── md/
├── sample_1/
├── sample_2/
└── presub.sh

You must place lmprun.in and nep.txt in the generated md/ directory.

Active-Learning Style Workflow

A common NEP improvement loop is:

  1. run MD with the current nep.txt;
  2. collect dump.xyz or active.xyz;
  3. remove unphysical structures;
  4. select diverse or uncertain structures;
  5. run DFT single-point calculations;
  6. convert DFT results to extxyz;
  7. append new structures to train.xyz;
  8. retrain the NEP model.

For the development active-learning script, see Active Learning.

Template-Based Temperature Series

For a simple temperature series, a shell loop is often enough:

for temp in 300 500 700 900; do
    mkdir -p "${temp}K"
    cp model.xyz "${temp}K/"
    cp nep.txt "${temp}K/"
    sed "s/TEMPERATURE/$temp/g" run_template.in > "${temp}K/run.in"
done

Then submit the generated folders according to your local cluster system.

Practical Notes

  • Workflow scripts are cluster- and template-dependent. Always test with a small number of structures first.
  • Keep template files (INCAR, KPOINTS, POTCAR, run.in, CP2K inputs) under version control if possible.
  • Check generated folders before submitting a large batch.
  • Use analyzers and sampling scripts before expensive DFT calculations.