Skip to content

🔧 Workflow

This directory (Scripts/workflow/) contains automation scripts for high-throughput computational workflows, particularly for NEP model development and active learning cycles.

Script Location: Scripts/workflow/

This section covers the workflow automation tools in GPUMDkit (Interactive Mode - Function 3).

Workflow scripts automate repetitive tasks in computational materials research:

  • SCF Batch preprocessin: Set up VASP/CP2K single-point energy calculations
  • MD sampling: Prepare molecular dynamics simulations with GPUMD/LAMMPS

Interactive Mode

gpumdkit.sh
# Select: 3) Workflow

You'll see the following menu:

 ------------>>
 301) SCF batch pretreatment
 302) MD sample batch pretreatment (gpumd)
 303) MD sample batch pretreatment (lmp)
 000) Return to the main menu
 ------------>>

Option 301: SCF batch pretreatment

This script automates the preprocessing of POSCAR or extxyz files for self-consistent field (SCF) calculations. The script includes the following steps:

  1. Converts a .xyz file to POSCAR format using GPUMDkit if no .vasp files are found in the current directory.
  2. Renames and organizes .vasp files into a struct_fp directory.
  3. Creates individual directories for each POSCAR file, setting up symbolic links to the necessary VASP input files.
  4. Generates a presub.sh script to automate running VASP SCF calculations.

Usage

Ensure all .vasp files or a single .xyz file are in the current directory.

Select option 301 from the menu:

301

You will see the following prompt:

 ------------>>
 Input the function number:
 301
 ------------>>
 1) VASP scf batch pretreatment
 2) CP2K scf batch pretreatment
 ------------>>
 Input the function number:

choose 1or 2:

 Starting SCF batch pretreatment...
 Found 8 .vasp files.
 >-------------------------------------------------<
 | This function calls the script in Scripts       |
 | Script: scf_batch_pretreatment.sh               |
 | Developer: Zihan YAN (yanzihan@westlake.edu.cn) |
 >-------------------------------------------------<

 We recommend using the prefix to locate the structure.
 The folder name will be added to the second line of XYZ.
 config_type=<prefix>_<ID>
 ------------>>
 Please enter the prefix of directory (e.g. FAPBI3_iter01)

Enter the prefix of the folder name:

Li7La3Zr2O12_iter01

The script scf_batch_pretreatment.sh in the Scripts will be called to perform the pretreatment.

You will see the following prompts:

 >-----------------------------------------------------<
 ATTENTION: Place POTCAR, KPOINTS and INCAR in 'fp' Dir.
 ATTENTION: Place POTCAR, KPOINTS and INCAR in 'fp' Dir.
 ATTENTION: Place POTCAR, KPOINTS and INCAR in 'fp' Dir.
 >-----------------------------------------------------<

You need to prepare the POTCAR, KPOINTS, and INCAR files and place them in a directory named fp.

Option 302: MD sample batch pretreatment (gpumd)

This script automates the preprocessing of POSCAR or extxyz files for MD sampling using GPUMD.

  1. If .vasp files are found in the current directory, it will convert them to extxyz format to prepare the model.xyz file for GPUMD. If .vasp files are not found, the .xyz file will be read and all frames in it will be split into a individual sample.
  2. Renames and organizes .xyz files into a struct_md directory.
  3. Creates individual directories for each model.xyz file, setting up symbolic links to the necessary GPUMD input files.
  4. Generates a presub.sh script to automate running MD simulations.

Usage

Ensure all .vasp files or a single .xyz file are in the current directory.

Select option 302 from the menu:

302

You will see the following prompt:

 Starting MD sample batch pretreatment...
 No .vasp files found, but found one XYZ file.
 Converting it to model.xyz using GPUMDkit...
 All frames from "NEP-dataset.xyz" have been split into individual model files.
 20 model.xyz files were generated.
 >-------------------------------------------------<
 | This function calls the script in Scripts       |
 | Script: md_sample_batch_pretreatment.sh         |
 | Developer: Zihan YAN (yanzihan@westlake.edu.cn) |
 >-------------------------------------------------<

You will see the following prompts:

 >----------------------------------------------------<
 | ATTENTION: Place run_*.in and nep.txt in 'md' Dir. |
 >----------------------------------------------------<
 | You need to provide MD control parameter files in  |
 | the format run_*.in (e.g., run_1.in, run_2.in),    |
 | each corresponding to a sample (e.g., sample_1,    |
 | sample_2) for molecular dynamics simulations.      |
 >----------------------------------------------------<
 Thank you for using GPUMDkit. Have a great day!

You need to prepare the run.in andnep.txt files and place them in a directory named md.

Option 303: MD sample batch pretreatment (lmp)

The same with option 302.

workflow_active_learning_dev.sh

Implements automated active learning cycles for iterative NEP model improvement.

Purpose: Automates the active learning pipeline:

  1. Generate candidate structures (MD sampling)
  2. Evaluate with current NEP model
  3. Select uncertain structures
  4. Run DFT calculations
  5. Add to training set
  6. Retrain NEP model

Status: Development version - under active development

Documentation: See detailed tutorial in docs/tutorials/

Contributing

To add new workflow capabilities, see CONTRIBUTING.md for detailed guidelines.

Thank you for using GPUMDkit! For questions about workflows or to share your workflow adaptations, please open an issue on our GitHub repository or contact Zihan YAN (yanzihan@westlake.edu.cn).