Skip to content

Latest commit

 

History

History
105 lines (79 loc) · 3.33 KB

README.md

File metadata and controls

105 lines (79 loc) · 3.33 KB
Raw

CatFlow

Parts of the code to be open source.

Python package codecov

Machine learning aided catalysis reaction free energy calculation and post-analysis workflow, thus, analyzer for catalysis.

As is known to all, cat is fluid and thus cat flows. 🐱

Former Miko-Analyzer and Miko-Tasker This repository is a temporary branch of original CatFlow. It would be merged into main repo after active refactor.

Analyzer

Installation

To install, clone the repository:

git clone https://github.com/chenggroup/catflow.git

and then install with pip:

cd catflow
pip install .

Acknowledgement

This project is inspired by and built upon the following projects:

  • ai2-kit: A toolkit featured artificial intelligence × ab initio for computational chemistry research.
  • DP-GEN: A concurrent learning platform for the generation of reliable deep learning based potential energy models.
  • ASE: Atomic Simulation Environment.
  • DPDispatcher: Generate and submit HPC jobs.
  • Metadynminer: Reading, analysis and visualization of metadynamics HILLS files produced by Plumed. As well as its Python implementation Metadynminer.py.
  • stringmethod: Python implementation of the string method to compute the minimum energy path.

Tasker

Potential of Mean Force Calculation

A simple workflow designed for free energy calculation from Potential of Mean Force (PMF).

Usage

Commandline

First, prepare a yaml file for workflow settings in detial. For example, config.yaml.

job_config:
  work_path: "/some/place"
  machine_name: "machine_name"
  resources:
    number_node: 1
    cpu_per_node: 1
    gpu_per_node: 1
    queue_name: gpu
    group_size: 1
    module_list:
      - ...
    envs:
      ...
  command: "cp2k.ssmp -i input.inp"

  reaction_pair: [0, 1] # select indexes of atoms who would be constrained
  steps: 10000000 # MD steps
  timestep: 0.5 # unit: fs
  restart_steps: 10000000 # extra steps run in each restart
  dump_freq: 100 # dump frequency
  cell: [24.0, 24.0, 24.0] # set box size for initial structure
  type_map: # should be unified with DeePMD potential
    O: 0
    Pt: 1
  model_path: "/place/of/your/graph.pb"
  backward_files:
    - ...

flow_config:
  coordinates: ... # a list of coordinations to be constrained at
  t_min: 300.0 # under limit of simulation temperature
  cluster_component:
    - Pt # select elements of cluster
  lindemann_n_last_frames: 20000 # use last 20000 steps to judge convergence by calculate Lindemann index
  init_artifact:
    - coordinate: 1.4
      structure_path: "/place/of/your/initial_structure.xyz"
    - coordinate: 3.8
      structure_path: "/place/of/your/initial_structure.cif"
job_type: "dp_pmf" # dp_pmf when using DeePMD

Then, just type command like this:

catflow tasker pmf config.yaml

And enjoy it!