Data-efficient modeling of catalytic reactions via enhanced sampling and on-the-fly learning of machine learning potentials

This repository contains the code needed to reproduce the workflow presented in:

S. Perego and L. Bonati, Data-efficient modeling of catalytic reactions via enhanced sampling and on-the-fly learning of machine learning potentials, ChemRxiv, doi:10.26434/chemrxiv-2024-nsp7n

[!NOTE]
We are preparing tutorials to illustrate more generally the different steps of our protocol (including DEAL) beyond the specific examples addressed in the manuscript (which require, e.g., running Quantum Espresso). If you are interested, turn on notifications!

Contents:

├── notebooks           # notebooks describing the different stages of the MLP construction 
│   ├── 0_preliminary
│   ├── 1_exploration
│   └── 2_convergence
├── configs             # configuration files for the different tasks
│   ├── deal
│   ├── flare
│   ├── lammps
│   ├── mace
│   ├── plumed
│   └── qe
└── mlputils            # python module containing the necessary functions

Requirements

The following software and versions have been used:

• ASE (v3.22.1-fix from luigibonati/ase) $\rightarrow$ modified ASE version with small fixes to espresso input/outputs. Note that version 3.23 is not compatible.
• FLARE (v1.3.3-fix from luigibonati/flare) $\rightarrow$ modified FLARE version with small fixes to ensure it work correctly with LAMMPS & PLUMED simulations as well as for the DEAL active learning selection scheme
• MACE (v0.35 from ACEsuit/mace)
• LAMMPS with support of:
  • MACE (patched version from ACEsuit/lammps)
  • FLARE (pair style from mir-group)
  • PLUMED (v2.9 from plumed/plumed2)