CARE: Catalysis Automated Reaction Evaluator

CARE (Catalysis Automated Reaction Evaluator) is a tool for generating and manipulating chemical reaction networks (CRNs) on catalytic surfaces. CARE is powered with GAME-Net-UQ, a graph neural network with uncertainty quantification targeting the DFT energy of relaxed species and transition states.

🪛 Installation

Installing CARE requires Conda and Git locally installed. The following instructions are optimized to install CARE on Linux systems, while for macOS we noticed a lower performance in the CRN generation mainly due to Python multiprocessing (see Contexts and start methods in the documentation)

1. Clone the repo:

git clone git@github.com:LopezGroup-ICIQ/care.git

2. Create a conda environment with Python 3.11 and activate it:

conda create -n care_env python=3.11
conda activate care_env

3. Enter the repo and install the package with pip:

cd care
python3 -m pip install .

NOTE: MacOS users might need to launch a new shell at this point in order for the entry points to work correctly.

4. Install pytorch and pytorch_geometric through conda. Ensure that the libraries are obtained from the pytorch and pyg channels, as shown here:

conda install pytorch cpuonly pytorch-scatter pytorch-sparse pyg -c pytorch -c pyg

NOTE: MacOS users might need to install pytorch geometric using pip.

5. (optional) Install Julia and the ODE packages required to perform microkinetic simulations. As alternative, simulations can run with the implemented Scipy solver.

curl -fsSL https://install.julialang.org | sh
python3 -m pip install juliacall  # Python-Julia bridge
julia -e 'import Pkg; Pkg.add("DifferentialEquations"); Pkg.add("DiffEqGPU"); Pkg.add("CUDA");'

NOTE: For some systems Julia may present some error while using sh. If that is the case, please install Julia by running instead:

curl -fsSL https://install.julialang.org | sh -s -- -y

6. (optional) Install the different evaluators available (MACE, fairchem), through the following command:

python3 -m pip install fairchem-core==1.1.0
python3 -m pip install mace-torch torch-dftd

NOTE: There currently is a dependency clash during installation for the two evaluators related to the e3nn library (see: this issue for MACE). Installation might result in an incompatibility warning, but both libraries should work correctly if the installation order shown below is followed.

💥 Usage

Blueprint generation

The blueprint can be constructed in two ways, by providing (i) the network carbon and oxygen cutoffs ncc and noc, or (ii) the chemical space as list of SMILES.

gen_crn_blueprint -h  # documentation
gen_crn_blueprint -ncc 2 -noc 1 -o output_name  # Example from ncc and noc
gen_crn_blueprint -cs "CCO" "C(CO)O" -o output_name # Example from user-defined chemical space

The CRN blueprint is stored as pickle file. To access the blueprint, do:

from pickle import load

with open('path_to_blueprint_file', 'rb') as f:
    intermediates, reactions = load(f)

Evaluation of intermediate and reaction properties

The available energy evaluators are GAME-Net-UQ, OCP models, and MACE-MP potentials.

eval_crn -h  # documentation
eval_crn [-i INPUT] [-bp BP] [-o OUTPUT] [-ncpu NUM_CPU]

This script requires an input toml file defining the material/surface of interest, the model of choice and its settings. The output is a ReactionNetwork object stored as pickle file. You can find examples of input files here.

Microkinetic simulation

run_kinetic [-i INPUT] [-crn CRN] [-o OUTPUT]

This script runs microkinetic simulation starting from the evaluated reaction network and an input toml file defining the reaction conditions, solver, inlet conditions. The results are stored as a pickle object file.

Run all together

You can run the entire pipeline (blueprint generation -> properties evaluation -> kinetic simulation) running the care_run script:

care_run -h  # documentation
care_run -i input.toml -o output_name

This will generate a directory output_name containing a crn.pkl with the generated reaction network. Examples of input .toml files can be found here.

📖 Tutorials

We currently provide three tutorials, available in the notebooks directory:

• CRN generation and manipulation
  
• Energy evaluator interface implementation
  
• Microkinetic simulations

❗️Notes

The DFT database in ASE format used to retrieve available CRN intermediates will be uploaded soon in Zenodo.

✒️ License

The code is released under the MIT license.

📜 Reference

• A Foundational Model for Reaction Networks on Metal Surfaces Authors: S. Morandi, O. Loveday, T. Renningholtz, S. Pablo-García, R. A. Vargas Hernáńdez, R. R. Seemakurthi, P. Sanz Berman, R. García-Muelas, A. Aspuru-Guzik, and N. López DOI: 10.26434/chemrxiv-2024-bfv3d