Machine Learning-Enabled Development of Accurate Force Fields for Refrigerants

Authors: Ning Wang, Montana Carlozo, Eliseo Marin-Rimoldi, Bridgette Belfort, Alexander W. Dowling, and Edward J. Maginn

Introduction

HFC-FFO is a repository used to rapidly calibrate the LJ parameters of HFC forcefields given experimental data. The key feature of this work is using machine learning tools in the form of Gaussian processes (GPs) which allow us to cheaply estimate the results of a molecular simulation given temperature state points and thermophysical property data.

Citation

This work has been published on JCTC, whose link is https://doi.org/10.1021/acs.jctc.3c00338. Please cite as:

Ning Wang, Montana N. Carlozo, Eliseo Marin-Rimoldi, Bridgette J. Befort, Alexander W. Dowling, and Edward J. Maginn*, “Machine Learning-Enabled Development of Accurate Force Fields for Refrigerants”, J. Chem. Theory Comput., 2023, 19, 14, 4546–4558

Available Data

HFC Parameter Sets

The non-dominated and best parameter sets for each HFC are provided under HFC-FFO/rXX/analysis/csv/. Where XX represents a different HFC. For example r-143a, r-14, r-170. The non-dominated sets are found in rXX-pareto.csv, and the best sets are found in rXX-final.csv. The parameter values in the CSV files are normalized between 0 and 1 based on the parameter bounds for each atom type (see manuscript, or HFC-FFO/rXX/analysis/utils/rXX.py for definitions of the upper and lower parameter bounds for each refrigerant).

Molecular Simulations Inputs

All molecular simulations were performed inside HFC-FFO/r##/runs where it exists. Each iteration was managed with signac-flow. Inside of each directory in runs, you will find all the necessary files to run the simulations. Note that you may not get the exact same simulation results due to differences in software versions, random seeds, etc. Nonetheless, all of the results from our molecular simulations are saved under HFC-FFO/analysis/csv/rXX-YY-iterZZ-results.csv, where XX is the molecule, YY is the stage (liquid density or VLE), and ZZ is the iteration number.

Surrogate Modeling Analysis

All of the scripts for the surrogate modeling are provided in HFC-FFO/r##/analysis, following the same naming structure as the CSV files.

Figures

All scripts required to generate the primary figures in the manuscript are reported under HFC-FFO/r##/analysis/final-figs and the associated PDF files are located under HFC-FFO/r##/analysis/final-figs/pdfs

Installation

To run this software, you must have access to all packages in the hfcs-fffit environment (hfcs-fffit.yml) which can be installed using the instructions in the next section.

This package has a number of requirements that can be installed in different ways. We recommend using a conda environment to manage most of the installation and dependencies. However, some items will need to be installed from source or pip.

Running the simulations will also require an installation of GROMACS. This can be installed separately (see installation instructions here <https://manual.gromacs.org/documentation/2021.2/install-guide/index.html>).

An example of the procedure is provided below:

# First clone hfcs-fffit and install pip/conda available dependencies
# with a new conda environment named hfcs-fffit
git clone https://github.com/emarinri/hfcs-fffit.git
git switch -c update-dependencies origin/update-dependencies
cd hfcs-fffit
conda env create -f ./devtools/conda-envs/hfcs-fffit.yaml
conda activate hfcs-fffit
cd ../

# Now clone and install  other dependencies
git clone git@github.com:dowlinglab/fffit.git
# Checkout the v0.1 release of fffit and install
cd fffit/
git checkout tags/v0.1
pip install .
cd ../
# Checkout the v0.1 release of block average and install
git clone git@github.com:rsdefever/block_average.git
cd block_average/
git checkout tags/v0.1
pip install .
cd ../

Usage

Liquid Density Optimization

NOTE: We use Signac and signac flow (<https://signac.io/>) to manage the setup and execution of the molecular simulations. These instructions assume a working knowledge of that software.

The first iteration of the liquid density simulations was performed under HFC-FFO/r##/runs/rXX-density-iter1/.

To run liquid density iterations, follow the following steps:

1. Create the initial configuration
   • Prepare rXX_gaff.xml
   • Go to the data folder and use the run.sh file

     conda activate hfcs-fffit
     cd HFC-FFO/rXX/run/rXX-density-iter1/data
     source run.sh
   

2. Initialize Signac workflow
   • Leave ''HFC-FFO/rXX/run/rXX-density-iter1/data'' untouched
   • Initialize files for simulation use

    cd HFC-FFO/rXX/run/rXX-density-iter1/
    python init.py
   

3. Check status a few times throughout the process

     python project.py status
   

4. Create force fields and generate inputs

     python project.py run -o create_forcefield
     python project.py run -o generate_inputs
   

5. Create systems
   • Note: rm -r workspace/ signac_project_document.json signac.rc will remove everything and allow you to start fresh if you mess up

     python project.py run -o create_system
   

6. Fix topology

     python project.py run -o fix_topology    
   

7. Run simulation

     python project.py submit -o simulate --bundle=24 --parallel
   

8. Calculate density

     python project.py submit -o calculate_density --bundle=24 --parallel
   

9. Extract density using the following after each LD iteration in the analysis/ folder

     python extract_rXX_density.py ZZ
   

10. Run GP optimization and get samples for the next iteration in the analysis/ folder

  module load gcc/11.2.0
  python id-new-samples.py
  python plotfig_gp_examples.py 

VLE Optimization

To run vapor-liquid-equilibrium iterations, follow the following steps:

1. Use analysis/csv/rXX-vle-iter1-params.csv to initialize files for simulation use

     cd HFC-FFO/rXX/run/rXX-vle-iter1/
     python init.py
   

2. Check status a few times throughout the process

     python project.py status 
   

3. Create force fields

     python project.py run -o create_forcefield
   

4. Calculate vapor/liquid box size

     python project.py run -o calc_vapboxl
     python project.py run -o calc_liqboxl
   

5. Run simulation

     python project.py submit -o equilibrate_liqbox --bundle=12 --parallel
     python project.py run -o extract_final_liqbox
     python project.py submit -o run_gemc --bundle=12 --parallel
   

6. Calculate VLE Properties

     python project.py run -o calculate_props
   

7. Extract VLE properties using the following after each VLE iteration in the analysis/ folder

     python extract_rXX_vle.py ZZ
   

8. Analyze Data

     module load gcc/11.2.0
     python id-new-samples.py
     python get-new-samples.py 
     python analysis.py 
   

Final Analysis

The nondominated parameter sets and final processing steps can be run using the following:

1. Summarize data

     cd HFC-FFO/rXX/run/rXX-vle-iterKK
     python id-pareto.py
     cd HFC-FFO/rXX/analysis/final-analysis
     python select_final.py
   

2. Plots in the paper were generated by codes in HFC-FFO/rXX/analysis/final-figs/

Known Issues

The instructions outlined above seem to be system-dependent. In some cases, users have the following error:

ImportError: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.29' not found

If you observe this, please try the following in the terminal

export LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH

which should fix the problem. This is not an optimal solution and is something we would like to address. We found that related projects 1, 2 have similar issues. If you are aware of a robust solution to this issue, please let us know by raising an issue or sending an email!

Credits

This work is funded by the National Science Foundation, EFRI DChem: Next-generation Low Global Warming Refrigerants, Award no. 2029354, and uses the computing resources provided by the Center for Research Computing (CRC) at the University of Notre Dame. The authors would like to thank Bridgette Befort as her work is used as the basis of this method.

Contact

Please contact Ning Wang (nwang2@nd.edu), Eliseo Marin Rimoldi (emarinri@nd.edu), or Montana Carlozo (mcarlozo@nd.edu) with any questions, suggestions, or issues.