Python Rapid Artificial Intelligence Ab Initio Molecular Dynamics

                              /\
   |\    /|                  /++\
   ||\  /||                 /++++\
   || \/ || ||             /++++++\
   ||    || ||            /PyRAI2MD\
   ||    || ||           /++++++++++\                    __
            ||          /++++++++++++\    |\ |  /\  |\/| | \
            ||__ __    *==============*   | \| /--\ |  | |_/

                          Python Rapid
                     Artificial Intelligence
                  Ab Initio Molecular Dynamics



                      Author @Jingbai Li
               Northeastern University, Boston, USA

                          version:   2.0 alpha
                          

  With contriutions from (in alphabetic order):
    Jingbai Li                 - Fewest switches surface hopping
                                 Zhu-Nakamura surface hopping
                                 Velocity Verlet
                                 OpenMolcas interface
                                 OpenMolcas/Tinker interface
                                 BAGEL interface
                                 Adaptive sampling
                                 Grid search
                                 Two-layer ONIOM (coming soon)
                                 Periodic boundary condition (coming soon)
                                 QC/ML hybrid NAMD

    Patrick Reiser             - Neural networks (pyNNsMD)

  Special acknowledgement to:
    Steven A. Lopez            - Project directorship
    Pascal Friederich          - ML directoriship>

Features

• Machine learning nonadibatic molecular dyanmics (ML-NAMD).
• Neural network training and grid search.
• Active learning with ML-NAMD trajectories.
• Support BAGEL, Molcas for QM, and Molcas/Tinker for QM/MM calculations.
• Support nonadibatic coupling and spin-orbit coupling (Molcas only)

Prerequisite

• Python >=3.7 PyRAI2MD is written and tested in Python 3.7.4. Older version of Python is not tested and might not be working properly.
• TensorFlow >=2.2 TensorFlow/Keras API is required to load the trained NN models and predict energy and force.
• Cython PyRAI2MD uses Cython library for efficient surface hopping calculation.
• Matplotlib/Numpy Scientifc graphing and numerical library for plotting training statistic and array manipulation.

Content

 File/Folder Name                                  Description                                      
---------------------------------------------------------------------------------------------------
 pyrai2md.py                                       PyRAI2MD interface                              
 PyRAI2MD                                          source codes folder
  |--variables.py                                  PyRAI2MD input reader                           
  |--method.py                                     PyRAI2MD method manager                         
  |--Molecule                                      atom, molecule, trajectory code folder
  |   |--atom.py                                   atomic properties class                         
  |   |--molecule.py                               molecular properties class                      
  |   |--trajectory.py                             trajectory properties class                     
  |   |--pbc_helper.py                             periodic boundary condition functions           
  |    `-qmmm_helper.py                            qmmm functions                                  
  |
  |--Quantum_Chemistry                             quantum chemicial program interface folder
  |   |--qc_molcas.py                              OpenMolcas interface                            
  |   |--qc_bagel.py                               BAGEL interface                                 
  |    `-qc_molcas_tinker                          OpenMolcas/Tinker interface                     
  |
  |--Machine_Learning                              machine learning library interface folder
  |   |--training_data.py                          training data manager                           
  |   |--model_NN.py                               neural network interface                        
  |   |--hypernn.py                                hyperparameter manager                          
  |   |--permutation.py                            data permutation functions                      
  |   |--adaptive_sampling.py                      adaptive sampling class                         
  |   |--grid_search.py                            grid search class                               
  |   |--remote_train.py                           distribute remote training                      
  |    `-pyNNsMD                                   neural network library                         
  |
  |--Dynamics                                      ab initio molecular dynamics code folder
  |   |--aimd.py                                   molecular dynamics class                        
  |   |--mixaimd.py                                ML-QC hybrid molecular dynamics class           
  |   |--single_point.py                           single point calculation                        
  |   |--hop_probability.py                        surface hopping probability calculation         
  |   |--reset_velocity.py                         velocity adjustment functions                   
  |   |--verlet.py                                 velocity verlet method                          
  |   |--Ensembles                                 thermodynamics control code folder
  |   |   |--ensemble.py                           thermodynamics ensemble manager                 
  |   |   |--microcanonical.py                     microcanonical ensemble                         
  |   |    `-thermostat.py                         canonical ensemble                              
  |   |
  |    `-Propagators                               electronic propagation code folder
  |       |--surface_hopping.py                    surface hopping manager                         
  |       |--fssh.pyx                              fewest switches surface hopping method          
  |       |--gsh.py                                generalized surface hopping method              
  |        `-tsh_helper.py                         trajectory surface hopping tools                
  |
   `-Utils                                         utility folder
      |--aligngeom.py                              geometry aligment and comparison functions      
      |--coordinates.py                            coordinates writing functions                   
      |--read_tools.py                             index reader                                    
      |--bonds.py                                  bond length library                            
      |--sampling.py                               initial condition sampling functions            
      |--timing.py                                 timing functions                                
       `-logo.py                                   logo and credits                                    

Installation

Download the repository

git clone https://github.com/lopez-lab/PyRAI2MD.git

Specify environment variable of PyRAI2MD

export PYRAI2MD=/path/to/PyRAI2MD

Test PyRAI2MD

Copy the test script and modify environment variables

cp $PYRAI2MD/Tool/test_PyRAI2MD.sh .
bash test_PyRAI2MD.sh

Or directly run if environment variables are set

$PYRAI2MD/pyrai2md.py quicktest

Run PyRAI2MD

$PYRAI2MD/pyrai2md.py input

User manual

We are currently working on the user manual.

Cite us

• Jingbai Li, Patrick Reiser, Benjamin R. Boswell, André Eberhard, Noah Z. Burns, Pascal Friederich, and Steven A. Lopez, "Automatic discovery of photoisomerization mechanisms with nanosecond machine learning photodynamics simulations", Chem. Sci. 2021. DOI: 10.1039/D0SC05610C
• Jingbai Li, Rachel Stein, Daniel Adrion, Steven A. Lopez, "Machine-learning photodynamics simulations uncover the role of substituent effects on the photochemical formation of cubanes", ChemRxiv, preprint, DOI:10.33774/chemrxiv-2021-lxsjk