Usage of Automatic Force Field Generation (Self-learning Hybrid Monte Carlo Method)

Sequence of Use

1. Initial Structure Creation

One initial structure is created by the user using Advance/NanoLabo, etc.

2. Automatic generation of force field

Based on the initial structure, the generation of training data and the learning of the force field are automatically proceeded.

3. Molecular dynamics calculation with LAMMPS

Molecular dynamics calculations are performed using the created force field file. LAMMPS, to which we have added Neural Network Potential functionality, is used for the calculation.

Explanation

Step 1. Initial Structure Creation

Prepare the atomic structure of the sample that would like to be used as training data, and create an input file for SCF calculations of Quantum ESPRESSO based on it. Our product Advance/NanoLabo can easily generate an input file from atomic structure files such as cif.

The path to the pseudopotential in the file must be the one used for the actual calculation in step 2. For example, when running the calculation on a calculation server, the path must point to the pseudopotential directory in the calculation server.

Once the input file is prepared, add the following settings to have the necessary information output.

&CONTROL
   tprnfor = .true.
   tsannp  = .true.
   disk_io = 'none'
/
&SYSTEM
   nosym   = .true.
/

Save the created input file with the file name espresso.in.

You can also create the file by editing espresso.in output by slhmc --template without creating it from scratch.

Step 2. Automatic Generation of Force Field

When using Quantum ESPRESSO bundled with NanoLabo Tool, setup of the library search path and others are required. Please refer to the following for the settings.

• Calculation Engines included in Nanolabo Tool
  • Windows version
  • Linux version
  • GPU version of LAMMPS

The automatic generation of the force field proceeds by calling Quantum ESPRESSO, LAMMPS, and NeuralMD sequentially from the executable slhmc. Therefore, first set up the system so that these can be executed.

slhmc --template

Execute the above to output the template of the required configuration file.

slhmc.prop

Configuration file for the self-learning Hybrid Monte Carlo method. Please refer to the following for the settings.

• Format of slhmc.prop

espresso.sh, lammps.sh, sannp.sh(.bat for Windows)

Execution files for Quantum ESPRESSO, LAMMPS, and NeuralMD. These files will be executed from slhmc. If necessary, add the environment variable settings required for execution (path: PATH, LD_LIBRARY_PATH, license: ADVANCED_LICENSE_FILE), parallel execution settings (command: mpirun, OpenMP parallel number: OMP_NUM_THREADS), etc. Do not change the input/output file names written in the file, which are fixed.

Example for MPI parallel execution in espresso.sh

OMP_NUM_THREADS=1
mpirun -n 4 pw.x -in espresso.in 1> espresso.out 2> espresso.err

Note

NeuralMD of Windows version does not support MPI parallel execution while OpenMP parallel is available.

sannp.prop

NeuralMD configuration file. Refer to the Format of sannp.prop to edit as needed.

When calculating the charge, set withCharge 1 in sannp.prop and uncomment out the line of sannp --train-charge in sannp.sh.

When using Δ-NNP(LJ-like force field), set withLJlike 1 in sannp.prop and uncomment out the line of sannp --classical in sannp.sh.

When using Δ-NNP(ReaxFF), set withReaxFF 1 in sannp.prop and place the parameter definition file ffield.reax in the same folder.

After preparation, place espresso.in created in step 1 in the same folder and run

slhmc --calc

to initiate force field creation.

Note

Parallel execution of slhmc itself is not available. Set up parallel execution in each sh(bat) file for Quantum ESPRESSO, LAMMPS, and NerualMD.

When using a job management system, make a job script to run slhmc --calc and post the job. In this case, too, do not parallel execute slhmc itself but set up parallel execution in each sh file.

Hint

An already created force field is available as the initial force field alternatively while in default that based on the first-principles molecular dynamics calculations is used which is performed at first. Set initialTrain 0 in slhmc.prop, place the force field file ffield.sannp in the same folder, and then run slhmc.

Hint

To perform calculations using GPU, replace sannp with sannp_gpu in sannp.sh , and lammps with lammps_gpu in lammps.sh . Also, if you place configuration files for GPU sannp.mpi2gpu , gpu.conf in the same folder, the settings are applied at execution.

Running slhmc --stop aborts the force field creation. Setting restart 1 in slhmc.prop and running slhmc --calc again resumes the force field creation from where it is interrupted.

During execution, the force field file field.sannp is output progressively when the force field is updated. Also, the history of the structure is output as slhmc.xyz.

The input/output files for each calculation engine are stored in the subfolder slhmc_dat.

If the calculation does not proceed due to errors etc, check the file slhmc.CRASH in which input/output files of relevant calculation engine are saved.

Step 3. Molecular Dynamics Calculations with LAMMPS

Run molecular dynamics calculations with LAMMPS using the output force field file field.sannp. See Step 5. Molecular Dynamics Calculations with LAMMPS on the Basic Usage page.

Hint

To create a force field using several different initial structures with a self-learning hybrid Monte Carlo method, first perform up to step 2 for each initial structure. Then, combine the resultant training data sannp.train and use it to train the neural network again (see Step 4. Training of Neural Network (Optimization) on the Basic Usage page).