README.md

This repo is archieved along with AIMS_Tools. Please check ms-tools for latest development.

ffoptimizer

Automatically optimize the LJ parameters for TEAM Force Field against experimental data of density and enthalpy of vaporization.
AIMS_Tools, DFF, Packmol and GROMACS are required.

Steps

1. Configure environment in config.py

• Specify paths for required packages. The version of mstools should be 0.1.

  MS_TOOLS_DIR = '/home/gongzheng/GitHub/AIMS_Tools'
  PACKMOL_BIN = '/share/apps/tools/packmol'
  DFF_ROOT = '/home/gongzheng/apps/DFF/Developing'

• Specify the DFF table for assigning atom types

  DFF_TABLE = 'MGI'

• Specify the Slurm partitions and corresponding GROMACS binaries. The default in config.py is good for gtx partition.

2. Initialize optimization

• Prepare data file which contains SMILES and experimental density and Hvap data. An example is provided as example_LJ/data.txt. For better GPU performance, make sure that there are even lines in expt data file.
• Prepare PPF file which contains initial parameters. The parameters to be optimized should be unfrozen. An example is provided as example_LJ/initial.ppf.
• Prepare an empty directory WORKDIR for running simulation
• Init

  cd example_LJ
  ./run.py init task_name data.txt initial.ppf WORKDIR

3. Start optimization

  ./run.py optimize task_name

Examples

Three examples are provided for optimized LJ, temperature-dependent LJ and dihedral parameters

1. Optimize LJ-12-6 parameters: example_LJ

Three files are required

• data.txt lists the molecules and experimental data and their weight for optimization. The names of molecules are arbitrary but should only contain alphabets and numbers.
• initial.ppf is the initial parameters exported from DFF. The N12_6 lines for c_4h3, c_4h2 and h_1 are unfrozen so that the epsilon and sigma for these three atom types will be optimized.
• run.py is the controlling script. No modification is required for this script.

Start optimization by

  mkdir WORKDIR
  ./run.py init LJ data.txt initial.ppf WORKDIR
  ./run.py optimize LJ

Check the generated log file. Make sure RSQ is decreasing and new parameters are reasonable. It usually converges in less than 6 iterations.

2. Optimize temperature-dependent LJ-12-6 parameters: example_LJ_T

data.txt and initial.pff are the same as previous example. Modifications should be made in run.py to optimize temperature parameters.

• Two variables drde_dict and drde_atoms should be specified in run.py to optimize the temperature parameter \lambda for different atom types.
• optimizer.drde_dict lists temperature parameters that will be fixed during the optimization.
• optimizer.optimize(drde_atoms=...) lists temperature parameters that are subject to optimization.
• the \lambda parameters are always named as xxx_dl. It will match all atom types starting with xxx. For example, c_4_dl is the \lambda parameter for c_4h2, c_4h3, c_4o...

Note that

• If a parameter appears in both drde_dict and drde_atoms, it will be subject to optimization.
• It is better to have at most one free \lambda parameter to get rid of coupling between parameters.

Start optimization and check the log file. Make sure RSQ is decreasing and new parameters are reasonable. The RSQ should be smaller than previous example because of the introduction of temperature-dependence.

3. Optimize temperature-dependent LJ-12-6 parameters together with dihedral: example_LJ_T_dihedral

data.txt is the same as previous example. Modifications should be made in initial.pff and run.py to optimize dihedrals. The dihedrals are fitted to QM energy surface by using DFF.

• Unfreeze the dihedral parameters in initial.ppf so that they can be optimized.
• Variable torsions lists the dihedrals subject to optimization.
• Corresponding MSD and QMD files should be provided.

Note that

• Dihedral fitting here only make sense for the backbone of linear molecules. For example, alkanes, diphenyl...

Start optimization and check the log file. Make sure RSQ is decreasing and new parameters are reasonable. Check the dft file to make sure dihedrals are correctly fitted.