Prepare Configuration¶

Source code:

Purpose¶

bff prepare stages everything needed for downstream workflows:

equilibrated GROMACS systems
reusable training assets under training/system-XXX/
CP2K reference assets under reference/system-XXX/

Minimal Example¶

project:
  directory: ./ace-colvars
  log: ./out.log

gromacs:
  command: gmx

defaults:
  nsteps:
    npt: 0
    prod: 100000

reference:
  n_single_point_snapshots: 1000

systems:
  - topology: ./topol.top
    templates:
      ACE: ./ace.gro
    mdp:
      em: ../../../data/mdp/em.mdp
      npt: ../../../data/mdp/npt.mdp
      prod: ../../../data/mdp/nvt.mdp
    charge: -1
    multiplicity: 1
    box: [15.7107, 15.7107, 15.7107, 90, 90, 90]

Top-Level Keys¶

project Project directory output. A string is accepted as shorthand for project.directory.
project.directory Output directory for equilibration/, training/, and reference/.
project.log Optional workflow log file.
gromacs.command GROMACS executable, usually gmx.
defaults.nsteps.npt Default NpT equilibration length for systems that do not override it.
defaults.nsteps.prod Default production-preparation run length for systems that do not override it.
reference.n_single_point_snapshots Number of evenly spaced snapshots staged for CP2K single-point and short-MD reference jobs.
systems Non-empty list of prepared systems.

`systems[]` Keys¶

topology GROMACS topology describing residue counts.
templates Mapping from residue name to coordinate template file for non-standard residues.
charge Total system charge for staged CP2K reference inputs.
multiplicity Spin multiplicity for staged CP2K reference inputs.
box Optional box dimensions. Accepts 3 values or full 6-value triclinic format.
bias Optional bias specification. Use either plumed_file or colvars_file.
nsteps.npt Optional per-system NpT override.
nsteps.prod Optional per-system production-preparation override.
mdp.em Energy minimization MDP file.
mdp.npt NpT equilibration MDP file.
mdp.prod Production MDP file staged into the downstream training assets.

Outputs¶

The most important output for downstream BFF workflows is:

PROJECT/training/system-XXX/

Each such directory contains one prepared system with:

system-XXX.top
system-XXX.gro
system-XXX.ndx
system-XXX.em.mdp
system-XXX.npt.mdp
system-XXX.mdp
optional bias file

Those directories are consumed directly by both trainset and validate.

The reference tree is system-centered:

PROJECT/reference/system-XXX/system.gro
PROJECT/reference/system-XXX/system.top
PROJECT/reference/system-XXX/system.xyz
PROJECT/reference/system-XXX/md/ CP2K direct-MD input files.
PROJECT/reference/system-XXX/single-atoms/ One isolated-atom CP2K input.inp and centered pos.xyz per unique element in the prepared system. These jobs use a neutral atom in a nonperiodic 10 Å cubic box with the same revPBE-based CP2K setup as the staged MD inputs.
PROJECT/reference/system-XXX/snapshots/ Decorrelated XYZ snapshots, sp.inp, and a 100-step md.inp.

bff prepare only stages these reusable reference inputs. The runnable output tree, per-snapshot runs/ directories, optional Slurm staging, and final train.extxyz / valid.extxyz split are created later by bff reference, typically from a separate 02-reference-data/ example stage.