Acetate Walkthrough

Source files:

• examples/acetate/

Goal

The acetate example optimizes acetate partial charges using three systems:

• aqueous acetate
• acetate with calcium at contact distance
• acetate with calcium in a solvent-shared configuration

Stage Order

cd examples/acetate/01-prepare/colvars
bff prepare config.yaml

cd ../../02-reference-data
bff reference config-local.yaml

cd ../02-training-data
bff trainset config-local.yaml

cd ../03-qoi
bff qoi config.yaml

cd ../04-train-lgp
bff train config.yaml

cd ../05-learning
bff learn config.yaml

cd ../07-validate
bff validate config.yaml

Important Files

• prepare stage overview: 01-prepare/README.md
• shared prepare assets: 01-prepare/common/
• Colvars prepare config: 01-prepare/colvars/config.yaml
• local reference config: 02-reference-data/config-local.yaml
• reference CP2K overrides and trajectories: 02-reference-data/inputs/, 02-reference-data/trajectories/
• PLUMED prepare config: 01-prepare/plumed/config.yaml
• local force-field trajectory config: 02-training-data/config-local.yaml
• QoI config: 03-qoi/config.yaml
• train config: 04-train-lgp/config.yaml
• learn config: 05-learning/config.yaml
• validate config: 07-validate/config.yaml

Notebooks

• 05-learning/interactive.ipynb walks through surrogate training, posterior sampling, posterior sample export, and basic posterior plots interactively.
• 06-visualize/visualize.ipynb focuses on plotting and inspection only.

Notes

• The walkthrough uses the Colvars preparation variant by default.
• Shared topologies, force-field files, template coordinates, and MDP inputs live in 01-prepare/common/.
• bff reference consumes the staged CP2K reference inputs under 01-prepare/colvars/reference/system-XXX/ and writes runnable outputs plus collected extxyz files under 02-reference-data/reference-assets/system-XXX/.
• 02-reference-data/inputs/ contains optional CP2K input overrides, and 02-reference-data/trajectories/ contains the ab initio reference trajectories used by QoI.
• 02-training-data/ contains force-field trajectory campaign configs and generated trainset outputs.
• Prepared training assets live under 01-prepare/colvars/ace-colvars/training/system-XXX/.
• The QoI step demonstrates both builtin routines and a custom routine loaded from ./restraint.py:distance_distribution.
• The learn stage writes posterior chains in 05-learning/ and the interactive notebook can also export posterior-samples.yaml for 07-validate.