| name | pfd-finetuning |
| description | Iterative workflow that fine-tunes pre-trained ML force fields using DFT-labeled data via active learning. |
| metadata | {"dependent_skills":["machine-learning-force-field","dft-calculation","molecular-dynamics","atomic-structure","deepmd","abacus"],"tags":["iterative","active-learning","fine-tune"]} |
Coordinate iterative fine-tuning of ML force fields from a pre-trained model.
Standard loop:
- Structure building
- MD exploration
- Entropy-based data curation
- DFT Labeling
- Training
- Convergence check
- Repeat until criteria or max iterations.
Key parameters to confirm:
- max iterations (default:1), convergence criterion (default: RMSE_energy = 5 meV/atom)
- initial structure, supercell size (default: 1 x 1 x 1), number of perturbations (default: 1), perturbation settings.
- MD ensemble (default: NVT/NPT), temperature (default: 300 K), timestep (default: 2 fs), steps (default: 1 ps), save interval (default: 50), pressure (for NPT ensemble, default: 0 GPa)
- curation max_sel(number of frames sent for DFT calculation, default:30), chunk_size(default: 10)
- DFT parameters: kspacing (default: 0.14 Angstrom)
- training epochs (default: 100) and train/test split (default: 0.9/0.1)
