| name | molecular-dynamics |
| description | Concept skill for Molecular Dynamics (MD) simulation. Describes MD ensembles, key parameters, and which tool skills to invoke for MD runs. Use this to understand MD concepts before selecting a specific simulation tool. |
| metadata | {"dependent_skills":["ase-deepmd","mattersim"],"tags":["MD","molecular-dynamics","NVT","NPT","simulation","exploration"]} |
Molecular Dynamics (MD) Simulation
Molecular Dynamics simulates the time evolution of a system of atoms by numerically integrating Newton's equations of motion under a given interatomic potential (force field). In the MLFF/PFD context, MD is primarily used for configuration space exploration — generating diverse, physically relevant structures for labeling and training.
Ensembles
| Ensemble | Fixed Variables | Use Case |
|---|
| NVT | N, V, T | Fixed-volume exploration; most common for PFD |
| NPT | N, P, T | Variable-cell exploration; use when volume relaxation matters |
| NVE | N, V, E | Microcanonical; rarely used for exploration |
Key Parameters
| Parameter | Default (PFD) | Description |
|---|
| Temperature | 300 K | Higher T increases structural diversity |
| Timestep | 2 fs | Standard for most solids and liquids |
| Total steps | 500 (≈1 ps) | Minimum for initial exploration |
| Save interval | 50 steps | Frames saved for entropy-based curation |
| Pressure | 0 GPa | NPT only |
When to Use MD in a Workflow
- PFD fine-tuning / distillation: Step 2 — run MD with the current MLFF to generate candidate structures.
- Materials validation: Run short MD to check structural stability at target temperature.
- Property sampling: Generate an ensemble of configurations for property averaging.
Available Tool Skills
| Skill | Best For |
|---|
ase-deepmd | MD with a trained DeePMD potential via ASE |
mattersim | MD with the pre-trained MatterSim universal potential |
Load the appropriate tool skill for step-by-step instructions (e.g., load_skill("ase-deepmd")).
