q734738781/md-diffusion-analysis

md-diffusion-analysis

Overview

Use this skill to prepare a VASP MD stage, dispatch it, and summarize the resulting trajectory with MSD, RDF, and diffusion-fit artifacts. Do not use it for a generic VASP run log, one-off thermalization checks, or when the trajectory is too short to support diffusion claims. For MACE MD sampling, delegate to dynamics_worker and mace-md-sampling.

Quick Start

1. Confirm this is a VASP MD run. If it is MACE MD, hand off to dynamics_worker.
2. Prepare inputs with vasp_prepare(preset="md", ...) and make controls explicit through user_incar_patch.
3. Submit the prepared VASP stage with managed remote execution.
4. Decide what part of the trajectory is equilibration and what part is production.
5. Analyze the collected trajectory with analyze_trajectory.

Allowed tools

• vasp_prepare
• remote_submission
• remote_submission_batch
• analyze_trajectory

Workflow

1. Use the MD preset as a starter template

• Keep the default Nose-Hoover/NVT starter only when it matches the scientific question.
• Use user_incar_patch in the same vasp_prepare call to set the actual timestep, thermostat, or temperature schedule needed for the run.
• Surface the intended ensemble semantics explicitly; do not leave the reader guessing whether the run is being interpreted as NVT-like sampling, annealing, or another protocol.

For MACE MD, do not run from this materials-side skill. The dynamics-side mace-md-sampling skill owns mace_md_dir, grouped MD controls, trajectory output, and trajectory analysis.

2. Separate equilibration from production

• Report how much trajectory is being discarded as warmup before any diffusion fit.
• If no equilibration window is removed, say that explicitly instead of implying a production-only trajectory.

3. Keep trajectory provenance visible

• Report the exact MD control overrides, not just “ran MD”.
• Keep the output root clean so the trajectory analyzer sees one clear result directory.

4. Analyze before interpreting

• analyze_trajectory emits MSD, RDF, and any temperature/energy time-series it can recover.
• Do not claim diffusion coefficients if the fit window or sampled trajectory length is obviously inadequate.
• If diffusion is species-specific or anisotropic, set the species filter and diffusion dimension explicitly.

5. Report finite-time limitations

• If the MSD never reaches a clear diffusive regime, report that the run is qualitative only.
• Do not compare diffusion coefficients across runs that changed ensemble, temperature, composition, or cell size silently.

Method-critical defaults

• MD controls are not one-size-fits-all; always surface the patched thermostat and timestep knobs when they affect interpretation.
• If diffusion is species-specific, pass the species filter explicitly at analysis time and keep rdf_species / diffusion_dimension aligned with the scientific question.
• Do not report a diffusion coefficient just because the analyzer can fit one numerically; the fit still needs a physically credible production window.
• Keep ensemble, temperature schedule, equilibration cut, and production length visible in the final summary.

Output Contract

Return:

• MD input/output root
• key MD control overrides
• chosen ensemble interpretation and equilibration cut
• trajectory summary JSON
• MSD/RDF artifact paths
• whether a diffusion coefficient is being reported or withheld as unreliable

References

• The canonical one-call customization path is user_incar_patch, not a later patching pass after preset generation.