Concepts

Walkers, iterations, and spawning

Each iteration runs a batch of short MD walkers. Saved frames are projected into a CV space; a spawner then chooses which frames to restart the next iteration's walkers from. Spawners:

spawn_scheme	Strategy
density	Restart from under-populated regions of a regular grid.
voronoi	Restart from sparse Voronoi (k-means) cells; scales better than a grid in higher-dimensional CV spaces.
lof	Restart from statistical outliers (local outlier factor).
fps	Farthest-point sampling for maximal coverage.
we	Weighted-ensemble split/merge resampling. Experimental (beyond manuscript scope); see MSM & kinetic seeding on weight validity.
msm	MSM least-counts / uncertainty-driven spawning (needs msm.enabled). Experimental (beyond manuscript scope).

Any spawner can also be pointed toward a target region of the CV space (search_mode: target), balancing exploration with progress toward the target.

CV spaces

A run uses either:

• Fixed CVs (space_mode: fixed) — a user project_file returning physical CVs (dihedrals, distances, …), or
• Learned CVs (space_mode: pca | tica | tvae | deep-tica) — trained on the fly from input features and periodically retrained (retrain_freq). See Collective variables.

Input features

Learned CVs are trained on input features extracted from the trajectories: pairwise distances, fitted_coords, or system-specific dihedrals, restricted by the feature_selection atom mask.

Convergence

Sampling proceeds for the configured iteration budget, or stops early when grid/Voronoi bin occupancy plateaus (resolution_check_patience, convergence_patience in spawning). Because a retrained learned CV space can rotate, shift, or scale, bin boundaries are recalculated in the newly projected space whenever the model retrains.

After a campaign, representative structures can seed longer production runs for post-hoc MSM construction — see MSM & kinetic seeding.

Execution

Walkers are dispatched by an execution backend: local (multi-GPU workstation) or slurm / pbs (HPC array jobs). The choice is purely a config setting and does not affect the science. See Execution.

Reproducibility & provenance

• Global deterministic seeding (random_seed).
• Per-iteration checkpoints (checkpoint_freq) with --resume.
• Every frame carries lineage (iteration:walker:frame + parent), enabling connected-path reconstruction with trails-md-path.