Output Data Description

DeepPseudopot generates a rich set of artefacts as it initializes, trains, and exports neural network pseudopotentials. Use this guide to understand what appears in the results directory and how to interpret each file.

Initialization Artefacts

File	Description
oldFunc_plotBS.pdf	Baseline band structure produced by the analytic Zunger potentials prior to neural-network fitting.
initZunger_plotBS.pdf, initZunger_plotPP.pdf	Diagnostics comparing initialized neural potentials against the reference data in its prediction of band structure and pseudopotential itself.
initZunger_pot.dat, initZunger_qSpace_pot.dat	Real- and reciprocal-space potential files derived from the initialized network.
initZunger_PPmodel.pth, initZunger_epoch_<N>_PPmodel.pth	Neural-network checkpoints saved during the optional Zunger pre-fit stage.

Training and Refinement Outputs

File	Description
epoch_<N>_plotBS.{pdf,png}, epoch_<N>_plotPP.{pdf,png}	Snapshots of band-structure and potential convergence during training epochs or Monte Carlo iterations.
training_cost.dat, validation_cost.dat	Data files of the loss values; suitable for plotting and inspecting convergence trends.
mc_iter_<N>_*	Temporary records for each Monte Carlo proposal; retained only when monitoring in-flight runs.
mc_checkpoint.pth, best_pot.*, best_plotPP.*	Best-found models and potentials when Monte Carlo refinement is active.

Final Deliverables

File	Description
final_plotBS.pdf, final_plotPP.pdf	Final visualizations for last-epoch (or best) band structures and pseudopotentials.
final_pot.dat, final_qSpace_pot.dat	Exported potentials ready for downstream simulation pipelines.
movie_BS.mp4, movie_PP.mp4	Optional animations summarizing training progress; fun for visualization.

Suggested Post-processing Workflow

1. Inspect final_plotBS.pdf for agreement with experimental or ab initio references.
2. Compare final_pot.dat against prior versions to quantify improvements.
3. Archive training_cost.dat and validation_cost.dat for regression tracking across tuning runs.
4. For Monte Carlo campaigns, promote mc_checkpoint.pth or best_pot.* into the next initialization bundle.

For real-time monitoring, restart strategies, and troubleshooting tips, see Troubleshooting Guide.