Selvauma/ps-polysulfide-deepmd

Li–S Polysulfide Conversion on Carbon Substrates (DeepMD)

What this is

DeepMD-kit potentials supporting a Li–S battery polysulfide (PS) shuttle/conversion study: capacity fade in Li–S batteries is driven by soluble lithium polysulfides (Li₂Sₓ, x = 1–8) dissolving out of the cathode and shuttling to the anode. This project characterizes Sₓ/Li₂Sₓ formation energetics and their binding to carbon substrates (a mitigation strategy — strong PS–carbon binding suppresses the shuttle), feeding into a mesoscale cathode-precipitation model (MonaLiSa, Purdue collaboration).

Reference DFT-level results already converged and manuscript-ready (PS_docs/MANUSCRIPT_PLAN.md):

• Gas-phase Sₓ (x=1–8) formation energies
• Li₂Sₓ (x=1,2,4,6,8) + S₈ binding energies on a carbon "flake" substrate

Note: per an explicit 2026-09-02 scope decision, no further DFT/AIMD/MLMD is planned for this project — what's here is the final set.

Files & Validation

File	RMSE energy (eV/atom)	RMSE force (eV/Å)	Training steps	Size
model/ps_composite_variant1.pb	0.00718	0.0903	1,000,000	89.1 MB
model/ps_reaction.pb	0.00139	0.1	400,000	81.0 MB

RMSE values are validation-set (held-out) energy/force error, read directly from each run's DeepMD-kit lcurve.out at its final training step — not re-derived or estimated.

2 of 5 potentials from the composite/porous-carbon/polysulfide training set variants, selected by lowest validation force RMSE; several early small-cell/test iterations are archived and not included here.

Training pipeline

VASP/PAW, PBE-GGA, ENCUT 420–450 eV, force convergence ±10 meV/Å (dense k-mesh for periodic carbon substrates, Γ-only for molecular Sₓ/Li₂Sₓ clusters) → DeepMD-kit train/freeze/compress.

Citation

Selva Chandrasekaran Selvaraj, University of Illinois Chicago. Manuscript in preparation: "Atomic Reaction Barrier of PS Conversion", in collaboration with the MonaLiSa mesoscale-modeling group (Purdue University).