Skip to content

Reactive MD application setup

For readers

This protocol covers practical setup and use of already-published reactive force fields for production application trajectories. It is for deployment and interpretation, not new parameter fitting; route fitting tasks to protocols/reaxff-parameterization-workflow.

Summary

This workflow helps operators run reactive MD in a way that is reproducible, scoped to known force-field validity, and auditable before conclusions are reported. The core idea is to front-load sanity checks, enforce explicit validation gates, and only then scale to long trajectories and interpretation.

In this corpus, successful application studies repeatedly pair (i) chemistry/scope declarations, (ii) short pre-production checks for numerical and chemical stability, and (iii) paper-grounded interpretation rules for products, intermediates, and trends. That pattern appears across high-temperature pyrolysis, oxidation/surface chemistry, electrolyte decomposition, and defect-mediated intercalation contexts.

Inputs and prerequisites

  • A published reactive FF parameter file with documented element coverage and known intended chemistry domain (for example, contexts represented in 2014castro-marcano-journal-of-a-pyrolysis-large-scale, 2014zou-acta-materia-molecular-dynamics, 2020hossain-j-chem-phys-lithium-electrolyte-solvation).
  • MD engine support for the selected reactive FF and charge equilibration workflow (engine/version and relevant pair style settings recorded in run metadata).
  • Initial structures consistent with target phase, density, defect state, and composition; no impossible overlaps or unsupported atom types.
  • A reaction/species analysis plan defined before production (what species, bonds, events, and observables are required to claim success).
  • A paper-grounded acceptance matrix tying each planned claim to at least one corpus anchor in source_refs.

Procedure

  1. Declare applicability boundary.
    Record what chemistry the chosen FF is expected to represent, plus explicit "do not interpret" zones (for example, extreme oxidation state or composition regimes not covered in source studies).
  2. Preflight topology and typing.
    Validate atom typing, stoichiometry, boundary conditions, and initial geometry quality; fail fast if unsupported species or unrealistic close contacts appear.
  3. Run micro-sanity trajectories.
    Execute short trajectories at target and bracket temperatures to verify stable integration, charge convergence behavior, and absence of immediate unphysical fragmentation/polymerization.
  4. Calibrate control parameters.
    Tune timestep and thermostat/barostat coupling for stable yet physically interpretable dynamics; keep settings fixed once accepted for production.
  5. Define observables before scaling.
    Lock analysis scripts/criteria for species tracking, bond-event counting, diffusion/reaction proxies, and trajectory quality diagnostics.
  6. Execute staged production.
    Use staged runs (equilibration then production windows) with checkpointing and reproducible seeds; log all environment and input hashes.
  7. Apply validation gates after each stage.
    Continue only if all validation gates pass (below). If any gate fails, stop, classify failure mode, and revise setup rather than extending runtime.
  8. Interpret with scope discipline.
    Report only trends/mechanisms supported by passed gates and corpus-consistent behavior; flag extrapolative claims as hypotheses.

Validation checks and acceptance criteria

  • Gate 1 - Numerical stability: no runaway temperature/pressure behavior beyond expected fluctuations for the chosen ensemble, and no persistent integration instability across replicate seeds.
  • Gate 2 - Charge-equilibration health: charge solution converges reliably with no repeated solver failures; large oscillatory charge artifacts trigger rejection.
  • Gate 3 - Chemical plausibility baseline: early-time chemistry is qualitatively consistent with expected regime from relevant corpus papers (for example, no dominant impossible products in the first analysis window).
  • Gate 4 - Replicate consistency: key trend directions (major species evolution, relative pathway ranking, or interfacial reactivity ordering) are stable across independent replicas.
  • Gate 5 - Sensitivity sanity: modest perturbations (temperature window, timestep sanity adjustment, initial velocity seed) do not invert central conclusions.
  • Gate 6 - Claim-to-evidence traceability: every reported mechanistic statement is tied to observable definitions and at least one source-backed comparator from source_refs.

A run is accepted for interpretation only when all six gates pass and all deviations are documented.

Failure modes and mitigations

  • Unphysical early chemistry (explosive decomposition or inert deadlock): reduce timestep, re-check initial geometry/typing, and verify chemistry lies inside known FF scope before retrying.
  • Charge solver instability or extreme charge swings: tighten solver tolerances if available, inspect problematic local motifs, and test whether artifacts persist across minimized/pre-equilibrated starting states.
  • Replica divergence dominated by rare events with no stable trend: increase replica count and report distributional outcomes rather than a single exemplar trajectory.
  • Product network dominated by species absent from paper-grounded expectations: pause interpretation; treat as potential out-of-domain behavior unless independently validated by higher-level evidence.
  • Boundary/finite-size artifacts (surface systems, defects, confined phases): test larger cells or alternative boundary setups before assigning mechanism-level conclusions.
  • Over-interpretation of accelerated/high-T trajectories: explicitly separate "pathway accessibility under simulated conditions" from real-world rate claims unless validated against external evidence.

Variants and when to choose them

  • Reproducible simulation package: input decks, FF file identity, runtime metadata, random seeds, and checkpoint lineage.
  • Validation dossier documenting pass/fail status for each gate, plus mitigation notes for any failed attempts.
  • Analysis bundle: species/bond-event outputs, trend summaries, and uncertainty/replicate spread metrics.
  • Interpretation notes mapped to evidence anchors for use in paper pages, theme hubs, and debate pages.
  • Recommended downstream pages:
  • theme-reactive-md-corpus
  • reaxff-family
  • transferability-reactive-ff
  • protocols/reaxff-parameterization-workflow

Evidence anchors