Skip to content

Evaluation and comparison of classical interatomic potentials through a user-friendly interactive web-interface

Summary

This entry records a Scientific Data descriptor PDF path associated with Tavazza et al. (journal citation 4, 160125, 2017) that is bibliographically the same dataset release as 2016choudhary-venue-untitled (shared DOI 10.1038/sdata.2016.125). The project publishes a database and interactive web interface comparing 0 K energetics and elastic properties for a large panel of materials computed with many classical interatomic potentials—including EAM, MEAM, Tersoff, Stillinger–Weber, AIREBO, COMB, and ReaxFF—against Materials Project DFT references and available experimental elastic data. The purpose is benchmarking at scale: making force-field errors auditable, versioned, and comparable across potential families rather than reporting a single application study.

Methods

This Scientific Data descriptor (DOI 10.1038/sdata.2016.125, 2017 publication metadata) is bibliographically the same dataset release as 2016choudhary-venue-untitled; this slug exists because the corpus ingested a 2017 PDF filename variant (papers/Others/Tavazza_Interatomic_potential_comparison_2017.pdf).

1 — MD application (atomistic dynamics)

The project runs high-throughput LAMMPS workflows at 0 K (structure relaxation / finite-difference strain) to compute elastic tensors and related mechanical diagnostics for many materials × many classical interatomic potentials (EAM, MEAM, Tersoff, Stillinger–Weber, AIREBO, COMB, ReaxFF, …) compared against Materials Project (MP) DFT references and experimental elastic data where available.

  • Engine / code: LAMMPS for automated 0 K energy/strain evaluations (descriptor text + 2016choudhary-venue-untitled).
  • System size & composition: Crystalline MP-consistent unit/supercells per material entry in the database (exact sizes live in dataset records).
  • Boundaries / periodicity: 3D PBC for bulk crystal elastic calculations (standard MP-style setups mirrored in the pipeline).
  • Ensemble: N/A — 0 K static limits; no finite-T NVT production MD in the benchmark core described for elastic scoring.
  • Timestep: N/A — no finite-timestep MD time integration in the 0 K elastic workflow summarized here.
  • Duration / stages: N/A — not a ps/ns trajectory study; single-point / strain-step sequences instead.
  • Thermostat: N/A — 0 K statics.
  • Barostat: N/A — elastic constants from stress–strain or energy–strain derivatives without NPT dynamics.
  • Temperature: 0 K static reference comparisons (explicitly flagged as the baseline scope).
  • Pressure: N/A — hydrostatic NPT trajectories are not the headline workflow for the elastic database.
  • Electric field: N/A — not used.
  • Replica / enhanced sampling: N/A — high-throughput grid over potentials/materials, not enhanced sampling MD.

2 — Force-field training

N/A — the publication benchmarks existing parameter files; it does not introduce a new ReaxFF fit in this descriptor.

3 — Static QM / DFT-only

Materials Project DFT supplies reference elastic tensors / hull data used to score each classical potential’s 0 K predictions against QM and, where present, experiment.

Findings

Across thousands of calculations, the database makes clear that elastic quality is multi-dimensional: a potential may reproduce one tensor component while failing on another, and PCA highlights structured correlations among errors rather than IID noise. For ReaxFF, inclusion alongside fixed-bond models situates reactive potentials in the same comparative ecosystem—useful for parameterization QA even though ReaxFF is often deployed in chemically reactive contexts beyond the 0 K elastic benchmarks emphasized here. Limitations include the baseline focus on 0 K elastic properties; temperature-dependent phonon/anelastic effects are explicitly flagged as outside the first release’s scope. Maintainers should treat this PDF as bibliographically redundant with 2016choudhary-venue-untitled: the DOI is the stable identifier, while filename differences track publisher packaging or ingest duplicates without implying a second dataset. When citing elastic comparisons from the web portal, record the potential file version and LAMMPS build metadata alongside Materials Project MP IDs, because elastic tensors for a given crystal can shift slightly as MP DFT settings evolve across database generations. PCA plots in the descriptor are pedagogical: they summarize correlation structure but are not a substitute for inspecting raw error vectors on your material of interest.

Relevance to group

Provides a benchmarking ecosystem adjacent to ReaxFF development: the same comparative mindset (QM/experiment vs empirical models) used when curating reactive force fields.

Citations and evidence anchors

  • reaxff-family
  • Classical force-field repositories and validation