Review of force fields and intermolecular potentials used in atomistic computational materials research
Summary¶
Atomistic materials modeling relies on a zoo of interatomic potentials ranging from pairwise Lennard-Jones liquids to chemically reactive bond-order models and, increasingly, machine-learned surfaces fit to electronic-structure data. This Applied Physics Reviews survey organizes analytic potentials by functional complexity, typical accuracy envelopes, and computational cost, with dedicated coverage of classical fixed-charge force fields (AMBER, CHARMM, OPLS-AA, GROMOS, TraPPE), polarizable and embedded approaches, empirical metal potentials (EAM/MEAM), carbon and hydrocarbon bond-order models (REBO, AIREBO, COMB, qAIREBO), ReaxFF-style reactive chemistry, and machine-learning potentials that interpolate quantum energies. The pedagogical goal is to help practitioners match models to phenomena—mechanical properties, transport, chemistry at interfaces—without overfitting complexity where simpler models suffice.
Methods¶
Reviews and perspectives (non-simulation primary data). This Applied Physics Reviews entry is a survey, not a single-system benchmark trajectory study. Its Methods are the authors’ literature scope and taxonomy: they organize interatomic potentials for atomistic materials modeling using equations and primary citations rather than prescribing one MD protocol for readers to reproduce wholesale.
Coverage. The review surveys fixed-charge biomolecular/organic fields (AMBER, CHARMM, OPLS-AA, GROMOS, TraPPE); polarizable and embedded-atom approaches; EAM/MEAM metals; carbon/hydrocarbon bond-order models (REBO, AIREBO, COMB, qAIREBO); ReaxFF-class reactive chemistry; and machine-learning potentials fit to electronic-structure data.
How to use this page. Map a phenomenon (mechanics, transport, interface chemistry) to an appropriate potential family using the review’s section headings, then follow primary literature for parameters and validation—do not treat this survey as a substitute for system-specific ReaxFF or classical benchmarks in the van Duin corpus.
Literature comparison protocol. Subsections cite benchmark studies where experiment or high-level QM disagrees with inexpensive FF choices, highlighting when ReaxFF or ML potentials are preferable to fixed-bond models for oxidation or fracture problems.
MD / DFT production blocks. N/A — the article does not center on one LAMMPS/VASP timestep, ensemble, or production nanosecond trajectory; per AGENTS.md, those slots are not applicable as a single numbered protocol here.
Findings¶
Across categories, the review stresses that accuracy, transferability, and simulation speed form a triangle of competing objectives: richly parameterized nonreactive fields can improve thermodynamic fits for specific phases but may fail outside training manifolds, whereas bond-order reactive models encode chemistry at the cost of more elaborate functional forms and parameter training workflows. ReaxFF is positioned within the reactive bond-order lineage as a widely used option for bond formation and rupture in oxides, organics, and metals when quantum molecular dynamics is too expensive. Machine-learning sections emphasize promise and pitfalls—high accuracy near training distributions but careful data governance required for extrapolation.
Sensitivity and outlook. Practitioners must match temperature, pressure, and phase to the FF training manifold; future work in the review points to tighter validation loops as ML potentials proliferate.
Corpus honesty. This page summarizes the Applied Physics Reviews survey (pdf_path); it does not reproduce tabulated benchmark numbers from underlying literature.
Limitations¶
Survey articles cannot replace application-specific validation; numerical comparisons are illustrative rather than exhaustive, and rapidly evolving ML potential literature may outdate specific benchmarks soon after publication.
Relevance to group¶
The review gives external framing for where ReaxFF sits among empirical and machine-learned alternatives, supporting curriculum-style linking from the knowledge base to broader methods discourse.
MAS / retrieval notes¶
Treat this entry as a methods survey rather than a single-system benchmark: cite the review’s internal section headings when users ask which potential class fits ceramics, polymers, or metals, and defer quantitative parameters to primary literature referenced there. Stable locator: DOI 10.1063/1.5020808; page numbers should be copied from the Applied Physics Reviews PDF because pagination can differ across mirrors.