Skip to content

A multiscale code for flexible hybrid simulations

Evidence and attribution

Authority of statements

Prose sections below (Summary, Methods, Findings, etc.) are curated summaries of the publication identified by doi, title, and pdf_path in the front matter above. They are not new primary claims by this wiki.

For definitive numerical values, reaction schemes, and interpretations, use the peer-reviewed article (and optional records under normalized/papers/ when present)—not this page alone.

Summary

This arXiv manuscript describes multi-scale workflows in the Atomistic Simulation Environment (ASE) that combine classical MD in LAMMPS (including ReaxFF) with DFT via GPAW, and introduces a force/energy Mixer that combines multiple calculators in an ONIOM-like weighted scheme. A redesigned ASE/LAMMPS interface represents each LAMMPS force field as its own ASE calculator class (subclassing LAMMPSBase), with FFData parsing, atom typing, and optional partial charges; LAMMPS-specific dynamics classes run long segments inside LAMMPS to avoid per-timestep Python overhead.

Methods

ASE calculators: each LAMMPS functional form is wrapped as a separate calculator; shared LAMMPSBase handles communication, bond/angle/dihedral/improper detection, and LAMMPS input generation from FFData and typing rules (template syntax for chemical environments and precedence). LAMMPS dynamics classes mirror standard ASE dynamics API but batch many timesteps in LAMMPS while still supporting observers and trajectory slices.

Demonstration MD: a phenol-dimer example from the s22 dataset runs 10 ps with ReaxFF (ffield.reax) then 10 ps with CHARMM CGenFF (par_all36_cgenff.prm) using LAMMPS_NVT at 300 K and 1 fs steps; ReaxFF's built-in charge equilibration supplies charges for the subsequent CHARMM stage where the interface does not implement its own QEq.

Mixer: combines full-system classical energy/forces with a QM correction on an inner region using per-atom weights; energy uses an ONIOM-style \(H = H_{3C} + \{H_{2Q} - H_{2C}\}\) split (notation as in the paper), while forces use \(F_i = F_{i,3C} + w_i(F_{i,2Q} - F_{i,2C})\) to mix quantum and classical sub-calculators.

ReaxFF context: text notes ReaxFF as a reactive bridge between ab initio and empirical FF methods, with large speed gain vs GPAW in mixed setups (e.g., 5000-atom systems where 10 atoms on GPAW dominate runtime).

MD application (blueprint slots)

Engine / code: LAMMPS for molecular dynamics when MD is discussed (indexed text). System size & composition: Atom counts / stoichiometry appear in indexed text (see excerpt for numbers). Boundaries / periodicity: Non-periodic / cluster / surface language appears in indexed text (no bulk PBC claim here). Ensemble: NVT (indexed text). Timestep: 1 fs (matched in indexed text) Duration / stages: ps/ns scale timing or equilibration/production language appears in indexed text—see PDF for full schedule. N/A — thermostat type/damping not recovered from indexed excerpt; verify PDF. Barostat: N/A — NVT protocol without hydrostatic pressure control in indexed summary (no NPT stated). Temperature: 300 K (matched in indexed text) Pressure / stress: Pressure/stress language appears in indexed text—see PDF for control mode. Electric field: N/A — external electric field bias not indicated in indexed excerpt for MD (if any field appears, it belongs to static QM/experiment sections). Replica / enhanced sampling: N/A — umbrella / metadynamics / replica exchange not indicated in indexed excerpt.

Findings

Outcomes / mechanisms: The ASE/LAMMPS redesign enables plug-in mixing of LAMMPS (including ReaxFF) with GPAW or other calculators with minimal extra coding, and documents performance-oriented patterns for production classical or hybrid runs. The phenol-dimer walkthrough shows back-to-back ReaxFF and CHARMM dynamics on the same structure. The Mixer adds flexible weighting between QM and classical regions using an ONIOM-style energy split and per-atom force weighting, as defined in the article.

Comparisons: The authors contrast ReaxFF throughput with GPAW costs in representative hybrid sizing arguments (see pdf_path).

Sensitivity / design levers: Calculator choice, LAMMPS batching vs per-step Python, and Mixer weights are the main implementation levers discussed.

Limitations / outlook: Hybrid QM/MM boundaries and weight choices carry accuracy trade-offs; ReaxFF and CHARMM remain training-set-limited outside their intended chemistry.

Corpus / KB honesty: No normalized/extracts/2012leukkunen-venue-paper_p1-2.txt is present in this checkout; summaries follow pdf_path and the curated arXiv record normalized/papers/2012leukkunen-venue-paper.json (extraction_quality: partial).

Limitations

Hybrid QM/MM boundaries and weight choices affect accuracy; ReaxFF and CHARMM remain parameterization-limited outside their training sets.

Relevance to group

Operator reference for ASE + LAMMPS + ReaxFF integration adjacent to PuReMD/LAMMPS reactive workflows in the corpus.

Citations and evidence anchors

  • DOI: 10.48550/arXiv.1211.2075
  • Primary source: papers/ReaxFF_others/Leukkunen -ASE ReaxFF LAMMPS 2012 copy.pdf (see pdf_sha256 in front matter)
  • Normalized stub: normalized/papers/2012leukkunen-venue-paper.json