# Reactive molecular dynamics

Reactive molecular dynamics is a molecular dynamics simulation approach in which chemical bonds can break and form during the trajectory, unlike standard fixed-topology force fields, so that reaction pathways, intermediate species, and product distributions emerge directly from the simulation. The [ReaxFF](https://www.edgepedia.org/reaxff) reactive force field was developed to make practical the simulation of reactive chemical systems of thousands of atoms<sup>[1](https://doi.org/10.1021/jp004368u)</sup> by describing reaction chemistry without explicit quantum mechanical calculation.<sup>[2](https://www.nature.com/articles/npjcompumats201511)</sup> Output includes time-resolved chemical species populations, reported for example with LAMMPS `fix reaxff/species`.<sup>[3](https://lammpstutorials.github.io/sphinx/build/html/tutorial5/reactive-silicon-dioxide.html)</sup>

| Key fact | Value |
|---|---|
| Core principle | Bond order computed from interatomic distance; bond energy from bond order, so bonds dissociate continuously to separated atoms<sup>[1](https://doi.org/10.1021/jp004368u)</sup> |
| Cost vs classical MD | About 10–50 times slower than non-reactive force fields<sup>[4](https://www.lammps.org/workshops/Aug13/vanDuin/LAMMPS_workshop_vanDuin.pdf)</sup>; ~18 times slower in an independent benchmark<sup>[5](https://pmc.ncbi.nlm.nih.gov/articles/PMC8145783/)</sup> |
| Cost vs quantum chemistry | 2–5 times faster than PM3 or SCC-DFTB; about three orders of magnitude faster than HF or DFT (B3LYP/6-31G*)<sup>[5](https://pmc.ncbi.nlm.nih.gov/articles/PMC8145783/)</sup> |
| Time step | Typically 0.25–0.5 fs; smaller steps at high temperature<sup>[6](https://www.engr.psu.edu/ADRI/Upload/reax_um.pdf)</sup> |
| System size | More than 1,000,000 atoms reported with parallel LAMMPS/ReaxFF<sup>[4](https://www.lammps.org/workshops/Aug13/vanDuin/LAMMPS_workshop_vanDuin.pdf)</sup> |
| Parameterization targets | Heats of formation within 4.0 kcal/mol, bond lengths within 0.01 Å, bond angles within 2°<sup>[1](https://doi.org/10.1021/jp004368u)</sup> |
| Main software | LAMMPS, PuReMD/sPuReMD, GULP, AMBER, ADF<sup>[5](https://pmc.ncbi.nlm.nih.gov/articles/PMC8145783/)</sup><sup> • </sup><sup>[2](https://www.nature.com/articles/npjcompumats201511)</sup> |

## How it works

The enabling idea is the bond-order formalism: bond order is calculated empirically from interatomic distances, and bond energy follows from bond order, so a stretching bond loses order smoothly and dissociates properly to separated atoms.<sup>[1](https://doi.org/10.1021/jp004368u)</sup> Electronic interactions driving bonding are treated implicitly, which allows reaction chemistry without explicit quantum mechanical treatment.<sup>[2](https://www.nature.com/articles/npjcompumats201511)</sup> In ReaxFF the bond order is a sum of sigma, pi, and pi-pi contributions, each an exponential function of distance; the sigma term is near unity below about 1.5 Å and negligible above about 2.5 Å, and the raw bond orders are corrected for overcoordination before use.<sup>[1](https://doi.org/10.1021/jp004368u)</sup><sup> • </sup><sup>[7](https://doi.org/10.1137/100808599)</sup>

The total energy is a sum of continuous terms: bond energy, over- and undercoordination corrections, lone-pair energy, valence angle, torsion, hydrogen bond, van der Waals, and Coulomb terms.<sup>[2](https://www.nature.com/articles/npjcompumats201511)</sup><sup> • </sup><sup>[7](https://doi.org/10.1137/100808599)</sup> Valence angle energies are made bond-order dependent, so the equilibrium angle shifts smoothly from about 109.47° for sp3 to 120° for sp2 to 180° for sp hybridization as pi-bond order grows, and angle terms vanish on dissociation.<sup>[1](https://doi.org/10.1021/jp004368u)</sup> Design rules forbid any discontinuity in energy or forces during a reaction, use one atom type per element, and require no pre-definition of reactive sites or pathways.<sup>[6](https://www.engr.psu.edu/ADRI/Upload/reax_um.pdf)</sup> Partial charges are not fixed: they are re-derived each step by charge equilibration (QEq or EEM), so charges respond to local environment, for example shifting after fracture creates new surfaces.

## How it is done

A practical ReaxFF workflow in LAMMPS uses `pair_style reaxff` with a parameter file such as ffield.reax.CHOFe, `fix qeq/reaxff` to perform charge equilibration at every step (with low and high cutoffs, a tolerance, and a maximum iteration count), and `fix reaxff/species` to report which chemical species are present at intervals, which is how bond breaking and formation are tracked. Time steps of 0.25–0.5 fs retain reasonable energy conservation at 0–1500 K; higher temperatures need smaller steps.<sup>[6](https://www.engr.psu.edu/ADRI/Upload/reax_um.pdf)</sup> The per-step charge equilibration requires solving a large sparse linear system with shielded electrostatic kernels, which dominates cost and motivates specialized solvers such as GMRES with ILUT preconditioning.<sup>[7](https://doi.org/10.1137/100808599)</sup> Force-field selection or training against quantum chemistry data for the chemistry of interest precedes production runs.<sup>[8](https://www.ctcms.nist.gov/potentials/Download/2010Workshop/NIST_vanDuin_July2010.pdf)</sup>

## Origin

Reactive molecular dynamics grew out of bond-order potentials for hydrocarbons. The second-generation reactive empirical bond order (REBO) potential, published by Donald W. Brenner and colleagues in 2002 in the *Journal of Physics Condensed Matter*, allows covalent bond breaking and forming with changes in atomic hybridization within a classical potential, improving on an earlier 1990 version.<sup>[9](https://doi.org/10.1088/0953-8984/14/4/312)</sup> AIREBO, reported by Steven J. Stuart, Alan B. Tutein, and Judith A. Harrison in 2000 in *The Journal of Chemical Physics*, extended REBO with adaptive nonbonded and dihedral interactions so that condensed-phase hydrocarbons could be modeled alongside their reactions.<sup>[10](https://doi.org/10.1063/1.481208)</sup> ReaxFF itself was reported by Adri C. T. van Duin and colleagues in 2001 in *The Journal of Physical Chemistry A*.<sup>[1](https://doi.org/10.1021/jp004368u)</sup> A widely used extension to hydrocarbon oxidation (the CHNO parameterization) was reported by Kimberly Chenoweth, Adri C. T. van Duin, and [William A. Goddard](https://www.edgepedia.org/william-a-goddard) in 2008 in *The Journal of Physical Chemistry A*.<sup>[11](https://doi.org/10.1021/jp709896w)</sup> On the software side, the PuReMD implementation and its algorithms were described by Hasan Metin Aktulga and colleagues in 2012 in the *SIAM Journal on Scientific Computing*,<sup>[7](https://doi.org/10.1137/100808599)</sup> with GPU implementations following in 2013 and 2014.<sup>[12](https://doi.org/10.1016/j.jmgm.2013.02.001)</sup><sup> • </sup><sup>[13](https://doi.org/10.1016/j.jcp.2014.04.035)</sup>

## Variants

ReaxFF and the charge-optimized many-body (COMB) potential are two widely used reactive potentials with variable-charge schemes suited to multicomponent systems.<sup>[14](https://www.annualreviews.org/content/journals/10.1146/annurev-matsci-071312-121610)</sup> Variants address specific weaknesses: eReaxFF adds explicit electrons (assigned a mass of 1 amu) with ACKS2 charges to treat charge transfer and redox chemistry,<sup>[2](https://www.nature.com/articles/npjcompumats201511)</sup> and ReaxFF-nn replaces part of the analytic form with a small neural network computing bond order and bond energy, implemented in GULP and LAMMPS.<sup>[15](https://pubs.rsc.org/en/content/articlelanding/2025/cp/d4cp00535j)</sup> The IFF-R (reactive INTERFACE force field) uses Morse-potential bonds as a faster alternative for predefined dissociation chemistry.<sup>[16](https://nature.com/articles/s41467-024-50793-0.pdf)</sup> A different philosophy is LAMMPS `fix bond/react`, which applies predetermined topology changes using distance-based probabilistic criteria rather than bond-order potentials, intended for building polymeric, amorphous, or cross-linked systems.<sup>[17](https://docs.lammps.org/latest/fix_bond_react.html)</sup> Machine learning potentials (MLPs) with neural network and kernel-based algorithms have also been developed for reactive systems involving bond breaking and formation; reactive MLPs speed up calculation of reactive dynamics and facilitate study of reaction trajectories, reaction rates, and free energies, with data sampling and active learning used to capture rare events.<sup>[18](https://www.annualreviews.org/content/journals/10.1146/annurev-physchem-062123-024417)</sup>

## Applications

Published ReaxFF applications span pyrolysis and oxidation of gas, liquid, and solid fuels, catalytic reactions, soot formation, [flame synthesis](https://www.edgepedia.org/flame-synthesis) of nanoparticles, droplet collision and evaporation, and CO2 capture under subcritical and supercritical conditions.<sup>[19](https://discovery.ucl.ac.uk/id/eprint/10169357/)</sup> Reviews also document studies of organic reactions under extreme conditions relevant to high-energy materials, hydrocarbons, and coals; nanomaterials including graphene oxides, carbon nanotubes, silicon nanowires, and metal nanoparticles; catalytic mechanisms of metals and metal oxides; and electrochemical mechanisms in fuel cells and lithium batteries.<sup>[20](https://academic.hep.com.cn/fcse/EN/10.1007/s11705-015-1545-z)</sup> Hybrid reactive/nonreactive schemes such as ReaxFF/AMBER extend these uses to biomolecular systems with local reactive regions.<sup>[5](https://pmc.ncbi.nlm.nih.gov/articles/PMC8145783/)</sup>

## Limitations and alternatives

Charge equilibration is the best-documented weakness: EEM/QEq can produce unphysical long-range charge transfer, non-integer molecular charge at large separations, lack of out-of-plane polarization, and lack of energy conservation; the ACKS2 model, which solves a \( (2N+2) \times (2N+2) \) matrix and enforces integer charges after dissociation, was designed to impede these artifacts.<sup>[21](https://par.nsf.gov/biblio/10284475-recent-advances-improving-accuracy-transferability-xe-fficiency-reactive-force-fields)</sup> Accuracy is chemistry-dependent: one published comparison reports ReaxFF bond dissociation curve deviations of typically 10–50% from reference data, exceeding 1000% for iron,<sup>[16](https://nature.com/articles/s41467-024-50793-0.pdf)</sup> and ReaxFF shows well-documented shortcomings modeling nitromethane explosion.<sup>[18](https://www.annualreviews.org/content/journals/10.1146/annurev-physchem-062123-024417)</sup> Parameterization typically relies on DFT-level data, and LDA/GGA functionals can themselves be insufficient for metal oxide heats of formation, so parameter sets should not be used as a black box.<sup>[21](https://par.nsf.gov/biblio/10284475-recent-advances-improving-accuracy-transferability-xe-fficiency-reactive-force-fields)</sup> Reparameterization is generally required for new reaction or chemistry types.<sup>[18](https://www.annualreviews.org/content/journals/10.1146/annurev-physchem-062123-024417)</sup> Because classical dynamics on a reactive potential rarely crosses energy barriers on its own, enhanced sampling such as umbrella sampling or metadynamics is used for rare reactive events.<sup>[18](https://www.annualreviews.org/content/journals/10.1146/annurev-physchem-062123-024417)</sup> ReaxFF is 10–50 times slower than non-reactive force fields; its computational scaling is implementation- and solver-dependent, with the charge-equilibration step a major cost factor,<sup>[7](https://doi.org/10.1137/100808599)</sup> in contrast to the \( N^{3} \) scaling at best for quantum methods,<sup>[4](https://www.lammps.org/workshops/Aug13/vanDuin/LAMMPS_workshop_vanDuin.pdf)</sup> and parallel LAMMPS/ReaxFF has been reported on more than 1,000,000 atoms.<sup>[4](https://www.lammps.org/workshops/Aug13/vanDuin/LAMMPS_workshop_vanDuin.pdf)</sup> Alternatives: ab initio MD gives higher fidelity but is limited to a few thousand atoms by \( O(N^{3}) \)–\( O(N^{7}) \) scaling;<sup>[7](https://doi.org/10.1137/100808599)</sup> `fix bond/react` suits known mechanisms and crosslinking but reproduces only end results, and quantum mechanical modeling, path-integral MD, or reactive empirical potentials are more appropriate when predicting reactivity is the objective.<sup>[22](https://pmc.ncbi.nlm.nih.gov/articles/PMC7608055/)</sup> For speed on predefined dissociation chemistry, IFF-R runs on average about 30 times faster than ReaxFF in LAMMPS for 1,000–10,000 atom systems using about 10% of the memory.<sup>[16](https://nature.com/articles/s41467-024-50793-0.pdf)</sup>

## References

1. [Adri C. T. van Duin and colleagues (2001). ReaxFF: A Reactive Force Field for Hydrocarbons. The Journal of Physical Chemistry A.](https://doi.org/10.1021/jp004368u)
2. [The ReaxFF reactive force-field: development, applications and future directions (npj Computational Materials, 2015)](https://www.nature.com/articles/npjcompumats201511)
3. [LAMMPS tutorial: Reactive silicon dioxide (lammpstutorials.github.io)](https://lammpstutorials.github.io/sphinx/build/html/tutorial5/reactive-silicon-dioxide.html)
4. [Development and application of the ReaxFF reactive force field method (van Duin, LAMMPS workshop, August 2013)](https://www.lammps.org/workshops/Aug13/vanDuin/LAMMPS_workshop_vanDuin.pdf)
5. [ReaxFF/AMBER, A Framework for Hybrid Reactive/Nonreactive Force Field Molecular Dynamics Simulations](https://pmc.ncbi.nlm.nih.gov/articles/PMC8145783/)
6. [ReaxFF User Manual (Penn State, Adri van Duin group)](https://www.engr.psu.edu/ADRI/Upload/reax_um.pdf)
7. [Hasan Metin Aktulga and colleagues (2012). Reactive Molecular Dynamics: Numerical Methods and Algorithmic Techniques. SIAM Journal on Scientific Computing.](https://doi.org/10.1137/100808599)
8. [Development and applications of ReaxFF reactive force fields for combustion, catalysis and material failure (van Duin, NIST workshop slides, 2010)](https://www.ctcms.nist.gov/potentials/Download/2010Workshop/NIST_vanDuin_July2010.pdf)
9. [Donald W Brenner and colleagues (2002). A second-generation reactive empirical bond order (REBO) potential energy expression for hydrocarbons. Journal of Physics Condensed Matter.](https://doi.org/10.1088/0953-8984/14/4/312)
10. [Steven J. Stuart, Alan B. Tutein, Judith A. Harrison (2000). A reactive potential for hydrocarbons with intermolecular interactions. The Journal of Chemical Physics.](https://doi.org/10.1063/1.481208)
11. [Kimberly Chenoweth, Adri C. T. van Duin, William A. Goddard (2008). ReaxFF Reactive Force Field for Molecular Dynamics Simulations of Hydrocarbon Oxidation. The Journal of Physical Chemistry A.](https://doi.org/10.1021/jp709896w)
12. [Mo Zheng, Xiaoxia Li, Li Guo (2013). Algorithms of GPU-enabled reactive force field (ReaxFF) molecular dynamics. Journal of Molecular Graphics and Modelling.](https://doi.org/10.1016/j.jmgm.2013.02.001)
13. [S.B. Kylasa, H.M. Aktulga, A.Y. Grama (2014). PuReMD-GPU: A reactive molecular dynamics simulation package for GPUs. Journal of Computational Physics.](https://doi.org/10.1016/j.jcp.2014.04.035)
14. [Reactive Potentials for Advanced Atomistic Simulations (Liang et al., Annu. Rev. Mater. Res. 43, 2013)](https://www.annualreviews.org/content/journals/10.1146/annurev-matsci-071312-121610)
15. [ReaxFF-nn: a reactive machine-learning potential in GULP/LAMMPS and its applications in the thermal conductivity calculations of carbon nanostructures (Phys. Chem. Chem. Phys., 2025)](https://pubs.rsc.org/en/content/articlelanding/2025/cp/d4cp00535j)
16. [Reactive INTERFACE Force Field (IFF-R), Nature Communications 2024](https://nature.com/articles/s41467-024-50793-0.pdf)
17. [fix bond/react command, LAMMPS documentation](https://docs.lammps.org/latest/fix_bond_react.html)
18. [Machine Learning of Reactive Potentials (Annual Review of Physical Chemistry)](https://www.annualreviews.org/content/journals/10.1146/annurev-physchem-062123-024417)
19. [Classical and reactive molecular dynamics: Principles and applications in combustion and energy systems (UCL Discovery deposit)](https://discovery.ucl.ac.uk/id/eprint/10169357/)
20. [Development, applications and challenges of ReaxFF reactive force field in molecular simulations (Frontiers of Chemical Science and Engineering, 2015)](https://academic.hep.com.cn/fcse/EN/10.1007/s11705-015-1545-z)
21. [Recent Advances for Improving the Accuracy, Transferability and Efficiency of Reactive Force Fields (Leven et al., J. Chem. Theory Comput. 17, 2021)](https://par.nsf.gov/biblio/10284475-recent-advances-improving-accuracy-transferability-xe-fficiency-reactive-force-fields)
22. [Chemical Reactions in Classical Molecular Dynamics (fix bond/react, LAMMPS)](https://pmc.ncbi.nlm.nih.gov/articles/PMC7608055/)

---
*Topic: Encyclopedia › Physical world and mathematics › Chemistry › Chemical principles and methods*

*Initially written Sep 29, 2026 · Reviewed: Sep 30, 2026 · Edited: Sep 30, 2026 · Last review: Sep 30, 2026*

*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*

License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license