This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| ReaxFF | |
|---|---|
| Name | ReaxFF |
| Type | Reactive force field |
| Developers | Adri van Duin, William A. Goddard III, Bengt J. van Duin |
| First release | 2001 |
| Latest release | 2020s |
| License | Academic / open-source variants |
| Programming languages | C, C++, Fortran, Python |
| Operating system | Cross-platform |
ReaxFF ReaxFF is a reactive force field approach for atomistic simulations that enables bond formation and bond breaking within classical molecular dynamics. It bridges atomistic modeling methods used at institutions such as California Institute of Technology, Massachusetts Institute of Technology, Stanford University, Lawrence Livermore National Laboratory and Sandia National Laboratories with quantum chemistry references from groups like IBM Research, Bell Labs, Max Planck Society and Los Alamos National Laboratory. Developed in the early 2000s and applied across materials science, catalysis, and combustion, ReaxFF has been advanced by collaborations involving researchers at Purdue University, Pennsylvania State University, University of California Berkeley, Rutgers University, University of Oxford, University of Cambridge, ETH Zurich, Imperial College London, and University of Tokyo.
ReaxFF was introduced to provide a transferable reactive potential for large-scale simulations where methods from Hartree–Fock, Density Functional Theory, Coupled Cluster, Møller–Plesset perturbation theory or Quantum Monte Carlo are computationally prohibitive. It targets systems studied by experimental facilities such as Argonne National Laboratory, Brookhaven National Laboratory, Oak Ridge National Laboratory, European Synchrotron Radiation Facility and Diamond Light Source. Influenced by earlier force fields like CHARMM, AMBER, OPLS-AA, COMPASS and GROMOS, ReaxFF emphasizes dynamic bonding and charge equilibration for applications related to projects at NASA, DARPA, DOE and industrial partners including Dow Chemical Company, BASF, DuPont and Shell.
The theoretical foundations combine concepts from reactive empirical potentials and electronic structure theory developed in the wider literature of Walter Kohn and John Pople. ReaxFF is built on energy terms inspired by bond-order theories used by researchers in Pauling-style chemical bonding and parametrization strategies akin to those employed by groups like Perdew and Becke. Charge transfer and polarization are handled using schemes related to the Electronegativity Equalization Method and methods developed by Rappe and Godby, with training data derived from calculations using B3LYP, PBE0, HSE06 and other Density Functional Theory functionals. Validation often references benchmark datasets from consortia such as Materials Project, Open Quantum Materials Database, NIST and computational campaigns at NERSC and XSEDE.
ReaxFF expresses the total potential energy as a sum of bond, valence, torsion, van der Waals, Coulombic and specialized terms; this mirrors decomposition approaches used in classical models like AMBER and CHARMM. Bond orders are continuous variables allowing smooth transitions akin to approaches by Tersoff and Brenner in reactive carbon potentials; angular and torsional contributions recall treatments in MMFF94. Nonbonded interactions are modulated by distance-dependent functions with screening reminiscent of methods used at Sandia National Laboratories and analytic forms compared to those in Lennard-Jones and Buckingham potentials. Charge equilibration schemes are related to algorithms from Rappe and Goddard groups and implemented with numerical solvers common in high-performance computing environments at LLNL and ORNL.
Parameter sets are derived by fitting to reference data produced by quantum chemistry calculations and experimental observables from facilities like NIST, JPL, AFRL and European Research Council-funded projects. Optimization techniques borrow from algorithms associated with Levenberg–Marquardt, Genetic Algorithm approaches, Bayesian optimization workflows championed at Google DeepMind and high-throughput frameworks developed by Materials Project and AFLOW. Parameter libraries exist for chemistries involving elements studied at institutions such as CNRS, RIKEN, Max Planck Institute for Coal Research and Shanghai Jiao Tong University.
ReaxFF is implemented in major molecular simulation packages including LAMMPS, GROMACS (via plugins), AMBER (plugins), GULP, Materials Studio, AIMD wrappers and in custom codes at Lawrence Berkeley National Laboratory. Interfaces to workflow managers and environments like ASE, VMD, Ovito, CMake, MPI and CUDA enable integration with HPC systems such as Frontera, Summit, Perlmutter and cloud platforms provided by Amazon Web Services and Google Cloud. Community distributions and forks are maintained by research groups at Purdue University, University of Utah, University of Illinois Urbana-Champaign and national labs.
ReaxFF has been applied to studies of hydrocarbon combustion relevant to SAE International standards, heterogeneous catalysis as investigated at Max Planck Institute for Chemical Energy Conversion, biomass pyrolysis projects supported by DOE Bioenergy Technologies Office, corrosion processes relevant to NACE International standards, lithium-ion battery electrode reactions in collaborations involving Toyota Research Institute and Argonne National Laboratory, and shock and detonation chemistry relevant to Naval Research Laboratory and Lawrence Livermore National Laboratory programs. It has been used to simulate graphene growth studied at IBM Research, oxide formation relevant to Siemens research, and polymer degradation examined by groups at Dow Chemical Company and BASF.
Validation commonly compares ReaxFF predictions to DFT benchmarks, experimental thermochemistry from NIST and spectroscopy data from facilities like APS and ESRF. Limitations include transferability challenges noted in studies published by groups at UC Berkeley, MIT, ETH Zurich and Oxford, dependence on quality of reference data as highlighted by Materials Project teams, and computational cost overhead relative to nonreactive potentials in codes used at Lawrence Livermore National Laboratory and Sandia National Laboratories. Ongoing work addresses long-range electrostatics, dispersion corrections similar to developments by Grimme, and integration with machine-learned potentials from efforts by DeepMind, Google AI, and academic consortia.
ReaxFF originated in 2001 with seminal work led by researchers affiliated with Purdue University, Caltech, Sandia National Laboratories and NASA-funded collaborations. Subsequent milestones include parameter extensions for hydrocarbons, oxides and halogens developed through projects at DOE laboratories and European counterparts such as CNRS and Max Planck Society. Major software integrations into LAMMPS and community benchmarking initiatives occurred during the 2000s and 2010s with contributions from LLNL, ORNL, Berkeley Lab and universities including Rutgers University and Penn State. Recent years have seen hybridization with machine learning methods pursued by teams at Stanford University, MIT, Carnegie Mellon University and industry partners like Intel and NVIDIA.