REAXFF reactive force field for disulfide mechanochemistry, fitted to multireference ab initio data
Authority of statements
Prose summarizes J. Chem. Theory Comput. 2016, 12, 3913–3925 (DOI 10.1021/acs.jctc.6b00461).
Summary¶
Müller and Hartke reparametrize ReaxFF for disulfide mechanochemistry using multireference ab initio reference data. The motivation is single-molecule AFM-style pulling: routine single-reference DFT is a weak match for S–S bond breaking under load, and ab initio MD cannot reach millisecond-scale traces. They fit with a nondeterministic global evolutionary algorithm, then illustrate the field with exploratory ReaxFF trajectories on large multifunctional mechanophores (DSM-C and DSM-PEG in the article) where disulfides are intentional weak links—sizes and solvent/handle motifs that are impractical for first-principles dynamics.
Methods¶
Force-field training. The work refits ReaxFF for mechanochemical disulfide chemistry in designed organomechanophores (Figure 1: DSM-C for conducting/solution AFM scenarios; DSM-PEG with polyethylene glycol handles), separate from earlier protein disulfide studies that relied on DFT-only training data. QM references combine multireference ab initio energy surfaces along stretch and bond-breaking paths with single-determinant HF/DFT comparisons where single-reference models fail (section 3.1). The training set covers those disulfide pathways plus additional geometries in the Supporting Information. Optimization uses the authors’ evolutionary-algorithm global search, motivated by their estimate that roughly 80 ± 20 parameters per atom must be matched to reliable reference energies for a usable reactive field. Validation is tied to those multireference surfaces and to literature context on ReaxFF practice for related chemistry.
MD application. Exploratory trajectories use LAMMPS with the authors’ newly fitted ReaxFF parametrization for the targeted C/H/O/S mechanophores (the article contrasts its error against legacy LAMMPS CHOS parametrizations such as Singh et al. and Mattsson et al. when discussing starting points for optimization). Systems include multifunctional mechanophores in vacuo and in a toluene solvent box with periodic boundary conditions; peripheral anchor atoms mimic attachment to a surface and AFM tip. Integration uses a 0.5 fs time step at 300 K in the NVT ensemble with a Berendsen thermostat (damping factor 100 in the paper’s units). The authors seed random velocities per trajectory to avoid redundant phase-space sampling. Illustrative pulling segments reach 25 ps in one quoted comparison to AFM timescales, and rupture scouting runs search for events within 50 ps windows when ramping end forces. N/A — barostat / hydrostatic pressure control — not used for these constant-volume gallery simulations. N/A — applied electric field; umbrella / metadynamics / replica exchange — not reported for these exploratory runs.
Static QM. Multireference and DFT/HF calculations generate the training surfaces; they are not presented as a standalone dynamics substitute for AFM time scales.
Findings¶
The fitted ReaxFF is reported to track multireference energetics along disulfide mechanochemical pathways and to allow reactive MD of large mechanophores with disulfide breaking points. The exploratory trajectories are framed as narrowing part of the time-scale gap between AFM experiments and atomistic simulation when bond-order MD remains affordable for solvent-bearing constructs. Compared with common practice, the article stresses that training on multireference surfaces—not only routine DFT—matters for quantitative S–S rupture under load. Mechanophore design is a sensitivity lever: conducting DSM-C versus PEG-functionalized DSM-PEG motifs are proposed so that force-extension traces isolate disulfide events from competing pathways. The discussion flags transferability limits—each new reaction class still needs its own reference set and optimization cycle—and notes that agreement with a given AFM trace will still depend on laboratory spring constants, loading rates, and solvent chemistry.
Limitations¶
Parameter fits are specific to the chosen disulfide mechanochemistry targets; extending to other reaction classes requires new training data and optimization cycles. Quantitative agreement with a given AFM experiment still depends on instrument spring constants, loading rates, solvent environment, and anchor chemistry that are not universal across laboratories.
Relevance to group¶
The article shows a QM-reference upgrade path—multireference training targets—for ReaxFF in mechanochemistry, adjacent to broader reactive FF development practice in polymer pulling and sonochemistry communities that already use bond-order models. It is a useful citation when arguing that high-quality reference data, not only system size, gates accuracy for S–S rupture under load in van Duin-style ReaxFF workflows.
Citations and evidence anchors¶
- DOI: 10.1021/acs.jctc.6b00461 — J. Chem. Theory Comput. 12, 3913–3925 (2016); Supporting Information documents additional training geometries referenced in the main text.