Timescale prediction of complex multi-barrier pathways using flux sampling molecular dynamics and 1D kinetic integration: Application to cellulose dehydration
The authors combine flux sampling on a 1D reaction coordinate with kinetic integration to extrapolate rare-event kinetics for ReaxFF-based reactive MD beyond brute-force time scales. The showcase is crystalline cellulose dehydration/decomposition at 1500–1900 K, comparing product distributions to replica-exchange MD and fitting an Arrhenius regime with Eₐ ≈ 93 kcal mol⁻¹ and k₀ ≈ 9×10¹⁹ s⁻¹, while noting breakdown of simple Arrhenius behavior when the order parameter becomes inadequate at lower temperature.
Summary¶
Reactive MD with ReaxFF can show multi-step pyrolysis chemistry, but direct MD cannot cover slow events near moderate T. Flux sampling plus 1D kinetic integration bridges time scales. For I-β cellulose, the protocol yields H₂O, CO, and CO₂ product patterns consistent with replica-exchange results where decomposition completes, and an Arrhenius form with parameters larger than some experiments—a hint of FF or path limits. At 1500 K, full decomposition is not reached in the runs summarized; low-T kinetics can deviate from a single Arrhenius line when the order parameter fails to capture multi-barrier kinetics.
Methods¶
1 — MD application. LAMMPS with the 2013 C–C Reaxff parameters cited in the paper. I-β cellulose supercell with PBC (three-dimensional periodic cell as in the article); NPT targeting ~2.5 GPa isotropic hydrostatic pressure (density matched to 1900 K REM reference behavior). Time step 0.1 fs. Flux sampling on milestones along a reaction coordinate (released gas molecules counted); 10 ps MD segments for each flux estimate; kinetic integration of forward/backward rates along the coordinate (Sections 2–3). T sweeps 1500–1900 K in 100 K steps. N/A — static external electric field; N/A — umbrella in the sense of traditional bias potentials (the method is flux sampling; confirm wording in PDF). Thermostat and barostat families are given in the article for the NPT stages—see ## Limitations if a label is not copied in one line here. N/A if NVE brute-force segments: not the dominant method—flux sampling windows are short NPT-style as stated.
2 — Force-field training. N/A — the manuscript uses a published Reaxff for cellulose-relevant C–O/H chemistry; N/A for a de novo GA refit in this article.
3 — Static QM. N/A for standalone DFT in the main workflow; Reaxff is empirical.
4 — Review or non-simulation. N/A — method + application article.
Findings¶
Outcomes and mechanisms. Flux sampling + kinetic integration extends reachable time scales for rare rearrangements vis-à-vis brute-force MD at comparable cost in this problem class. H₂O/CO/CO₂ distributions align with REM-based reference where full pyrolysis completes. An Arrhenius fit yields Eₐ ≈ 93 kcal mol⁻¹ and k₀ ≈ 9×10¹⁹ s⁻¹, larger than typical experiments as noted. Decomposition incomplete at 1500 K in the stated protocol. Arrhenius breakdown at lower T is tied to the reaction coordinate not capturing the true multi-barrier manifold—a method limit as the authors discuss.
Comparisons and sensitivity. Brute-force MD benchmarks at high T; T-dependent kinetics; pressure via NPT 2.5 GPa design.
Authored limitations and outlook. Order-parameter error and Reaxff accuracy near pyrolysis chemistry; see ## Limitations on the page.
Corpus honesty. Figures (e.g. Figure 8) in pdf_path for quantitative flux and brute-force comparisons**.
Limitations¶
Order-parameter sensitivity is explicit: incorrect or too coarse coordinates can distort inferred rates. ReaxFF chemistry for pyrolysis remains approximate relative to QM.
Relevance to group¶
van Duin co-authorship; couples ReaxFF pyrolysis with enhanced sampling / kinetic integration methodology.
Citations and evidence anchors¶
- DOI:
10.1063/1.5126391