In the indirect coal liquefaction process, the residual wax produced by the Fischer-Tropsch synthesis reaction is a key byproduct that restricts system efficiency and economic viability. These residual waxes are classified into three primary sources: heavy oil filter cakes, generated during periodic filtration of slurry-bed reactors, which contain deactivated iron-based catalysts and high-carbon-number heavy wax oils; heavy wax filter cakes, formed in the separation unit from liquid products, comprising residual catalyst particles and waxy components; and catalyst replacement filter cakes, produced in the filtration unit after additive incorporation, primarily composed of heavy wax, additives, and trace catalysts. Efficient conversion of these residual waxes is one of the core challenges in optimizing the Fischer-Tropsch synthesis process. Notably, the oxidative pyrolysis behavior of the paraffin components (C₁₄−C₄₅) directly determines their energy conversion efficiency and resource recovery potential. Therefore, elucidating the oxidative pyrolysis mechanism of paraffins at the atomic scale holds dual significance for addressing wax deposition, regulating product selectivity, and developing clean fuel technologies-it pertains both to process intensification and the high-value utilization of byproducts.

However, existing research faces two major bottlenecks. First, traditional experimental methods struggle to capture transient free-radical reaction pathways (e.g., nanosecond-scale chain reactions) at high temperatures (>2000 K), as the lifetimes of key intermediates (such as peroxy radicals) are extremely short (<1 ns), exceeding the spatiotemporal resolution limits of conventional characterization techniques. Second, current molecular dynamics (MD) simulations are mostly limited to single-component models or narrow temperature ranges, failing to systematically resolve the correlation mechanisms between complex carbon chain distribution (C₁₄−C₄₅) in actual residual wax and oxidative pyrolysis behavior across a wide temperature range. Reactive force field (ReaxFF), which was constructed based on bond order concepts, can accurately describe chemical bond breaking/formation processes, providing an atomic-scale dynamic perspective for paraffin oxidative pyrolysis. Meanwhile, high-precision MD trajectory data can drive machine learning models to optimize force field parameters and predict product distributions, offering a new paradigm for industrial intelligent regulation.

To address these challenges, the paraffin oxidative pyrolysis dataset (see Table 1) constructed in this study achieves three pioneering innovations: (1) Multi-scale model adaptation: Based on the actual compositional characteristics of Fischer-Tropsch synthesis residual wax, three carbon chain distribution models (Simu1–Simu3) were constructed to directly correlate with compositional variations in industrial residual wax. (2) Wide temperature range dynamic trajectory coverage: The full oxidative pyrolysis pathways within the 2100−2500 K temperature range were recorded in detail (0.25 fs time step), filling the gap in high-temperature transient reaction mechanisms. (3) Intelligent analysis compatibility: Data formats are pre-adapted for ReacNetGenerator reaction network analysis and OVITO visualization tools, supporting machine learning-driven reaction prediction and coking mechanism exploration.

As a core demonstration of hydrocarbon molecular dataset, this dataset provides a theoretical cornerstone for the efficient conversion of Fischer-Tropsch synthesis residual wax by sharing high-fidelity simulation data, driving a paradigm shift in the coal chemical industry from process optimization to mechanism-driven advancement.

Through its multidimensional design, this dataset delivers systematic data support and analytical frameworks for investigating the mechanisms of paraffin oxidative pyrolysis. Its core value is first demonstrated by a multi-model comparison framework, in which users can quantitatively assess the influence of carbon chain length on reaction pathways based on three initial structures with distinct carbon chain distributions (Simu1−Simu3). This capability offers mechanistic insights into the differences in reactivity among residual waxes. Second, the dataset enables comprehensive reaction trajectory analysis, recording atomic motion with a temporal resolution of 0.1 ps. This high-resolution data allows for the tracking of dynamic evolution processes of key intermediates-such as alkyl and peroxy radicals-thereby supplying dynamic evidence for elucidating the microscopic mechanisms of free-radical chain reactions. Third, the dataset covers five temperature points (2100−2500 K), forming a complete kinetic temperature series. Users can extract species concentration data from the “thermostat” folder to construct Arrhenius curves, accurately determine the activation energies of elementary reactions such as C−C bond cleavage, and identify potential reaction pathway transition thresholds around 2300 K. Examples include the shift from dehydrogenation-dominated mechanisms at lower temperatures to direct bond cleavage mechanisms at higher temperatures. These findings provide a theoretical basis for optimizing oxidative pyrolysis temperatures to mitigate excessive oxidation. In terms of analytical compatibility, the dataset’s output formats are directly supported by reaction network analysis tools such as ReacNetGenerator, and are also compatible with bond-order visualization modules in OVITO software. This enables intuitive identification of the spatiotemporal distribution characteristics of coking precursors-such as polycyclic aromatic hydrocarbons-facilitating cross-scale analysis from the atomic level to mesoscopic aggregation behavior. Furthermore, the dataset can be utilized to construct macroscopic kinetic models for residual wax oxidative pyrolysis or to train machine learning models for rapid prediction of product yields under various operating conditions, thereby directly supporting process optimization and reactor design. In summary, by integrating multiple models, high-resolution trajectories, a broad temperature range, and standardized analytical interfaces, this dataset offers a comprehensive and reliable data platform for studying reaction kinetics in paraffin oxidative pyrolysis and for developing coking inhibition strategies.

The data in this dataset were obtained via reactive molecular dynamics simulations. The construction of the dataset involved an integrated workflow comprising three main stages: initial data preparation, data generation, and data processing. A schematic overview of the overall procedure is presented in Figure 1.

Based on the experimental data (see Supplementary Table S1), we constructed three three-dimensional periodic paraffin models (Simu1, Simu2, and Simu3), each exhibiting distinct hydrocarbon distributions. The initial simulation box dimensions were 126.88 Å × 126.88 Å × 126.88 Å. Periodic boundary conditions were applied in all three spatial directions throughout the simulation to accurately represent bulk behavior. Only straight-chain alkane structures were considered in the paraffin models. The distribution of carbon chain lengths, ranging from C₁₄ to C₄₅, was consistent with the experimental observations (see Table 2). Following the addition of 2000 O₂ molecules to each model, structural optimization was first performed using the COMPASS force field ^[1−2]. The systems were then subjected to annealing by heating from 300 to 1000 K under an NVT ensemble. Finally, the optimized structures were equilibrated in an NPT ensemble at 1000 and 0.8 GPa. After relaxation, the model density was determined to be 0.92 g/cm³, and the final box dimensions were approximately 60.369 Å × 60.39 Å × 60.39 Å.

All reactive molecular dynamics simulations were carried out using the LAMMPS^[3] software. The CHO reactive force field^[4] was employed with a time step of 0.25 fs. Temperature control was achieved via an NVT ensemble combined with a Nosé-Hoover thermostat, using a damping constant of 100 fs. The Verlet algorithm was used for MD integration. Initially, all models were structurally relaxed for 100 ps at 300 K. The relaxed configurations were subjected to a heating protocol, starting from 300 K and increasing to final temperatures of 2100, 2200, 2300, 2400 and 2500 K. The heating phase was conducted over 400 ps. The simulation temperatures were set significantly above the experimental values to accelerate reaction kinetics, thereby enabling the observation of a larger number of chemical events within the limited simulation timeframe.

Subsequently, the high-temperature structures obtained at each target temperature were maintained under isothermal conditions for 600 ps to simulate oxidative pyrolysis. Molecular products were identified using a bond-order cutoff of 0.3. Species information and trajectory data were recorded every 400 steps (corresponding to 0.1 ps). Reaction analysis was performed using ReacNetGenerator^[5], and molecular structures were visualized with the OVITO^[6] software.

The data in this dataset are organized according to a three-level hierarchical structure, corresponding to the three paraffin component distributions (Simu1, Simu2, Simu3). For each component model, simulation results are provided for five temperature points: 2100, 2200, 2300, 2400, 2500 K. The data for each temperature point are further categorized into two phases: the heating phase (heat) and the constant-temperature oxidative pyrolysis phase (thermostat). Within each phase folder, a frame data directory (frames-heat or frames-thermostat) contains 4001 and 6001 structure files (named frame*.data), respectively. These files record atomic configurations and evolution of products at different time steps. In addition, the remaining 10 files with the ".dump.*" suffix store post-processing information derived from the frame data, including reaction network topologies and species statistics. Detailed descriptions of the data storage structure and analysis reports are provided in Table 3 and Figure 2−5. For definitions of numerical values and symbols, readers are referred to the methodology established in reference [5].

To ensure the reliability and reproducibility of reactive molecular dynamics simulation data for paraffin oxidative pyrolysis, systematic quality control measures were implemented throughout the entire simulation workflow. For the force field, the high-accuracy CHO-ReaxFF reactive potential, parameterized using first-principles calculations, was employed to accurately describe the reaction pathways of carbon, hydrogen, and oxygen species at elevated temperatures. Prior to reaction simulations, all systems underwent structural relaxation for 100 ps to eliminate initial configuration deviations and ensure that both energy and structure reached equilibrium. Temperature control was maintained using a Nosé-Hoover thermostat with a damping constant of 100 fs. Fluctuations were verified to remain below ±1% at 2100 K (Figure 6), satisfying the stability requirements for reactive dynamics simulations. To guarantee the integrity of trajectory data, full atomic coordinates and bond-order information were recorded every 400 steps (0.1 ps).

A uniform bond-order cutoff of 0.3 was applied to ensure consistent species identification across all simulations. Furthermore, the energy conservation error under a 0.25 fs integration timestep was maintained below 0.5% (Figure 6), meeting the numerical precision criteria for short-range dynamics simulations in LAMMPS. Collectively, these measures ensure that the simulation results fulfill the requirements for scientific research in terms of physical plausibility and data reliability.

To maximize the utility of this dataset in investigating the mechanisms of paraffin oxidative pyrolysis, users are advised to adopt the following integrated analytical workflow. This workflow comprises a full-chain methodology spanning atomic-scale trajectory visualization, reaction network reconstruction, thermodynamic analysis, and machine learning model development, with the aim of systematically elucidating the oxidative cracking behavior and regulatory principles of paraffin at high temperatures. Atomic motion trajectories can be parsed and visualized using OVITO software (example file path: Simulation/Simu1_2100K/thermostat/frames-thermostat/frames.*.data). By utilizing its built-in bond-order analysis function (recommended bond-order threshold: 0.3), users can dynamically track the bond-breaking and bond-forming sequences of key elements (C, H, and O) during the reaction process, thereby directly observing the evolution pathways and lifetimes of radical intermediates (e.g., alkyl radicals).

In the reaction network analysis stage, users can directly open the generated .html format report files in a web browser. These files, automatically generated by ReacNetGenerator, present reaction pathways, key species, and their occurrence frequencies in an interactive graphical format. By examining the topological changes in network structures under different temperatures or models-such as high-frequency reaction pathways and dominant reaction channels-users can identify the formation pathways of coking precursors and the competitive relationship between oxidation and oxidative pyrolysis reactions.

To quantify the effect of temperature on reaction pathways, it is recommended to extract and compare product distribution data files (e.g., processed_species-thermostat.csv) from folders corresponding to different temperatures (e.g., Simu1/2100K and Simu1/2500K). Analyzing the concentration variations of specific products (e.g., C₂H₄, CO₂) with temperature enables the calculation of thermodynamic parameters such as apparent activation energies, thereby revealing mechanistic shifts with temperature (e.g., the transition from peroxy radical-dominated oxidation at lower temperatures to direct C–C bond cleavage at higher temperatures). For detailed analytical procedures, readers are referred to the literature “Atomic-scale insight into thermal decomposition behavior of polypropylene: A ReaxFF method”. All raw .data trajectory files in this dataset can serve as input for machine learning models. Users may convert these files into formats suitable for model training (e.g., feature vectors describing local atomic environments) and subsequently develop predictive models. Such models can be employed to rapidly predict the equilibrium product distributions of paraffin oxidative pyrolysis under varying initial compositions and temperature conditions, providing an efficient computational tool for materials design and process optimization.

This dataset constructs an oxidative pyrolysis trajectory dataset via molecular dynamics simulations, covering three paraffin components across a temperature range of 2100−2500 K, providing multiscale atomic dynamic information for elucidating hydrocarbon oxidative pyrolysis mechanisms. Its multi-model design supports quantitative analysis of how carbon chain length influences reaction pathways; high-resolution trajectories comprehensively record the evolution of free radicals; wide temperature coverage facilitates the extraction of kinetic parameters and identification of mechanistic transition thresholds; and the standardized storage structure paired with compatible formats ensures the ready-to-use nature and reproducibility of the data. Not only does this dataset provide an atomic-scale basis for fundamental research on paraffin oxidative pyrolysis, but it also lays a data foundation for industrial process optimization and the exploration of machine learning-driven new paradigms.