Thermal conductivities of monolayer graphene oxide from machine learning molecular dynamics simulations

TL;DR

This work presents NEP-GO, a neural-network potential trained on DFT data to enable large-scale reactive MD of graphene oxide reduction. By coupling thermal-reduction simulations with non-equilibrium MD and quantum-corrected spectral analysis, it reveals that GO’s thermal conductivity is governed by the initial oxidation state, with two main pathways: lattice restoration at higher OH/O and carbon-etching at higher O/C. Quantum corrections suppress high-frequency modes, reducing κ by roughly 45–60% across cases and yielding a range of 1.28–13.71 W m$^{-1}$ K$^{-1}$ for monolayer GO, significantly lower than pristine graphene. The methodology provides a tractable framework for defect-engineering heat transport in heterogeneous carbon materials and can be extended to multi-layer and bulk systems.

Abstract

Graphene oxide (GO) exhibits rich chemical heterogeneity that strongly influences its structural, thermal, and mechanical properties, yet quantitatively linking reduction chemistry to heat transport remains challenging. In this work, we develop a machine-learned neuroevolution potential (NEP) trained on an existing density functional theory dataset (\textit{Angew.\ Chem.\ Int.\ Ed.}, \textbf{63} , e202410088 (2024)), achieving reasonable accuracy at a computational cost much lower than the existing machine-learned and empirical potentials. Leveraging this potential, we perform large-scale molecular dynamics (MD) simulations to model the thermal reduction of GO across realistic structural domains. Using the homogeneous nonequilibrium MD method with a proper quantum-statistical correction scheme, we find that reduced GO exhibits strongly suppressed thermal conductivities, ranging from a few to tens of Wm$^{-1}$K$^{-1}$, substantially lower than pristine GO without defects and far below graphene. Moreover, the thermal conductivity of reduced GO increases moderately with increasing OH/O ratio, except at the highest oxidation level (O/C=0.5) where this trend inverts, while decreasing significantly with increasing O/C ratio, a trend strongly correlated with the fraction of recovered graphene-like structures. Our work provides a computationally tractable and predictive atomistic machine learning framework for exploring how chemical structure governs heat transport in heterogeneous carbon materials.

Figures (8)

• Figure 1: Comparison of accuracy and computational speed for , MACE, and models. (a) Correlation between reference data and predictions for total energy across the training set. (b-d) Comparison of predicted forces against data for (b) , (c) MACE and (d) across the train dataset. In panels (a)--(d), color intensity represents the local density of data points, and the corresponding and values are provided. (e) Computational speed as a function of the number of atoms in the structure. and MACE benchmarks were performed on a single RTX 4090 GPU (24 GB memory) using gpumd (version 4.3) and ase (version 3.23.0) packages, respectively; benchmarks were performed on 64 Xeon Platinum 8358P CPU cores (512 GB memory) using lammps package (version 29 Aug 2024)). The MACE and models were obtained from Refs. El-Machachi24accelarated and kowalik2019atomistic, respectively. (f) predictions for the structure with an O/C ratio of 0.4 and OH/O ratio of 0.5. The profile was calculated for annealed at 900K and equilibrated at 300K, using prediction server provided by Ref. golze2022accurate. The data were shifted horizontally to align with the experimental data from Ref. valentini2023tuning, while the MACE data were extracted from Ref. El-Machachi24accelarated.
• Figure 2: Evolution of gaseous byproducts during thermal reduction of at 900K with a fixed O/C ratio of 0.4. (a)-(e) Time-resolved production of H$_2$O, CO$_2$, CO, and other species during thermal reduction simulations, for initial OH/O ratios ranging from 0.1 to 0.5: (a) 0.1, (b) 0.2, (c) 0.3, (d) 0.4, and (e) 0.5. Each panel shows the average of five independent simulations (solid lines) with individual trajectories shown in translucence. (f) The ratio of graphene-like structures in the final configurations at 2000ps as a function of initial OH/O ratio.
• Figure 3: Evolution of gaseous byproducts during thermal reduction of at 900K with a fixed OH/O ratio of 0.3. (a)-(e) Time-resolved production of H$_2$O, CO$_2$, CO, and other species as during thermal reduction simulations, for initial O/C ratios ranging from 0.1 to 0.5: (a) 0.1, (b) 0.2, (c) 0.3, (d) 0.4, and (e) 0.5. Each panel shows the average of five independent simulations (solid lines) with individual trajectories shown in translucence. (f) The ratio of graphene-like structures in the final configurations at 2000ps as a function of initial O/C ratio.
• Figure 4: Thermal conductivity of structures obtained from simulations with a fixed initial O/C ratio of 0.4 and varying initial OH/O ratios. (a–e) Running thermal conductivity for OH/O ratio of (a) 0.1, (b) 0.2, (c) 0.3, (d) 0.4, and (e) 0.5, respectively. In each panel, translucent lines represent ten independent trajectories, while the solid dashed line shows the ensemble mean. The annotated values represent the average thermal conductivity and corresponding standard error calculated at 3ns. (f) Summary comparing the average classical (hollow circles with error bars) and quantum-corrected (solid dots) thermal conductivity as a function of the initio OH/O ratio.
• Figure 5: Thermal conductivity of structures obtained from simulations with a fixed initial OH/O ratio of 0.3 and varying initial O/C ratios. (a–e) Running thermal conductivity for O/C ratio of (a) 0.1, (b) 0.2, (c) 0.3, (d) 0.4, and (e) 0.5, respectively. In each panel, translucent lines represent ten independent trajectories, while the solid dashed line shows the ensemble mean. The annotated values represent the average thermal conductivity and corresponding standard error calculated at 3ns. (f) Summary comparing the average classical (hollow circles with error bars) and quantum-corrected (solid dots) thermal conductivity as a function of the initio O/C ratio.