Thesis Defense of Gilberto ALOU

Alou Gilberto - PhLAM Laboratory - UMR 8523 - PCMT research team

Title : From Ab Initio Molecular Dynamics to Machine Learning Potential Energy Surface-Based Simulations: NO Oxidation on Oxygen-Functionalized Graphite

Thesis committee : M. MONNERVILLE (PhLAM, supervisor), C.TOUBIN (PhLAM, supervisor), A. SANTAMARIA RIVERO (PhLAM, supervisor), C. CRESPOS (Université de Bordeaux, reviewer), C. CLAVAGUERA (Université Paris-Saclay, reviewer), P. LARREGARAY (Université de Bordeaux, committee member), D. PETITPREZ (Université de Lille, committee member), S. MORISSET (Université Paris-SAcaly, committee member)

Summary : 

Oxygen-functionalized graphite surfaces (O@HOPG) are commonly found in the atmosphere due to exposure to oxidizing pollutants such as ozone (O₃) and hydrogen peroxide (H₂O₂). These surface-bound oxygen groups play a crucial role in atmospheric chemistry, as they can react with a broad range of pollutants, including hydrocarbons and nitrogen oxides (NOₓ). Among these, NOₓ are major contributors to smog and acid rain; nitric oxide (NO) is a precursor to nitrogen dioxide (NO₂), which also promotes the formation of secondary organic aerosols—known to negatively impact both human health and the environment.

This thesis presents a theoretical investigation of the oxidation of NO molecules on epoxy-functionalized HOPG, with relevance to heterogeneous atmospheric chemistry and pollution abatement. The study relies on an approach combining Density Functional Theory (DFT)-based ab initio Molecular Dynamics (AIMD) with a machine learning potential energy surface (ML-PES). Four collision energies (0.025, 0.05, 0.1, and 0.3 eV) and two different orientations were analyzed where the reaction, adsorption, and scattering probabilities were computed.

AIMD simulations reveal that NO₂ formation can occur even at the lowest collision energy investigated (0.025 eV), approximately equivalent to room temperature (300 K), which agrees qualitatively with the experimental results. Additionally, simulations show that scattered NO molecules exhibit low specular reflection, lose half of their initial translational energy, and remain vibrationally cold with minimal rotational excitation. Furthermore, a statistical analysis of all reactive trajectories, elucidated the structural requirements for the reaction to efficiently occur. Finally, this theoretical study demonstrates the potential of oxygen-doped carbon surfaces for the conversion of NO to NO₂highlighting the importance of explicitly accounting for dynamical effects on the NO oxidation.

To circumvent the high computational cost of AIMD simulations and the limited statistical sampling, a ML-PES was constructed and trained on configurations extracted from AIMD trajectories. The accuracy of the ML-PES was rigorously validated by direct comparison with AIMD results, confirming its ability to reproduce both reactive and non-reactive events with high fidelity. The ML-PES successfully captures the essential dynamical and energetic features of the NO + O@HOPG -> NO₂ + HOPG reaction, accurately identifying reaction channels and characterizing energy transfer mechanisms, while reducing the computational cost by approximately two orders of magnitude compared to conventional AIMD simulations. The high efficiency of ML PES simulation makes it possible to enhance the sampling of reactive events, thereby providing statistically robust insights into the underlying mechanisms of NO oxidation. Nevertheless, limitations persist in regions of the configurational space that are underrepresented in the training data. To overcome this, an active learning strategy is proposed to iteratively refine the ML-PES by incorporating dynamically important configurations, thereby improving the model's robustness and transferability.

This work establishes a robust computational framework that combines AIMD and ML-PES for the atomistic simulation of gas–surface reactions. While this thesis concentrated on the NO oxidation on O@HOPG, the developed methodology is broadly transferable and can be extended to study other gas-surface reactions of atmospheric relevance within the context of depollution or catalysis.