Development and application of multi-scale and machine learning approaches to investigate carbon-based tribofilm formation and nano-scale kinetic friction

Friction and wear arise from chemical and physical processes localized at nanoscale asperities, yet their macroscopic consequences extend over larger spatial and temporal scales. This thesis advances computational tribology through multiscale and data-driven simulation strategies capable of bringing quantum accuracy to relevant length and time scales under realistic operating conditions. A hybrid Green’s Function/Quantum Mechanical (GF/QM) framework is developed to describe reactive sliding interfaces with ab initio fidelity while embedding physically consistent dissipation into surrounding solids. By representing the bulk through surface Green’s functions, the method captures elastic response and phonon-mediated energy transport without artificial thermostats or prohibitively large atomistic models. Applications to hydrogen-passivated diamond–diamond contacts reveal that interfacial electronic structure controls frictional regimes, producing stick–slip motion along the minimum-energy path at low velocities and velocity-weakening continuous sliding at higher speeds. The thesis introduces the use of machine-learning interatomic potentials (MLIPs) for the simulation of complex tribological systems, extending near-ab initio accuracy to nanometric-size interfaces, lubricant mixtures, and tribochemical reactions. An open-source active-learning workflow is developed to automate dataset construction by identifying and labeling only the most informative configurations, enabling efficient exploration of chemically diverse and dynamically evolving interfaces. These methodological developments are applied to the tribochemistry of carbon-based lubricants. Multilevel simulations rationalize the mechanochemical activation of aromatic additives and their ability to form low-shear tribofilms. Emphasis is placed on water-based lubricants containing plant-derived aromatic molecules, investigated as eco-friendly alternatives to conventional oil-based formulations. Machine-learning simulations reveal the formation of nanoscale aggregates that evolve into resilient protective films under load, maintaining surface separation and reducing friction under severe conditions. Overall, the thesis establishes an atomistic computational framework for the design of sustainable, high-performance lubrication strategies, while providing simulation methods broadly transferable to a wide range of interfacial, reactive, and nonequilibrium material systems.