Abstract

Zinc dialkyldithiophosphate (ZDDP) remains one of the most effective antiwear additives, yet the tribochemical mechanisms leading to its characteristic tribofilm formation have been debated for over 80 years. A molecular-level understanding of these reactions is crucial for designing environmentally friendly alternatives with comparable performance. Although experimental studies have identified the main constituents of the ZDDP tribofilm—namely FeS and a glassy Zn/Fe polyphosphate network—the mechanistic details of its formation remain incomplete. Computational simulations offer a unique means to probe these buried interface phenomena; however, density functional theory (DFT) provides the required chemical accuracy only at prohibitive computational cost, while conventional molecular dynamics (MD) lacks the necessary fidelity. To address this challenge, we developed the first Machine Learning Potential (MLP) specifically tailored for modeling ZDDP tribofilm formation. Using this MLP, we investigated film growth on Fe interfaces of varying reactivity, including Fe(110), Fe(210), Fe surfaces with asperities, and oxidized Fe. Our results reveal that the formation of linkage isomers through alkyl-chain transfer—described as S-to-O substitution within the molecule—is a key step in generating the polyphosphate networks characteristic of ZDDP-derived tribofilms. These simulations provide the first comprehensive, atomistically resolved picture of the dynamic processes governing ZDDP tribofilm formation.