Computational Investigations of Surfactants at Interfaces: Molecular Modeling and Machine Learning-Driven Discovery

Abstract

Per- and polyfluoroalkyl substances (PFAS) have become ubiquitous in industrial and consumer products due to their exceptional performance. However, their environmental persistence and adverse effects on human health have led to global restrictions, creating an urgent need for advanced strategies in PFAS detection, treatment, and replacement. This dissertation addresses these challenges using molecular dynamics (MD) simulations and machine learning to elucidate the mechanisms of PFAS adsorption for their detection and mitigation, and to accelerate the discovery of environmentally friendly surfactants to replace PFAS surfactants in firefighting foams. The first study investigates the adsorption of perfluorooctanoic acid (PFOA), a model PFAS molecule, on gold electrodes to inform the design of electrochemical sensors. Metadynamics simulations elucidate the two-dimensional free energy landscape near the electrode, revealing the preferred coplanar orientation of PFOA molecules with a minimum free energy of -83.9 kJ/mol. Notably, while spontaneous adsorption does occur, PFOA molecules can face significant energy barriers when reaching the most stable adsorbed state. These findings demonstrate that van der Waals interactions are the primary drivers for PFOA adsorption on gold electrodes, enabling sensitive electrochemical detection even at low concentrations. The second study focuses on the adsorption of PFOA molecules on graphene oxide (GO) and its derivatives to evaluate their efficacy as adsorbents for PFAS mitigation. By quantifying the adsorption strength and surface diffusion of PFOA molecules as functions of GO's oxygen-to-carbon (O/C) ratio and solution salinity, it is shown that while pristine graphene offers the strongest binding, favorable adsorption persists even at an O/C ratio of 16.7%. Unexpectedly, the in-plane diffusion coefficient of PFOA exhibits a non-monotonic trend, decreasing more than 45 times as the O/C ratio reaches 8.3% before increasing at higher oxidation levels. These results establish specific design rules for engineering graphene oxide–based materials that enhance both the adsorption capacity and the removal kinetics of PFAS in aqueous environments. The third study develops an active learning workflow that leverages coarse-grained (CG) MD simulations, Gaussian process regression, and Bayesian optimization to explore the vast design space for PFAS-free alternative surfactants in firefighting foams. From an initial chemical space of 2.8 million hydrocarbon surfactant candidates, a targeted library of 12,124 surfactants was screened for stability and synthetic feasibility. With this active learning framework, top-performing candidates were identified after exploring only 2.5% of the space after four rounds of active learning. The best candidate discovered offers an octane transport resistance of 1641.5 s/m at surfactant-saturated oil-water interfaces, which represents a substantial improvement over the commercial surfactant benchmark. Together, these studies provide a molecular-level understanding of PFAS interactions with a wide range of substrates and highlight the power of machine‑learning–accelerated molecular discovery. By integrating fundamental interfacial science with practical engineering strategies, this dissertation establishes critical design principles for the rational development of high‑sensitivity sensors and high‑capacity GO‑based adsorbents. The identification of promising surfactant candidates for firefighting foams offers guidelines to support experimental discovery of PFAS alternatives. In addition, the high‑throughput computational workflow demonstrated here provides a powerful framework for the rapid discovery of sustainable surfactants. Finally, releasing the associated models, tools, and datasets as open‑access resources will help accelerate future research and development toward PFAS‑free surfactants.