Data-Driven Discovery of High-Performance EOR Surfactants: Coupling Molecular Dynamics with Active Learning

Abstract Chemical Enhanced Oil Recovery (EOR) is a central route to maximizing hydrocarbon recovery from mature reservoirs. Although its field-scale success depends on several factors─including surfactant cost, adsorption and retention on the reservoir rock, solubility, and phase behavior─achieving a low oil–water interfacial tension (IFT) is a key enabling requirement. However, the design of surfactants meeting this requirement is impeded by the vast combinatorics of the chemical space and the prohibitive cost of traditional trial-and-error screening. In this study, we present a synergistic materials informatics framework that integrates Machine Learning (ML) and Molecular Dynamics (MD) simulations within an Active Learning (AL) workflow to accelerate surfactant discovery. Unlike static screening approaches, our methodology utilizes MD simulations as a high-fidelity oracle, iteratively labeling a subset of candidates to update a regressor, enabling the identification of high-performance surfactants from the PubChem database. Through this workflow, we identified Dodecylammonium Chloride (DDACl) as a top candidate; subsequent high-fidelity MD simulations confirmed a very low IFT at maximum surface coverage, outperforming widely used analogs such as Cetyltrimethylammonium Bromide (CTAB). We validated these computational findings through experimental tensiometry, confirming the efficacy of DDACl in reducing IFT in several oils with varying API gravities. The convergence of theoretical predictions with experimental results underscores the power of this data-driven framework to overcome the “small data” bottleneck in materials science, offering a scalable pathway for the data-driven discovery of surfactants with low oil–water IFT as candidates for EOR formulation development.