Discovering Ice and Water Structures with High-Resolution AFM, Atomistic Modeling, and Machine Learning

Since its invention several decades ago, Atomic Force Microscopy (AFM) has become an irreplaceable technique for the investigation of matter at the nanoscale. Specifically, the development of three-dimensional AFM (3D-AFM) enabled the observation of hydration structures in solid-liquid interfaces, while the use of tip functionalization in ultra-high vacuum has been crucial for reaching atomically resolved imaging of individual molecules. However, in both scenarios, only completely flat structures can be fully characterized by AFM. In more threedimensional samples, interpreting the measurements can be challenging, as only partial structural information is available. In this thesis, atomistic simulation and machine learning techniques are combined to tackle this problem in various systems, in all of which water molecules have central importance. First, a structure discovery workflow is developed for the case of ice nanoclusters on Au(111) and Cu(111) surfaces, centered about the use of neural network potentials. Then, molecular dynamics simulations are carried out to uncover the atomistic structure of cellulose-Iα and α-chitin nanocrystals surfaces in water. Finally, a high-throughput workflow is developed to identify the arrangement of solid-binding peptides assemblies on a graphite surface.