Structure Discovery in Atomic Force Microscopy Imaging of Ice

Abstract

The interaction of water with surfaces is crucially important in a wide range of natural and technological settings. In particular, at low temperatures, unveiling the atomistic structure of adsorbed water clusters would provide valuable data for understanding the ice nucleation process. Using high-resolution atomic force microscopy (AFM) and scanning tunneling microscopy, several studies have demonstrated the presence of water pentamers, hexamers, and heptamers (and of their combinations) on a variety of metallic surfaces, as well as the initial stages of 2D ice growth on an insulating surface. However, in all of these cases, the observed structures were completely flat, providing a relatively straightforward path to interpretation. Here, we present high-resolution AFM measurements of several water clusters on Au(111) and Cu(111), whose understanding presents significant challenges due to both their highly 3D configuration and their large size. For each of them, we use a combination of machine learning, atomistic modeling with neural network potentials, and statistical sampling to propose an underlying atomic structure, finally comparing its AFM simulated images to the experimental ones. These results provide insights into the early phases of ice formation, which is a ubiquitous phenomenon ranging from biology to astrophysics.

Automated Structure Discovery for Scanning Tunneling Microscopy

“…The first option, which this work highlights, is to create a large, diverse, and descriptive molecular data set of various chemical species and structures and to train a versatile model with the aim of predicting the structure of almost any small organic molecule. This approach has also been chosen in most previous sample characterization efforts (e.g., , ), but recently, another method has been used in ice structure discovery, , where instead of a diverse data set, a tailored data set is utilized and perfected to make very accurate predictions possible in a constrained problem domain. That is, if the goal was to predict only the geometries of different hydrocarbons or triangulene-based molecules, the model would benefit from a tailored data set with a heavy emphasis on such structures.…”

“…However, the accuracy in the prediction was excellent, which suggests that even though images in the training set are not entirely representative of true images in terms of electronic structure, they capture the characteristic sharp lines coming from CO tip bending accurately, and this seems to be critical for structure discovery. Still, we do note that for systems experiencing significant charge transfer or orbital hybridization ASD-STM is most likely not a sufficient tool, and a more comprehensive approach is needed . Also, synthetic AFM images of isolated molecules in general correspond well to experimental AFM images where the substrate is naturally present, and since for some samples it is possible to simultaneously gather STM and AFM signals, the possibility of incorporating AFM data into the training process should be explored.…”

“…Still, we do note that for systems experiencing significant charge transfer or orbital hybridization ASD-STM is most likely not a sufficient tool, and a more comprehensive approach is needed. 41 Also, synthetic AFM images of isolated molecules in general correspond well to experimental AFM images where the substrate is naturally present, and since for some samples it is possible to simultaneously gather STM and AFM signals, the possibility of incorporating AFM data into the training process should be explored.…”