Generalization Across Experimental Parameters in Neural Network Analysis of High-Resolution Transmission Electron Microscopy Datasets.

Neural networks are promising tools for high-throughput and accurate transmission electron microscopy (TEM) analysis of nanomaterials, but are known to generalize poorly on data that is "out-of-distribution" from their training data. Given the limited set of image features typically seen in high-resolution TEM imaging, it is unclear which images are considered out-of-distribution from others. Here, we investigate how the choice of metadata features in the training dataset influences neural network performance, focusing on the example task of nanoparticle segmentation. We train and validate neural networks across curated, experimentally collected high-resolution TEM image datasets of nanoparticles under various imaging and material parameters, including magnification, dosage, nanoparticle diameter, and nanoparticle material. Overall, we find that our neural networks are not robust across microscope parameters, but do generalize across certain sample parameters. Additionally, data preprocessing can have unintended consequences on neural network generalization. Our results highlight the need to understand how dataset features affect deployment of data-driven algorithms.

Understanding the Influence of Receptive Field and Network Complexity in Neural Network-Guided TEM Image Analysis.

Trained neural networks are promising tools to analyze the ever-increasing amount of scientific image data, but it is unclear how to best customize these networks for the unique features in transmission electron micrographs. Here, we systematically examine how neural network architecture choices affect how neural networks segment, or pixel-wise separate, crystalline nanoparticles from amorphous background in transmission electron microscopy (TEM) images. We focus on decoupling the influence of receptive field, or the area of the input image that contributes to the output decision, from network complexity, which dictates the number of trainable parameters. For low-resolution TEM images which rely on amplitude contrast to distinguish nanoparticles from background, we find that the receptive field does not significantly influence segmentation performance. On the other hand, for high-resolution TEM images which rely on both amplitude and phase-contrast changes to identify nanoparticles, receptive field is an important parameter for increased performance, especially in images with minimal amplitude contrast. Rather than depending on atom or nanoparticle size, the ideal receptive field seems to be inversely correlated to the degree of nanoparticle contrast in the image. Our results provide insight and guidance as to how to adapt neural networks for applications with TEM datasets.

Driving energetically unfavorable dehydrogenation dynamics with plasmonics.

Nanoparticle surface structure and geometry generally dictate where chemical transformations occur, with higher chemical activity at sites with lower activation energies. Here, we show how optical excitation of plasmons enables spatially modified phase transformations, activating otherwise energetically unfavorable sites. We have designed a crossed-bar Au-PdH x antenna-reactor system that localizes electromagnetic enhancement away from the innately reactive PdH x nanorod tips. Using optically coupled in situ environmental transmission electron microscopy, we track the dehydrogenation of individual antenna-reactor pairs with varying optical illumination intensity, wavelength, and hydrogen pressure. Our in situ experiments show that plasmons enable new catalytic sites, including dehydrogenation at the nanorod faces. Molecular dynamics simulations confirm that these new nucleation sites are energetically unfavorable in equilibrium and only accessible through tailored plasmonic excitation.

In-situ observation of plasmon-controlled photocatalytic dehydrogenation of individual palladium nanoparticles.

Plasmonic nanoparticle catalysts offer improved light absorption and carrier transport compared to traditional photocatalysts. However, it remains unclear how plasmonic excitation affects multi-step reaction kinetics and promotes site-selectivity. Here, we visualize a plasmon-induced reaction at the sub-nanoparticle level in-situ and in real-time. Using an environmental transmission electron microscope combined with light excitation, we study the photocatalytic dehydrogenation of individual palladium nanocubes coupled to gold nanoparticles with sub-2 nanometer spatial resolution. We find that plasmons increase the rate of distinct reaction steps with unique time constants; enable reaction nucleation at specific sites closest to the electromagnetic hot spots; and appear to open a new reaction pathway that is not observed without illumination. These effects are explained by plasmon-mediated population of excited-state hybridized palladium-hydrogen orbitals. Our results help elucidate the role of plasmons in light-driven photochemical transformations, en-route to design of site-selective and product-specific photocatalysts.

Visualizing Facet-Dependent Hydrogenation Dynamics in Individual Palladium Nanoparticles.

Surface faceting in nanoparticles can profoundly impact the rate and selectivity of chemical transformations. However, the precise role of surface termination can be challenging to elucidate because many measurements are performed on ensembles of particles and do not have sufficient spatial resolution to observe reactions at the single and subparticle level. Here, we investigate solute intercalation in individual palladium hydride nanoparticles with distinct surface terminations. Using a combination of diffraction, electron energy loss spectroscopy, and dark-field contrast in an environmental transmission electron microscope (TEM), we compare the thermodynamics and directly visualize the kinetics of 40-70 nm {100}-terminated cubes and {111}-terminated octahedra with approximately 2 nm spatial resolution. Despite their distinct surface terminations, both particle morphologies nucleate the new phase at the tips of the particle. However, whereas the hydrogenated phase-front must rotate from [111] to [100] to propagate in cubes, the phase-front can propagate along the [100], [11̅0], and [111] directions in octahedra. Once the phase-front is established, the interface propagates linearly with time and is rate-limited by surface-to-subsurface diffusion and/or the atomic rearrangements needed to accommodate lattice strain. Following nucleation, both particle morphologies take approximately the same time to reach equilibrium, hydrogenating at similar pressures and without equilibrium phase coexistence. Our results highlight the importance of low-coordination number sites and strain, more so than surface faceting, in governing solute-driven reactions.

Weakly explosive percolation in directed networks.

Percolation, the formation of a macroscopic connected component, is a key feature in the description of complex networks. The dynamical properties of a variety of systems can be understood in terms of percolation, including the robustness of power grids and information networks, the spreading of epidemics and forest fires, and the stability of gene regulatory networks. Recent studies have shown that if network edges are added "competitively" in undirected networks, the onset of percolation is abrupt or "explosive." The unusual qualitative features of this phase transition have been the subject of much recent attention. Here we generalize this previously studied network growth process from undirected networks to directed networks and use finite-size scaling theory to find several scaling exponents. We find that this process is also characterized by a very rapid growth in the giant component, but that this growth is not as sudden as in undirected networks.