In situ transmission electron microscopy (TEM) provides a means for capturing atomic-resolution micrographs under reactive environments and has been widely employed to identify active site candidates in heterogeneous catalysis. We highlight here that the limited temporal resolution of in situ TEM fundamentally restricts its ability to resolve transient surface structures that may govern catalytic activity. We demonstrate that in situ TEM cannot capture dynamic active sites that evolve on sub-nanosecond time scales by considering two systems: (i) Cu adatoms on Cu(111) at 0 and 100 bar CO pressure at 573 K and (ii) N2 decomposition on Fe(111) at 800 K. We study both systems via molecular dynamics using machine learning interatomic potentials trained on density functional theory data. Our results show that rapid surface rearrangements are averaged out in simulated TEM micrographs even under idealized imaging conditions. As a result, the resulting transient sites are rendered effectively invisible under in situ conditions. The findings of this study highlight an intrinsic limitation of in situ TEM and emphasize the necessity of additional in situ characterization as well as computational techniques to understand catalyst dynamics.

Idealized models such as the Wulff construction provide inaccurate structures for small supported nanoparticles (NPs). Global optimization via machine learning interatomic potentials (MLIPs) enables ...