Nanocatalysts have found broad applications, ranging from fuel cells to catalytic converters and hydrogenation reactions. Understanding the active sites and local environment of nanocatalysts has attracted considerable attention from theoretical, computational and experimental studies. However, characterizing the three-dimensional (3D) atomic and chemical arrangement of different constituents, as well as structural reconstructions driven by catalytic reactions, remains elusive. Although a number of experimental methods have been used to study nanocatalysts, each has its limitations. X-ray absorption spectroscopy (XAS) is an important in situ/operando technique for probing the structure and composition of nanocatalysts, but it globally averages a large number of samples and isotopically characterizes nanocatalysts from the surface to the core. In situ and environmental transmission electron microscopy (TEM) can image the local heterogeneous structure of nanocatalysts in gas- and liquid-phase reactions, but the resolution and contrast of the images are often hindered. To achieve the highest possible spatial resolution, scanning TEM (STEM) and TEM are the methods of choice and have been widely used to characterize the structure and composition of nanocatalysts in a vacuum. However, both in situ and environmental TEM, as well as high-resolution S/TEM, provide only two-dimensional projection images of 3D nanocatalysts. Currently, atomic electron tomography (AET) is the only experimental method that is capable of resolving all of the 3D atomic positions of individual nanoparticles.

Here we use AET to determine the 3D atomic structures of platinum (Pt) alloy nanocatalysts for the oxygen reduction reaction (ORR) in proton exchange membrane fuel cells. We investigate the nanocatalysts before and after catalytic activation, observing their 3D structural and compositional changes at the atomic scale (seeMethods). The experimental 3D atomic structures obtained after catalytic activation are used for density functional theory (DFT)-trained machine learning (ML) to identify the active sites of the nanocatalysts.