Deep Learning Accelerated Determination of Hydride Locations in Metal Nanoclusters

Although the coordinates of the metal atoms can be accurately determined by X-ray crystallography, locations of hydrides in metal nanoclusters are challenging to determine. In principle, neutron crystallography can be employed to pinpoint the hydride positions, but it requires a large crystal and a neutron source, which prevents its routine use. Here, we present a deep-learning approach that can accelerate determination of hydride locations in single-crystal X-ray structure of metal nanoclusters of different sizes. We demonstrate the efficiency of our method in predicting the most probable hydride sites and their combinations to determine the total structure for two recently reported copper nanoclusters, [Cu₂₅ H₁₀ (SPhCl₂ )₁₈ ]^3- and [Cu₆₁ (S^t Bu)₂₆ S₆ Cl₆ H₁₄ ]⁺ whose hydride locations have not been determined by neutron diffraction. Our method can be generalized and applied to other metal systems, thereby eliminating a bottleneck in atomically precise metal hydride nanochemistry.