Abstract
The microstructure determines properties paradigm applies well to crystalline materials but struggles with amorphous systems. While researchers have long sought to link amorphous structures to macroscopic properties, traditional analyses focus on geometric packing, which our study reveals to be insufficient. We demonstrate this using two Pd-based metallic glasses, [Formula: see text] and [Formula: see text], which exhibit nearly identical geometries but different secondary relaxations. Electronic structure analysis uncovers the key distinction: [Formula: see text] has weaker Cu-P bonds and a less developed covalent network, enabling string-like atomic motions that drive pronounced relaxation, whereas [Formula: see text]'s stronger Ni-P interactions create a more constrained network. These findings highlight the critical role of electronic interactions and bonding fluctuations-beyond geometry-in governing glass dynamics. By integrating experiments and deep-learning simulations, we bridge the gap between local bonding heterogeneity and macroscopic behavior, offering new design principles for amorphous materials that prioritize electronic structure over purely geometric order. This advances glass physics by emphasizing the need to incorporate chemical interactions into structural analyses.