The macroscopic activation energy for diffusion in glasses arises not from local rearrangement barriers but from back-and-forth correlated motion imposed by asymmetric energy landscapes. Molecular dynamics simulations across metallic glasses, silica, and Lennard-Jones systems decompose diffusion into random-walk and correlation components, revealing that the correlation contribution --- the tendency for atoms to reverse their displacements --- dominates the activation energy. The asymmetry between forward and reverse barriers in a disordered potential landscape makes return trips thermodynamically cheaper than continuation, trapping diffusing particles in oscillatory patterns that inflate the apparent activation barrier far above any individual hop cost.
The standard picture of activated diffusion --- a particle surmounting a barrier to reach a neighboring site --- assumes symmetric landscapes where forward and backward barriers are comparable. Disorder breaks this symmetry systematically: if the landscape is random, then the barrier you crossed to arrive is statistically different from the barrier to continue. The return penalty is lower, so the particle preferentially returns. This transforms a local picture (barrier height sets the rate) into a global one (landscape asymmetry sets the correlation, and correlation sets the rate). The mechanism is universal across material classes precisely because it depends on structural disorder itself rather than on any specific chemistry.
Reader note (2026-11-17): Checked against the full text of arXiv:2603.18317 (Annamareddy, Wang, Voyles, Szlufarska and Morgan; v1 of March 2026), which I had earlier held as matching on an abstract-level read. The mechanism matches: the activation energy of diffusion splits into a random-walk part and a correlation part, and the correlation part, which comes from forward barriers being higher than the reverse barriers the atoms then meet, dominates. The paper gives the extent more precisely than the essay. In its main system, Cu50Zr50, the total diffusion barrier is about 1.22 eV, the random-walk part about 0.43 eV and the correlation barrier about 65 percent of the total; in Ni80P20 the correlation barrier is about 50 percent. The simulations cover a narrow temperature window near the glass transition (640 and 700 K for Cu50Zr50), and the paper cautions that the apparent Arrhenius behavior there is a consequence of the limited window and should not be read as a single microscopic barrier. The silica and single-component Lennard-Jones glasses are studied in the supplementary information, and the paper's wording is that the correlation effects are general across the four chemically and structurally distinct systems it simulated; the essay's universal across material classes extends that to all materials.
Systems with disordered energy landscapes suffer a hidden tax: the cost of exploration is not the height of individual barriers but the statistical tendency of random landscapes to make retreat cheaper than advance. This principle applies wherever navigation occurs through heterogeneous terrain --- the effective friction arises from the correlation structure of the landscape, not from the individual obstacles. Any agent moving through disorder will overestimate local barriers and underestimate landscape-scale trapping, because the return penalty is invisible to any measurement that considers only forward hops.
(arXiv:2603.18317)