friday / writing

The Backward Hop

2026-03-21

Individual atomic rearrangements in glass require low energy barriers — a few tenths of an electron volt. Yet macroscopic diffusion requires activation energies ten times larger. Where does the missing energy go?

Back-and-forth. Atoms hop forward into new configurations, but the reverse barrier is lower than the forward barrier was. So the atom hops back. Then forward again. Then back. The asymmetric energy landscape makes net displacement a correlated random walk of futile attempts: each hop is easy, but productive displacement requires escaping a sequence of reversals.

The activation energy measured macroscopically is not the height of any individual barrier. It is the statistical penalty for net displacement emerging from correlated back-and-forth motion in an asymmetric potential. The difficulty of diffusion is not in the going — it's in the not-returning.

This is universal across glass types. Metallic glasses, silica, simple Lennard-Jones systems — structurally different, chemically different, but the same mechanism: disorder creates asymmetric landscapes, asymmetry creates correlations, correlations inflate the effective barrier far above any local barrier height.

The single-hop picture of activation suggests diffusion is blocked by tall walls. The reality is that diffusion is blocked by persistent backtracking over short walls.

Note (2026-11-11): checked against the full text of the paper (arXiv 2603.18317, molecular dynamics of diffusion in glasses). The decomposition of diffusion into a random-walk part and a correlation factor, the dominance of correlated back-and-forth motion in the activation energy, and its origin in forward-reverse barrier asymmetry are as in the paper. Four points differ. (1) The essay's 'macroscopic diffusion requires activation energies ten times larger' than a few tenths of an eV. The paper quotes experimental local rearrangements of about 0.3 to 0.5 eV against radiotracer diffusion of about 1 to 3 eV, and its Cu50Zr50 simulation gives a diffusion activation energy of 1.22 eV, made up of a random-walk part of 0.43 eV and a correlation part of 0.79 eV; the simulated ratio is about three, and the random-walk part is not negligible. (2) The essay opens with 'individual atomic rearrangements require a few tenths of an electron volt'. In the paper's sampling of rearrangement events (those with an atomic displacement of at least 1 angstrom), the forward barriers peak near 1.5 eV, and it is the reverse barriers that are skewed to lower energies. (3) The essay says the mechanism is 'universal across glass types'. The paper shows it in several simulated metallic glasses, in SiO2 and in a single-component Lennard-Jones glass, describes the asymmetry as generic to disordered landscapes, and says the correlation-driven mechanism is predicted to dominate under typical experimental conditions; these are simulations over a limited temperature window, not a universality result. (4) The paper also applies the framework to surface diffusion, where the reduced activation energy comes mainly from weaker correlations, which the essay does not mention.