When glass shatters, the fragment sizes follow a distribution. That distribution depends on the impact energy, the residual stress in the glass, and the stress gradients — all the specific details of the particular breakage event. Different impacts produce different fragment size patterns.
Mohajerani and colleagues (arXiv:2602.20443) show that when fragment sizes from diverse impact conditions are normalized by their mean fragment area, the cumulative size distributions collapse onto a single master curve. One universal function describes all fragmentation events, regardless of the specifics. The details set the scale (the mean fragment size) but not the shape (the distribution around that mean).
The collapse is exponential: the probability of a fragment being larger than a given size decays exponentially above the mean. This holds across varying impact energies and stress conditions. The fracture topology changes — coarse networks at low energy, fine networks at high energy — but the normalized distribution is invariant.
At the microscopic level, the mechanism involves non-sequential bond breaking ahead of crack tips. Bonds break out of order — not in a smooth advancing front, but in a scattered pattern that produces local crack speeds exceeding the Rayleigh wave speed. This microscopic chaos produces the macroscopic universality: the specific pattern of bond breaking varies wildly between events, but the statistical properties of fragmentation are determined by the mean alone.
The result is another instance of the fragmentation universality observed across materials science, geology, and explosive dynamics: the microscopic details determine the mean fragment size, but the shape of the distribution is universal.
The general observation: when a complex process (fracture) has many microscopic realizations but one macroscopic parameter (mean size) that sets the scale, normalization by that parameter reveals a universal function. The microscopic chaos washes out in the aggregate. The only thing that survives from the specific event into the distribution is the mean — everything else is generic.