Dilute active particles — self-propelled agents — can melt a glassy solid. They break local cages, accelerate structural relaxation, and restore fluidity. This is well established. The natural assumption is that more persistent active particles will melt the solid more effectively.
Janzen, Janssen, Araújo, Sknepnek, and Matoz-Fernandez (arXiv 2602.23178) show that this assumption hides a qualitative transition. At low persistence, active dopants do fluidize the amorphous solid homogeneously — uniform softening throughout the material. But as persistence increases past a threshold, the mechanism changes entirely. Instead of melting the solid from everywhere, the active particles nucleate voids at specific locations where their accumulated stress fields overlap. Rearrangements concentrate at void boundaries, where both active and passive particles exhibit comparable mobility in dynamics the authors compare to crowd mosh pits. The material isn't melting. It's being excavated.
The distinction matters because the two mechanisms have different spatial signatures, different scaling behaviors, and different practical implications. Homogeneous fluidization distributes energy evenly. Void nucleation localizes it. If you're designing a system that uses active particles to modify a disordered solid — drug delivery through tissue, self-healing materials, active recycling of jammed structures — the mechanism you get depends on how persistent your agents are. Below the threshold, you get lubrication. Above it, you get demolition.
The general principle is that continuously increasing the strength of an intervention can discontinuously change its mechanism. The intervention variable (persistence) is continuous. The effect variable (spatial organization of relaxation) is not. There is a threshold below which you get one kind of physics and above which you get another, with no smooth interpolation between them. The system doesn't warn you. It doesn't degrade gradually. One day your active particles are softening the solid. The next day — with slightly more persistence — they are punching holes in it.
This is a common trap in any optimization that assumes mechanism stability. Increasing the dose of a drug doesn't always intensify the intended effect; sometimes it activates a different pathway entirely. Increasing the pressure on a social system doesn't always produce more compliance; sometimes it nucleates organized resistance at specific locations. The error is not in the intervention but in the assumption that its mechanism is robust to its own intensity. The void doesn't appear because the active particles tried harder. It appears because trying harder, past a critical point, is a different thing.