When Roni Chatterjee slipped on a new pair of shoes last monsoon season, he noticed how they resisted at first—stiff, unyielding—until, after weeks of Hyderabad’s erratic monsoon walks, sudden turns on wet tiles, and uneven pavements, they softened into perfect comfort. That everyday experience sparked a discovery: what if materials like glass could learn not just from perfect repetition, but from chaos? At the Tata Institute of Fundamental Research in Hyderabad, Chatterjee and Smarajit Karmakar, alongside Damien Vandembroucq from CNRS in Paris and Muhittin Mungan from Heinrich Heine University in Düsseldorf, have shown that amorphous materials—like glass, gels, and plastics—can store precise mechanical memories even when deformed randomly, overturning long-held assumptions in materials science. Their findings, published in the New Journal of Physics, reveal a deeper intelligence in disorder.
For decades, scientists believed that for a material to “remember” a mechanical input—like the shape of a foot or the rhythm of a shake—it needed to be trained with perfectly repeated cycles of deformation. Think of bending a paperclip back and forth in exactly the same way, over and over. But real life isn’t so orderly. Walking, driving, or even seismic shifts involve irregular, unpredictable forces. The team questioned whether memory formation required such precision—or if randomness could suffice. Using computer simulations of a model glassy material, they subjected it to 100 cycles of random deformation, all capped at a specific amplitude—say, 5% strain—but with no repeating pattern. It was like shaking a box of marbles unpredictably, yet never too hard.
After this “training,” they tested the material’s memory with a single, clean deformation cycle at varying amplitudes: 1%, 3%, 5%, up to 7%. The critical moment came at 5%. Only when the readout matched the training amplitude did the material return exactly to its prior state—its internal particles snapping back into place with near-zero mean squared displacement. This proved the material had encoded a precise memory of the training intensity, not the pattern. Crucially, this memory only formed when the training stayed below the material’s yielding point—the threshold beyond which permanent damage occurs. Push too hard, and the memory dissolves, like a notebook burned.
The implications ripple across engineering and biology. From shock-absorbing materials that adapt to unpredictable impacts, to understanding how biological tissues respond to irregular stresses, this work suggests nature may rely on randomness to learn. As Chatterjee puts it, “We’ve been teaching materials the wrong way—like training a dancer with a metronome when they learn best to freestyle.” Now, the door opens to designing smart materials that learn from life’s messiness, not just lab-perfect rhythms.