Memory in three-dimensional cyclically driven granular material

Dense granular packings are often subjected to cyclic forcing, and it is critical to understand whether such forcing will lead to irreversible rearrangements of particles, which can result in creep, yielding, or segregation. It is known that the response to forcing depends on the prior history of the packing, i.e. the system has a memory of prior rearrangements. The role of rotations is of particular interest, since most energy dissipation in driven dense granular packings occurs through frictional sliding during particle rotations. We measure a three-dimensional granular system's reversibility and memory under cyclic compression through a combination of experiments directly matched with simulations. Experimentally, we image the grains using a refractive-index-matched fluid, then analyze the images using the artificial intelligence of variational autoencoders. These techniques allow us to track all the grains' translations and three-dimensional rotations with accuracy sufficient to infer contact-point sliding and rolling. We match these experimental observations to simulations, which allow us to dissect the role of translations and rotations in granular memory. Our observations reveal unique roles played by three-dimensional rotations in granular flow, memory, and energy dissipation.