The work sits off on its own: its closest neighbours in the archive study cell division, heat transport in graphene, snow avalanches, threads in flow and friction in a fault. All of them touch in one way or another on collective rearrangements in disordered media, but this paper stands out because it systematically shows for the first time that energy can accumulate during an avalanche in a granular fault. Until now it was assumed that an avalanche always releases stress and compacts the material, yet here events are found that widen the layer and at the same time increase the elastic energy stored at the boundary.
The experimental setup was a single layer of roughly 4000 photoelastic disks of two diameters, confined in an annular gap 5 mm wide between two stationary cylindrical walls. The bottom plate rotated at a constant speed of 48 mm/h, while the top plate could move freely in the vertical direction under a load of about 20 kg. The driving force, the shear force on the top plate and its vertical displacement were recorded continuously at a frequency of 10 kHz. Six piezoelectric sensors recorded acoustic emission at 100 kHz, and 24 synchronised cameras took images of the structure every 4 seconds. Each experiment lasted about 100 hours and captured hundreds of thousands of avalanches.
Over the entire measurement time, about 400,000 acoustic bursts, 100,000 mechanical avalanches and 20,000 discrete slip events were identified. The avalanches were classified by the signs of the change in mechanical energy at the resisting boundary and of the vertical displacement: Q1 (energy release and compaction), Q2 (energy release and expansion), Q3 (energy accumulation and compaction), Q4 (energy accumulation and expansion). Pure energy releases turned out to be the most frequent, but the largest events occur in Q1 and Q4. The Q4 events, dubbed "negative avalanches", widen the layer and increase the energy transferred to the stationary boundary, and their acoustic signature is systematically stronger than that of energy-releasing events. Structural analysis shows that in Q1 and Q2 the motion of the grains is concentrated in the shear band with vortex structures, whereas in Q3 and Q4 the displacements are almost half as large but spread through the jammed region, blurring the edge of the shear band.
Unlike neighbouring studies, which treat avalanches as purely dissipative events, this research demonstrates that an avalanche can be a mechanism for redistributing stress into a more loaded state. This changes the picture of the role avalanches play in fault evolution: instead of simply releasing energy, they can prepare the system for subsequent instability. The four regimes obtained link dilatancy, force-chain rearrangement and elastic energy accumulation, which provides a new basis for interpreting seismic and laboratory data.
The topic could develop towards identifying structural motifs of the contact network that favour rupture or, conversely, stabilise the network. Graph neural networks and stability analysis could predict the type of avalanche from the instantaneous configuration of contacts. In addition, it is important to test how event rates and scaling relations depend on the confinement geometry, the loading protocol and detection thresholds before transferring the mechanism to natural faults.
The results matter for the physics of granular media, tribology and seismology, because they offer a new mechanism for stress accumulation in shear zones.
The next step will be to test the reproducibility of the observed regimes in independent experiments with varying geometry and loading rate, and to develop machine-learning algorithms for predicting the type of avalanche from the structure of the contact network.
The work is directly connected to the unsolved problem of the predictability of scale-invariant avalanches in granular media. The discovery that avalanches can accumulate energy and leave a fault in a more loaded state indicates that prediction should rely not so much on precursors as on the state of the system after an event. This opens a path to new approaches for predicting the failure of disordered materials.
The largest "negative avalanches" widen the layer and at the same time increase the elastic energy transferred to the stationary boundary, and their acoustic signature is systematically stronger than that of ordinary energy-releasing events.