ABSTRACT.
The averaging process is a Markovian mass redistribution model over the vertex set of a connected graph, in which masses sitting on nearest neighbor vertices get deterministically averaged out at random Poissonian times. Despite the latter randomness, this degenerate update rule drives, in the long run, the mass profile towards a deterministic flat configuration. This equilibrium state is also referred to as consensus when interpreting the averaging process as a basic model of opinion dynamics/gossip algorithm, in which vertices represent agents and masses their opinions. The mathematical interest in the model was recently revived by Aldous and Lanoue (2012), later followed by Chatterjee, Diaconis, Sly and Zhang (2022), Quattropani and Sau (2023), and Movassagh, Szegedy and Wang (2022+), all these works being concerned with determining sharp convergence rates to equilibrium for the process. In this talk, we review some basic properties of the averaging process, and present some recent results on mixing times and cutoff for this model. If time permits, we also discuss some recent developments on scaling limits of the process, particularly interesting in this degenerate setting due to the lack of a natural notion of local equilibrium. Based on joint works with Matteo Quattropani (Rome, Sapienza) and Pietro Caputo (RomaTre).