 |
Universität Augsburg
|
|
|
Universität Augsburg
|
|
Brune Massoulié
Paris-Dauphine
spricht am
Freitag, 8. Mai 2026
um
16:30 Uhr
im
Raum 2004 (L1)
über das Thema:
| Abstract: |
| Under suitable assumptions, a Markov chain will converge at large times to its stationary measure, over a time scale known as the mixing time. Markov chains can then be used to approximately sample from a complex distribution, which is the principle underlying Markov Chain Monte Carlo techniques. In this context, there has been strong interest both in the physics and the mathematics literature in lifting Markov chains, which is a way to modify an initial Markov chain so that that the lifted Markov chain converges faster to its stationary measure. We study a particle system which is both a lifting of the TASEP (Totally Asymmetric Simple Exclusion Process) and a periodic instance of self-repelling random walks. Using a graphical representation, we identify a coupling time with some delay, which allows us to study two notions of mixing times of the model. For the standard notion of mixing time, we show that the model does not converge abruptly to stationarity, whereas for a mixing time associated with an average of the distribution over the last few steps, convergence to stationarity is abrupt, which is known as the cutoff phenomenon. |
| Hierzu ergeht herzliche Einladung. |
| Prof. Dr. Dominik Schmid |