Theory seminar: Dynamical large deviations via tensor networks

Luke Causer (University of Nottingham):
Computing the statistics of dynamical observables in stochastic dynamics is a difficult task - often exponential in both system size and time.

Luke Causer (University of Nottingham)
Group seminar (hybrid format: lecture hall B0.32/zoom)
Wed, 18 May 2022, 10:30 am (MEZ)

Abstract:

Computing the statistics of dynamical observables in stochastic dynamics is a difficult task - often exponential in both system size and time. Here I give an overview of how recent applications of numerical tensor networks have allowed us to overcome this difficulty for kinetically constrained models. I will show how variational matrix product states can be used to calculate dynamical large deviations. The results of this method can then be used to construct an efficient sampling algorithm capable of generating the “rare events” associated to the dynamical large deviations on-hand. Finally, I will discuss how the approach can then be extended to finite times.

References:
[1] M. C. Bañuls, J. P. Garrahan, Phys. Rev. Lett. 123, 200601 (2019)
[2] L. Causer, M. C. Bañuls, J. P. Garrahan, Phys. Rev. E 103, 062144 (2021)
[3] L. Causer, M. C. Bañuls, J. P. Garrahan, Phys. Rev. Lett. 128, 090605 (2022)