Superfluid shear layer turbulence realised through machine learning optimisation

In recent years, machine learning has emerged as a powerful technique for optimising BEC experiments. Our experimental system consists of a 2D BEC confined using a potential derived from a digital micromirror device, permitting high resolution dynamic control of the condensate [1]. Using a ring-shaped configuration, we optimise the generation of counter-rotating persistent currents through machine-learner control of a stirring barrier. We find that the learner optimises stirring under several different constraints, including maximising winding number or minimising stirring time.

After preparing the optimised persistent current state, we study the dynamics of the system after in-trap interference. The interaction between the rotating and stationary superfluids realises a shear layer that rapidly decays, resulting in a ring of quantised vortices. This vortex ring is unstable and decays into vortex clusters, indicative of the superfluid Kelvin-Helmholtz instability [2]. By studying the cluster sizes with increasing hold time, we find analogous behaviour to decaying classical 2D turbulence.

[1] G. Gauthier, I. Lenton, N. McKay Parry, M. Baker, M. J. Davis, H. Rubinsztein-Dunlop, and T. W. Neely, Optica 3, 1136–1143 (2016).

[2] A. W. Baggaley and N. G. Parker. Kelvin-Helmholtz instability in a single-component atomic superfluid. Phys. Rev. A 97 053608 (2018).