Multiple transient memories are a remarkable and counterintuitive example of how information can be stored in driven disordered systems. A system with transient memories will ‘learn’ multiple driving amplitudes that can subsequently be read out. However, most of the memories will be forgotten after many driving cycles. Surprisingly, if noise is added, all of the memories are retained for much longer. This memory behavior has been observed in traveling charge-density waves and was predicted in simulations of sheared suspensions [Keim & Nagel, Phys. Rev. Lett. 107, 2011].
We have shown behavior in experiment that is consistent with multiple transient memories that become stabilized by noise. We apply low Reynolds-number cyclic shear to a neutrally buoyant, non-Brownian suspension. Starting from a random configuration, the particle trajectories are irreversible at first but (as has been shown before [Corte et al., Nat. Phys. 4, 2008]) eventually find a configuration where they retrace their paths exactly during each cycle. This comprises a memory of the driving, which is read out by measuring the degree of particle reversibility versus strain amplitude. We can also encode multiple memories, wherein smaller memories are forgotten when larger shear is applied. Finally, when noise is added, all the memories are retained for much longer.