Reproduce “Transitions in echo index and dependence on input repetitions” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Peter Ashwin, Andrea Ceni

Input repetition is shown to control whether an input-driven dynamical system retains multiple asymptotic responses or converges to a unique one. For a class of switching systems and a two-dimensional leaky-integrator echo state network, the analysis gives sufficient conditions for lower-bounding the echo index and for reducing it to one, while simulations reveal a transition near 30 repetitions of a single-attractor input regime.

Key insights

  • Long minimum repetitions allow trajectories to approach the attracting regions of individual maps, helping preserve distinct pullback-attracting responses; sufficiently long exposure to a single globally attracting map can instead eliminate multistability.
  • Run-length constraints are informative but do not fully determine the echo index: switching compositions and basin geometry can introduce additional attractors or alter the transition boundary.
  • Finite-time simulations may mistake slow transients for separate asymptotic responses, so longer observation windows and the frequency of informative input patterns are important when estimating the echo index.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.