IHCV: Discovery of Hidden Time-Dependent Control Variables in Non-Linear Dynamical Systems
Discovering non-linear dynamical models from data is at the core of science. Recent progress hinges upon sparse regression of observables using extensive libraries of candidate functions. However, it remains challenging to model hidden non-observable control variables governing switching between different dynamical regimes. Here we develop a data-efficient derivative-free method, IHCV, for the Identification of Hidden Control Variables. First, the performance and robustness of IHCV against noise are evaluated by benchmarking the IHCV method using well-known bifurcation models (saddle-node, transcritical, pitchfork, Hopf). Next, we demonstrate that IHCV discovers hidden driver variables in the Lorenz, van der Pol, Hodgkin-Huxley, and Fitzhugh-Nagumo models. Finally, IHCV generalizes to the case when only partial observational is given, as demonstrated using the toggle switch model, the genetic repressilator oscillator, and a Waddington landscape model. Our proof-of-principle illustrates that utilizing normal forms could facilitate the data-efficient and scalable discovery of hidden variables controlling transitions between different dynamical regimes and non-linear models.
PDF Abstract
