Filtrations Indexed by Attracting Levels and their Applications

In this paper, we introduce a framework for constructing two types of filtrations with respect to a reference object akin to an attractor. The forward directions of the filtrations characterize the level of control required to approach the attractor. The negative directions of the filtrations capture the robustness of the ``basin'' by quantifying how much perturbations can be tolerated without altering the tendency to be attracted to a given attractor. Furthermore, to accommodate real-world phenomena, we consider dynamical systems not on metric spaces with discrete-time maps, but on sets endowed with a more general cost function, where the dynamics are governed by partial maps. To illustrate the generality of the proposed construction, we apply it to ensemble forecasts of a tropical cyclone. The filtrations capture regions of heightened sensitivity, where small perturbations induce qualitative changes in asymptotic behavior. This example substantiates the theoretical framework in a data-driven but structurally rich context, emphasizing its relevance for analyzing control and robustness in intricate dynamical systems.