Collapses of dynamical systems into irrecoverable states are observed in ecosystems, human societies, financial systems and network infrastructures. Despite their widespread occurrence and impact, these events remain largely unpredictable. In searching for the causes for collapse and instability, theoretical investigations have been so far unable to determine quantitatively the influence of the structural features of the network formed by the interacting species. Here, we derive the condition for the stability of a mutualistic ecosystem as a constraint on the strength of the dynamical interactions between species and a topological invariant of the network: the k-core. Our solution predicts that when species located at the maximum k-core of the network go extinct as a consequence of sufficiently weak interaction strengths the system will reach the tipping point of its collapse. As a key variable involved in collapse phenomena, monitoring the k-core of the network may prove a powerful method to anticipate catastrophic events in the vast context that stretches from ecological and biological networks to finance.
I. INTRODUCTIONA complex dynamical system collapses when a small perturbation in the parameters characterizing the species interactions causes a large-scale extinction of the species in the system [1][2][3][4][5][6][7][8][9][10][11][12]. The tipping point at which the system suddenly shifts to the irrecoverable state is, for practical purposes, the most important quantity one wishes to know [5,6,13]. It is a function of the dynamical and structural parameters of the system determined by the fixed point solution of the nonlinear equations describing the system's dynamics [1]. However, the tipping point is hard to determine, due to the difficulties encountered in solving the nonlinear dynamical equations to quantify the dependence of the fixed point solution on the system parameters and, in particular, on the features of the underlying network of interacting species in the system [1,3,6,8]. Indeed, no exact analytical result exists, so far, that relates the network properties to the fixed points of the dynamical system. Here, we first study numerically the fixed point equations of a dynamical system of mu-