Extinctions as a vestige of instability: the geometry of stability and feasibility

Species coexistence is a complex, multifaceted problem. At an equilibrium, coexistence requires two conditions: stability under small perturbations; and feasibility, meaning all species abundances are positive. Which of these two conditions is more restrictive has been debated for many years, with many works focusing on statistical arguments for systems with many species. Within the framework of the Lotka-Volterra equations, we examine the geometry of the region of coexistence in the space of interaction strengths, for symmetric competitive interactions and any finite number of species. We consider what happens when starting at a point within the coexistence region, and changing the interaction strengths continuously until one of the two conditions breaks. We find that coexistence generically breaks through the loss of feasibility, as the abundance of one species reaches zero. An exception to this rule - where stability breaks before feasibility - happens only at isolated points, or more generally on a lower dimensional subset of the boundary. The reason behind this is that as a stability boundary is approached, some of the abundances generally diverge towards minus infinity, and so go extinct at some earlier point, breaking the feasibility condition first. These results define a new sense in which feasibility is a more restrictive condition than stability, and show that these two requirements are closely interrelated. We then show how our results affect the changes in the set of coexisting species when interaction strengths are changed: a system of coexisting species loses a species by its abundance continuously going to zero, and this new fixed point is unique. As parameters are further changed, multiple alternative equilibria may be found. Finally, we discuss the extent to which our results apply to asymmetric interactions.