How feedback loops lead systems to tip and collapse

A feedback loop is a chain of interactions in which a change in one part of a system alters other parts, which in turn affect the original change. A tipping point is the threshold at which amplifying feedbacks take over from stabilizing ones and drive large, self-propelling change that is often difficult to reverse. Collapse can look abrupt because a system may respond only slightly to growing pressure until the threshold is crossed, and then shift quickly. [1] [2]

According to the Center for Science Education, a negative or balancing feedback helps keep a system in equilibrium. Examples it gives are the ocean's capacity to store heat and the uptake of carbon dioxide by plants and soil. A positive or reinforcing feedback pushes changes further in the same direction. In the climate, a warmer planet evaporates more water, and the extra water vapor in the air traps heat and causes more warming. Melting ice is another example: as ice retreats, darker surfaces absorb more sunlight, which causes further warming and more ice loss. The word negative here does not mean bad, and a balancing loop is often beneficial. [1] [2]

A 2026 paper in Communications Sustainability defines a tipping point as the place where small extra changes shift the balance so that positive feedbacks dominate stabilizing ones. It names three essential properties: self-reinforcing feedback, threshold behavior and persistence. For a feedback to be self-propelling, one unit of change sent around the loop must produce at least one additional unit. The authors express this as a loop gain g, where perturbations grow when g is above 1 and fade when it is below 1. Weaker positive feedbacks settle at a finite change instead. The paper compares this to the squeal of a microphone and speaker, which keeps growing once the looped signal passes a critical gain. [2]

The authors also note that real thresholds are usually surfaces in a space of several variables, not single points. The stability of the Amazon rainforest, for example, depends on at least four parameters: temperature, precipitation, dry season intensity and deforestation. This is one reason exact thresholds are hard to identify in real data. [2]

Before a threshold, a system may show only a small, roughly linear response to changing conditions. Once positive feedback becomes self-propelling, the same system can change sharply. The Communications Sustainability paper describes two main routes. In bifurcation-induced tipping, gradual change removes a stable state. In the Stommel model of ocean circulation, rising freshwater input eventually eliminates the circulation's "on" state, and it collapses regardless of how fast the change came. In rate-induced tipping, forcing changes faster than the system can adjust, so a system can fail even though a stable state still exists. If the system is slow compared with the forcing, it may briefly overshoot a threshold without a lasting shift. [2]

Abrupt is relative. The paper notes that ice sheet collapse over thousands of years is abrupt compared with ice cap formation but slow compared with a human lifetime. The Conversation describes ecosystems as absorbing stress until they can no longer do so and then passing a point of no return, with a fishery example in which catching juvenile fish during shortages further undermines the stock. [2] [3]

The Center for Science Education describes crossing a tipping point as a transition to a new regime that cannot easily be reversed, using ice sheet loss and sea level rise as its example. The Global Tipping Points report describes tipping as change becoming self-sustaining after a forcing threshold is passed, producing a qualitative state change such as an ecological regime shift. The search results use slightly different wording, with some stressing irreversibility and others calling the change only often difficult to reverse. The Communications Sustainability authors also say mathematicians recognize many kinds of bifurcation, and that tipping points are a specific class defined by the combination of the three properties above. [1] [4] [2]