Infinitesimal phase response functions can be misleading

Phase response functions are the central tool in the mathematical analysis of pulse-coupled oscillators. When an oscillator receives a brief input pulse, the phase response function specifies how its phase shifts as a function of the phase at which the input is received. When the pulse is weak, it is customary to linearize around zero pulse strength. The result is called the infinitesimal phase response function. These ideas have been used extensively in theoretical biology, and also in some areas of engineering. I give examples showing that the infinitesimal phase response function may predict that two oscillators, as they exchange pulses back and fourth, will converge to synchrony, yet this is false when the exact phase response function is used, for all positive interaction strengths. For short, the analogue of the Hartman-Grobman theorem that one might expect to hold at first sight is invalid. I give a condition under which the prediction derived using the infinitesimal phase response function does hold for the exact phase response function when interactions are sufficiently weak but of positive strength. However, I argue that this condition may often fail to hold.