The Mechanism of Scale-Invariance
A system is invariant with respect to an input transformation if we can transform any dynamic input by this function and obtain the same output dynamics after adjusting the initial conditions appropriately. Often, the set of all such input transformations forms a Lie group, the most prominent examples being scale-invariant ($u\mapsto e^pu$, $p\in\mathbb{R}$) and translational-invariant ($u\mapsto pu$) systems, the latter comprising linear systems with transfer function zeros at the origin. Here, we derive a necessary and sufficient normal form for invariant systems and, by analyzing this normal form, provide a complete characterization of the mechanism by which invariance can be achieved. In this normal form, all invariant systems (i) estimate the applied input transformation by means of an integral feedback, and (ii) then apply the inverse of this estimate to the input before processing it in any other way. We demonstrate our results based on three examples: a scale-invariant "feed-forward loop", a bistable switch, and a system resembling the core of the mammalian circadian network.