The two decades before and after 1900 saw significant progress in both variational calculus and its application to physics. On the one hand, Weierstrass and Hilbert found, for the first time, sufficient conditions for a variational principle to be minimal and provided new means to put a field often plagued by counterexamples on a secure foundation. On the other hand, almost all newly discovered physical could be formulated in terms of a variational principle or a related minimal principle. This revitalized the old belief that there was something special about these ‘optimal forms’, but not by focusing on what had proven so pernicious, to wit, the idea that nature minimized a physical quantity. While empiricists, among them Ernst Mach, considered this as an unjustified return to metaphysics on a mathematical basis, Max Planck tried to avoid this charge by a two-tiered strategy. On the one hand, he took up the mathematical developments arguing that a principle of least action was meaningful only once all the possible motions and the boundary conditions had been specified. On the other hand, he diagnosed that the principle had weathered all scientific revolutions by representing an abstract form that – for each new scientific theory – had to be specified by a Lagrangian and a new constant of nature, among them the velocity of light (for special relativity) and the quantum of action bearing his name (for the older quantum theory). In a philosophical perspective, Planck’s move represented a return to Leibniz’s contention that the principle of perfection standing behind the idea of optimal forms corresponded to a belief about the architecture of nature. Until the advent of quantum mechanics, Planck’s view was highly influential on the German physics community. That it was finally abandoned, was also a product of the fact that the mathematical entities physicists initially applied to general relativity and quantum mechanics were of a different kind – even though, in the former case, such a formulation had been available since the beginning (Hilbert action) or, in the latter would be discovered two decades later (Feynman’s path integral).