The topological hypothesis on phase transitions: the simplest case

TL;DR

This paper investigates whether static topological properties of potential energy surfaces can reliably indicate phase transitions, finding that topology changes occur even without actual phase transitions, thus challenging the hypothesis.

Contribution

It provides a new test of the topological hypothesis on a simple model, showing topology changes without corresponding strong singularities or actual phase transitions.

Findings

Topology change correlates with phase transition in the model.

Euler characteristic variation is small despite phase transition.

Topology changes occur even when the phase transition is suppressed by external fields.

Abstract

We critically analyze the possibility of finding signatures of a phase transition by looking exclusively at static quantities of statistical systems, like e.g., the topology of potential energy sub-manifolds (PES). This topological hypothesis has been successfully tested in a few statistical models but up to now there is no rigorous proof of its general validity. We make a new test of it analyzing the, probably, simplest example of a non trivial system undergoing a continuous phase transition: the completely connected version of the spherical model. Going through the topological properties of its PES it is shown that, as expected, the phase transition is correlated with a change in their topology. Nevertheless this change, as reflected in the behavior of a particular topological invariant, the Euler characteristic, is small at variance with the strong singularity observed in other systems. Furthermore, it is shown that in the presence of an external field, when the phase transition is destroyed, a similar topology change in the sub-manifolds is still observed at the maximum value of the potential energy manifold, a level which nevertheless is thermodynamically inaccessible. This suggests that static properties of the PES are not enough in order to decide whether a phase transition will take place, some input from dynamics seems necessary.