Does the Kosterlitz–Thouless transition connect phases with different topological quantum numbers?

Question

The Kosterlitz-Thouless transition is often described as a "topological phase transition." I understand why it isn't a conventional Landau-symmetry-breaking phase transition: there is no local symmetry-breaking order parameter on either side of the transition, and all thermodynamic quantities remain continuous (though not analytic) at all derivative orders across the transition. I also understand why it is "topological" in the sense that it involves the binding of vortex topological defects.

But when I think of a "topological phase transition," I think of a phase transition between two phases characterized by different topological quantum numbers, which are topological invariants over the entire phase. For example, the transition between a trivial and topological band insulator corresponds to a change in the Chern number of the band structure (between zero and nonzero). The transition between a topologically trivial and topologically ordered phase corresponds to a change in the topologically protected ground-state degeneracy (between 1 and greater than 1).

Is the KT transition, in which neither side of the transition has either long-range entanglement or symmetry-protected topological order, a topological phase transition in this sense? If so, what are the quantum numbers on both sides of the transition? (I'm not counting trivial answers like "the topological quantum number defined as 0 if vortices are bound and 1 if they are unbound.")

Answer 1

We recently posted a paper (https://arxiv.org/abs/1808.09394) to address this issue systematically.

We may use disordered symmetry breaking states (which are described by non-linear $\sigma$-models) to realize a large class of topological orders. ... ... In this paper, we show that the phase transitions driven by fluctuations with all possible topological defects produce disordered states that have no topological order, and correspond to non-topological phase transitions. While transitions driven by fluctuations without any topological defects usually produce disordered states that have non-trivial topological orders, and correspond to topological phase transitions. Thus, it may be confusing to refer the transition driven by topological defects as a topological phase transitions, since the appearance of topological defects decrease the chance to produce topological phases of matter.

More precisely, if the fluctuating order parameter in a disordered state has no topological defects, then the corresponding disordered state will usually have a non-trivial topological order. The type of the topological order depends on the topology of the degenerate manifold $K$ of the order parameter (ie the target space of the non-linear $\sigma$-model). For example, if $\pi_1(K)$ is a finite group and $\pi_{n>1}(K)=0$, then the disordered phase may have a topological order described by a gauge theory of gauge group $G=\pi_1(K)$. If $\pi_1(K),\pi_2(K)$ are finite groups and $\pi_{n>2}(K)=0$, then the disordered phase may have a topological order described by a 2-gauge theory of 2-gauge-group $B(\pi_1(K),\pi_2(K))$.

It is the absence of topological defects that enable the symmetric disordered state to have a non-trivial topological order. When there are a lot of topological defects, they will destroy the topology of the degenerate manifold of the order parameter (ie the degenerate manifold effectively becomes a discrete set with trivial topology). In this case the symmetric disordered state becomes a product state with no topological order. Certainly, if the fluctuating order parameter contains only a subclass of topological defects, then only part of the topological structure of the degenerate manifold is destroyed by the defects. The corresponding symmetric disordered state may still have a topological order.