On the continuity of phase transition of three-dimensional square-lattice XY models

\Pi is the projective map that maps a paired graph to the corresponding unpaired one.

Let P=P(x\rightarrow y) be a fixed set of paired paths x\rightarrow y, y\rightarrow x, and paired loops x\rightarrow x and y\rightarrow y, except that, at the sites x and y, the edges are not paired. We say P is a switchable paired x\rightarrow y-graph if these paths and loops avoid the \delta-site.

For a fixed large integer N so that |x|+|y|<N, a partially paired graph G_{N} in the region |z|\leq N satisfies the following properties:

• (1) Its underlying unpaired graph, \Pi(G_N), is \Big(\mathcal{G}_{\delta}(x\rightarrow y), \mathcal{G}_0\Big)\Big|_{|z|\leq N} or \Big(\mathcal{G}_{\delta}, \mathcal{G}_0(y\rightarrow x)\Big)\Big|_{|z|\leq N}. Recall unpaired graphs \Big(\mathcal{G}_{\delta}(x\rightarrow y), \mathcal{G}_0\Big) or \Big(\mathcal{G}_{\delta}, \mathcal{G}_0(y\rightarrow x)\Big) from (?).

• (2) In the interior, at site z with |z|<N and z\not\in \{x,y\}, every incoming edge is paired with an outgoing one;

• (3) At the sites x and y, the edges are not paired;

• (4) On the boundary, at site z with |z|=N, the edges are not paired.

We denote this set of graphs by PP_{N}, i.e.

For any

we use them to represent two sets of paired infinite graphs, specifically,

and

For a fixed large integer N, a paired infinite graph G(x\rightarrow y) is called a finitely-paired graph in the region |z|\leq N if every edge in the region |z|\leq N is paired into a finite loop or a finite path.

Suppose that G_{\delta,\mathcal{P}}=G_{\delta,\mathcal{P}}(x\rightarrow y) is a paired infinite graph. And \mathcal{P}=\mathcal{P}(x\rightarrow y) is an unpaired subgraph in \Pi(G_{\delta,\mathcal{P}}).

We perform a surgical switching, using \mathcal{P}, on G_{\delta,\mathcal{P}} to obtain G_{0,\overline{\mathcal{P}}}=G_{0,\overline{\mathcal{P}}}(y\rightarrow x) as follows.

For G_{0,\overline{\mathcal{P}}}, its corresponding unpaired graph \Pi(G_{0,\overline{\mathcal{P}}}), is obtained from \Pi(G_{\delta}) by switching \mathcal{P} with the procedure described in (?).

Next, we consider the pairing in the new graph G_{0,\overline{\mathcal{P}}}.

In brief, some pairings remain unchanged, and some pairings must be modified depending on whether they are paired with edges in \mathcal{P}.

• (I) We start with an easy case. Specifically, suppose that \mathcal{P} is a x\rightarrow y path that does not contain any loop and is of the form

  \begin{aligned} \begin{split} x&=a_{0}\rightarrow a_{1}\rightarrow a_{2}\rightarrow \cdots \rightarrow a_{n}=y,\\ &\text{and}\ a_{k}\not=a_{l}\ \text{if}\ k\not=l. \end{split} \end{aligned}

  Fix a site v\in \mathbb{Z}^3, the pairing of an edge e_{v\rightarrow w} (or e_{w\rightarrow v}) at the site v will stay the same in the following two possibilities:

  • the edge is not paired with any edge of \mathcal{P},

  • the edge e_{v\rightarrow w} (or e_{w\rightarrow v}) is part of \mathcal{P} and is already paired to another edge in \mathcal{P} at v.

  Otherwise, v=a_k for some k=1,2,\cdots,n-1 there are two possible pairings of e_{a_k\rightarrow w} (or e_{w\rightarrow a_k}) at the site a_k, and we handle them differently. In what follows we only consider e_{a_k\rightarrow w} at the site a_k. That for e_{w\rightarrow a_k} is almost identical, and hence is omitted.

  • The edge e_{a_{k}\rightarrow w} is part of \mathcal{P}, thus w=a_{k+1} by (?). At a_k it is paired with an edge e_{b_{k-1}\rightarrow a_k}\not\in \mathcal{P}. And the edge e_{a_{k-1}\rightarrow a_k}\in \mathcal{P} is paired with e_{a_k\rightarrow c_{k+1}}\not\in \mathcal{P}.

    Then in the new graph G_{0,\overline{\mathcal{P}}} where we reverse the path \mathcal{P}, at a_k, e_{a_{k}\leftarrow a_{k+1}} is paired with e_{a_k\rightarrow c_{k+1}}; and e_{a_{k-1}\leftarrow a_k} is paired with e_{a_k\leftarrow b_{k-1}}.

  • The edge e_{a_{k}\rightarrow w} is not part of \mathcal{P}, and at a_k it is paired with e_{a_{k-1}\rightarrow a_k}\in \mathcal{P}, and the edge e_{a_k\rightarrow a_{k+1}}\in \mathcal{P} is paired with an edge e_{c\rightarrow a_k}.

    Then in the new graph G_{0,\overline{\mathcal{P}}} where we reverse the path \mathcal{P}, at a_k, e_{a_{k}\rightarrow w} is paired with e_{a_k\leftarrow a_{k+1}}, and e_{c\rightarrow a_k} is paired with e_{a_{k-1}\leftarrow a_k}.

  To illustrate the procedure, we use the Figures ? and ?. Both are paired graphs. In Figure ?, the paired path and loop are x\rightarrow a1\rightarrow a2\rightarrow y and b1\rightarrow a1\rightarrow b2\rightarrow b3\rightarrow a2\rightarrow b4\rightarrow b1. And the chosen (unpaired) path for surgical switching, \mathcal{P}, is

  \mathcal{P}=x\rightarrow a1\rightarrow b2\rightarrow b3\rightarrow a2\rightarrow y.

  Figure ? is obtained after the surgical switching, the paired path and loop are y\rightarrow a2\rightarrow b4\rightarrow b1\rightarrow a1\rightarrow x and a1\rightarrow a2\rightarrow b3\rightarrow b2\rightarrow a1.

• (II) For the general case, \mathcal{P} can be decomposed, disjointly, into

  \begin{aligned} \mathcal{P}=\mathcal{P}_1\cup \mathcal{P}_2\cup \cdots \mathcal{P}_{m} \end{aligned}

  and each of \mathcal{P}_k is a self-avoiding path or loop, then we make the surgical switching for each one of them similarly.