We use the notation \operatorname{Ad}_S and \Omega_{S} from Definition ?, ?.
We start by assuming that S\subset{G'} is abelian and that X\subset {\mathcal{E}'} is a non-empty open set with X\cap \mathcal{E}_{S}=\emptyset and such that
N < \left|
\operatorname{Ad}_{S}(\pi(e))
\right|
=
\left|
\operatorname{Ad}_{S}(e)
\right|
< \infty
\quad\text{for all }e\in X
for some N\in\{0,1,\ldots\}. For any z\in \mathbf{T} and e\in{\mathcal{E}'}, we have
\operatorname{Ad}_{S}(z\cdot e) = \{\, z\cdot e' : e'\in \operatorname{Ad}_{S}(e)\,\},
and thus in particular \Omega_{S}(z\cdot e)=\Omega_{S}(e). We can thus replace X with \pi^{-1}(\pi(X)) without changing its property
at (Proof). Note that (Proof) implies that U\coloneq r(X)=s(X) is contained in S^{(0)}, since \Omega_{S}(e)=0 for e\in {\mathcal{E}'} with s(e)\notin S^{(0)}.
Since X\cap \mathcal{E}_{S}=\emptyset by assumption, we have that V_0\coloneq\pi(X) is contained in G\setminus S.
By shrinking X, we can assume that V_0 is a (necessarily open) bisection in G. Since V_0\subset G\setminus S and V_0=\pi(X)\subset \pi(\operatorname{int}_{\mathcal{E}}\left({\mathcal{E}'}\right)), it follows from Lemma 2.11? that V_0\subset \operatorname{int}_{G}\left({G'}\right)\setminus S. Let g_0(\cdot)\colon U\to V_0 be the homeomorphism that sends u\in U\subset S^{(0)} to the unique element g_0(u) in V_0\cap G(u)\subset \pi(X).
Claim. The map u\mapsto |\operatorname{Ad}_{S}(g_0(u))| is lower semi-continuous from U to \mathbf{Z}_{\geq 0}.
Proof (Proof of Claim ?). Fix any \overline{u}\in U
\subset S^{(0)} and let g_0\coloneq g_0(\overline{u}). It will suffice to find an open neighborhood \tilde{U}\subset U of \overline u such that
| \operatorname{Ad}_{S}(g_0(u))|
\geq
|
\operatorname{Ad}_{S}(g_0(\overline{u}))|
for all u\in \tilde{U}.
If |\operatorname{Ad}_{S}(g_0)|=N+1, then since g_0(u)\in \pi(X), Inequality (?) follows automatically from the property at (Proof) of X, and we are done; therefore, we may assume that n\coloneq |\operatorname{Ad}_{S}(g_0)|-1
> N\geq 0,
so that
\operatorname{Ad}_{S}(g_0)
=
\{\, g_0,g_1,\ldots,g_{n}\,\}
\subset
G(\overline{u})
for pairwise distinct elements g_{i}. Since G is Hausdorff, there is an open neighborhood \tilde{U} of \overline u such that there exist open bisections V_{1},\ldots, V_{n} for which g_{i}\in V_{i} and \tilde{U}=s(V_{i}) for all i\in\{\,1,\ldots, n\,\}. We let g_i(\cdot) denote the homeomorphisms from \tilde{U} to V_i sending u to the unique element g_i(u)\in V_i\cap G_u; in particular, g_{i}(\overline{u})=g_{i}\in G(\overline{u}).
The paper has a figure here, which this page does not show.
For any two distinct i,j\in\{\,0,\ldots,n\,\}, we have g_{i}(\overline{u})=g_{i}\neq g_{j}=g_{j}(\overline{u}), so by continuity we can shrink \tilde{U} and assume that g_{i}(u)\neq g_{j}(u) for all u\in \tilde{U}, meaning that the bisections V_0\cap r^{-1} (\tilde{U}),V_{1},\ldots, V_{n} are pairwise disjoint; see Figure ?.
For each u\in \tilde{U}, we have s(g_0(u))=u=r(g_0(u)), so for each i\in\{\,0,\ldots, n\,\}, we may consider the following pairwise disjoint subsets of S(u):
\begin{aligned}
S_{i}(u)
=
\{\,
t\in S(u): t^{-1} g_0(u)t=g_{i}(u)
\,\},
\end{aligned}
so that
\begin{aligned}
\{\, t^{-1} g_{0}(u) t : t\in S_{i}(u)\,\}
=
\begin{cases}
\emptyset,
&
\text{if }S_{i}(u)=\emptyset,
\\
\{\,g_{i}(u)\,\}
&
\text{otherwise.}
\end{cases}
\end{aligned}
We claim that we can shrink \tilde{U} in such a way that each S_{i}(u) is non-empty for u\in \tilde{U}.
For each i\in\{\,1,\ldots,n\,\}, we know by assumption on g_0=g_0(\overline{u}) that S_{i}(\overline{u}) is non-empty, so pick s_{i}\in S_{i}(\overline{u}). By choosing small enough open bisections W_{i}\subset S
\subset {G'} around s_{i} and by shrinking the neighborhood \tilde{U} of \overline{u}, we get homeomorphisms s_i (\cdot) \colon
\tilde{U}
\to W_i sending u to the unique element s_i(u)\in W_i\cap G(u)\subset S(u); in particular, s_{i}(\overline{u})=s_{i}. Since u\in S_0(u) for all u\in \tilde{U}\subset S, we may further let W_0\coloneq \tilde{U} and let s_0(\cdot)\colon \tilde{U}\to W_0 be defined by s_0(u)=u.
Since g_{i}(u)\in G_u and s_{i}(u)\in S(u) for each u\in \tilde{U}, we may consider the continuous map
\kappa_i\colon \tilde{U}\to
G
\quad\text{ given by }\quad
u \mapsto
s_{i}(u)^{-1} g_0(u)s_{i}(u) g_{i}(u)^{-1}.
By construction of s_{i}(\cdot), we have that s_{i}(\overline{u})=s_{i} is an element of S_{i}(\overline{u}), meaning that
\kappa_{i}(\overline{u})=s_{i}^{-1} g_0s_{i}g_{i}^{-1}=\overline{u}\inG^{(0)},
so by étale ness, \kappa_{i}^{-1}(G^{(0)}) is an open neighborhood of \overline{u}. By replacing \tilde{U} with its open subset \kappa_{i}^{-1}(G^{(0)}), we can without loss of generality assume that \kappa_{i}(\tilde{U})\subset G^{(0)} for all i\in\{\,0,\ldots, n\,\}. In other words,
s_{i}(u)^{-1} g_{0}(u) s_{i}(u)
=
g_{i}(u)
\quad\text{for all } u\in \tilde{U},
as depicted in Figure ?. In particular, S_{i}(u) is non-empty, as claimed.
The paper has a figure here, which this page does not show.
We conclude for each i\in\{\,0,\ldots,n\,\} that
\begin{aligned}
{G'}
\supset
\{\,
t^{-1} g_{0}(u) t : u\in \tilde{U}, t\in S_{i}(u)
\,\}
\supset
\{\, g_{i}(u) : u\in \tilde{U}\,\}
=
V_{i}\cap r^{-1}(\tilde{U}).
\end{aligned}
Since g_{i}(u)\neq g_{j}(u) for all u\in \tilde{U} and i\neq j in \{\,0,\ldots,n\,\} and since S is abelian, it follows from s_{i}(u)^{-1} g_{0}(u) s_{i}(u)
=
g_{i}(u) that g_{i}(u)\in {G'}\setminus S, so V_{i}\cap r^{-1}(\tilde{U})
\subset \operatorname{int}_{G}\left({G'}\right)\setminus S, and that
|\operatorname{Ad}_{S}(g_0(u))|\geq n+1.
We have shown: for any fixed element \overline{u} of U, we have a neighborhood \tilde{U}\subset U such that
| \operatorname{Ad}_{S}(g_0(u))|
\geq
|
\operatorname{Ad}_{S}(g_0(\overline{u}))|
for all u\in \tilde{U}. Since \overline{u} was arbitrary, this finishes our proof. (Claim ?)
Claim. There exists n\geq N
and a non-empty open set \tilde{X}\subset X such that
\left|
\operatorname{Ad}_{S}(\pi(e))
\right|
=
\left|
\operatorname{Ad}_{S}(e)
\right|
= n+1
\quad\text{for all }e\in \tilde{X}.
Proof (Proof of Claim ?). As any e\in X already satisfies the equality of set-cardinalities, it suffices to force the equality
\left|
\operatorname{Ad}_{S}(\pi(e))
\right|
= n+1.
By lower semi-continuity (Claim ?), for each
n\geq N,
the set
C_{n} = \{ u\in U: |\operatorname{Ad}_{S}(g_0(u))|\leq n+1\}
is closed in U. By our assumption on X
at (Proof), we have
U = \bigcup_{
n\geq N
}
C_{n}.
Since G^{(0)} is locally compact, Hausdorff, and since U is nonempty and open in G^{(0)}, it is a Baire space [Kelley_book]*34 Theorem. Therefore at least one of the closed sets has an interior point [Schechter_book]*20.15. If we let n be the smallest element
in \{N,N+1,\ldots\}
for which \operatorname{int}_{G^{(0)}}\left(C_n\right)=\operatorname{int}_{U}\left(C_n\right) is non-empty, then there exists \overline{u}\in \operatorname{int}_{G^{(0)}}\left(C_n\right) with |\operatorname{Ad}_{S}(g_0(\overline{u}))|=n+1. By lower semi-continuity (Claim ?), there exists an open neighborhood \tilde{U}\subset U of \overline{u} such that, for all u\in \tilde{U}, we have
|\operatorname{Ad}_{S}(g_0(u))|
\geq
|\operatorname{Ad}_{S}(g_0(\overline{u}))|
= n+1
.
By definition of C_n, this implies that, for all u in the non-empty, open set \tilde{U}\cap \operatorname{int}_{G^{(0)}}\left(C_n\right), we have
|\operatorname{Ad}_{S}(g_0(u))|
=
n+1
.
We may now let \tilde{X}\coloneq
r_{\mathcal{E}}^{-1} \left( \tilde{U}\cap \operatorname{int}_{G^{(0)}}\left(C_n\right)\right)
\subset r_{\mathcal{E}}^{-1}(U) = X. (Claim ?)
Assume now that Condition (?) fails, i.e., there exists a non-empty set X\subset {\mathcal{E}'} that is open in \mathcal{E} and such that Equation (Proof) holds for N=1, i.e.,
1 < \left|
\operatorname{Ad}_{S}(\pi(e))
\right|
=
\left|
\operatorname{Ad}_{S}(e)
\right|
< \infty
\quad\text{for all }e\in X.
Note that (?) implies that X\cap \mathcal{E}_{S}=\emptyset, since |\operatorname{Ad}_{S}(s)|=1 for any s\in S because S is abelian. Thus, we may invoke the above claims; in particular, by
replacing X with its non-empty open subset \tilde{X} from Claim ? and by scavenging the proof of Claim ? (i.e., shrinking X even further), we can now make the following assumptions:
The number n\in \mathbf{Z}_{>0} is such that every element e in the open set \tilde{X}=X satisfies (?), (?) (not just (?));
for each i\in\{\,0,\ldots, n\,\}, we have pairwise disjoint open bisections V_{i}\subset
\operatorname{int}_{G}\left({G'}\right)\setminus S and W_{i}\subset S, and homeomorphisms g_{i}(\cdot)\colon U\coloneq r(X)\to V_{i} and s_{i}(\cdot)\colon U\to W_{i} such that \pi(X)=V_0 and
s_{i}(u)^{-1} g_{0}(u) s_{i}(u)
=
g_{i}(u)
\quad\text{for all } u\in U;
for i=0, we have W_{0}=U and s_{0}(u)=u.
By the property at (?), (?) of X=\tilde{X} and the fact that the V_{i} are mutually disjoint, we conclude from Equation (?) that
\operatorname{Ad}_{S}(g_0(u))
= \{\, g_0(u),\ldots, g_n(u)\,\}
\subset
G(u)
for all u\in U. This in turn implies that sets
\begin{aligned}
S_{i}(u)
=
\{\,
t\in S(u): t^{-1} g_0(u)t=g_{i}(u)
\,\},
\end{aligned}
which are non-empty by Equation (?), are a partition of S(u). Indeed, if s\in S(u), then s^{-1} g_0(u) s\in \operatorname{Ad}_{S} (g_0(u)), so s^{-1} g_0(u) s=g_i(u) for some i\in\{\,0,\ldots,n\,\}, meaning that s\in S_i (u). Thus,
\begin{aligned}
S(u) = \bigsqcup_{i=0}^{n} S_{i}(u)
\quad\text{for all } u\in U.
\end{aligned}
Next, we will deal with the \mathcal{E}_{S}-conjugates by ‘lifting’ the maps g_i(\cdot). By shrinking U, we can without loss of generality assume that the bisections V_0 and W_0,\ldots,W_n allow local trivializations of the principal \mathbf{T}-bundle \mathcal{E}, meaning that there exist \mathbf{T}-equivariant homeomorphisms \psi\colon \mathbf{T}\times V_{0} \to \pi^{-1}(V_{0}) and, for each i\in\{\,
1
,\ldots,n\,\}, \phi_i\colon \mathbf{T}\times W_{i}\to \pi^{-1}(W_{i})\subset \mathcal{E}_{S} such that the diagrams
\begin{tikzcd}[column sep = small]
\mathbf{T}\times V_{0} \ar[rr, "\psi"]\ar[rd,"\mathrm{pr}_{2}"'] && \pi^{-1}(V_{0})\ar[ld, "\pi"]
\\
& V_{0}
\end{tikzcd}
\quad\text{and}\quad
\begin{tikzcd}[column sep = small]
\mathbf{T}\times W_{i} \ar[rr, "\phi_{i}"]\ar[rd,"\mathrm{pr}_{2}"'] && \pi^{-1}(W_{i})\ar[ld, "\pi"]
\\
& W_{i}
\end{tikzcd}
commute. Since W_0=U\subset G^{(0)}, we may choose for i=0 the trivialization \phi_0\colon \mathbf{T}\times W_{0}\to \pi^{-1}(W_0) given by \phi_0 (z,u)=\iota(z,u). For i\in\{\,0,\ldots, n\,\}, the following maps are injective and continuous:
\begin{aligned}
\sigma_{i}(\cdot)&\colon U\to \pi^{-1} (W_{i})
&\text{ defined by }&&\sigma_{i}(u)&\coloneq \phi_{i} (1,s_{i}(u)), \text{ and}
\\
e_{i}(\cdot)&\colon U \to \pi^{-1}(V_i)
&\text{ defined by }&&
e_{i}(u) &\coloneq \sigma_{i}(u)^{-1} \psi (1,g_{0}(u))\sigma_{i}(u).
\end{aligned}
Since each \sigma_i(\cdot) takes values in \mathcal{E}_{S}, the e_{i}(u) are, by their very definition, \mathcal{E}_S-conjugates of each other. We therefore have
the following equality of sets:
\operatorname{Ad}_{S}(e_i(u))=\operatorname{Ad}_{S}(e_j(u)).
For any i, commutativity of the diagrams at (?) yields that
\begin{aligned}
\pi(e_{i}(u))
&=
\pi\left(
\sigma_{i}(u)^{-1} \psi (1,g_{0}(u))\sigma_{i}(u)
\right)
&&\text{by definition of $e_{i}(u)$}
\\
&=
\pi\left(
\phi_{i} (1,s_{i}(u))
\right)
^{-1}
\pi\left(\psi (1,g_{0}(u))\right)
\pi\left(
\phi_{i} (1,s_{i}(u))
\right)
&&\text{by definition of $\sigma_{i}(u)$}
\\
&=
s_{i}(u)^{-1} g_0(u) s_{i}(u)
&&\text{by~\eqref{diags:psi and phi_i}}
\\
&=g_{i}(u)
&&\text{by Equation~\eqref{eq:s_i conjugate}.}
\end{aligned}
Thus, for all i\neq j, it follows from \pi(e_{i}(u))=g_{i}(u)\neq g_{j}(u)=\pi(e_{j}(u)) that e_{i}(u)\neq e_{j}(u). Since (?), (?) implies for each u\in U that
\Omega_{S}(e_{i}(u))=\Omega_{S}(e_{0}(u))
=
|\operatorname{Ad}_{S}(g_0(u))|
=
n+1,
we conclude that
\begin{aligned}
\operatorname{Ad}_{S}(e_i(u))=\operatorname{Ad}_{S}(e_0(u))
=
\{\,e_{0}(u),\ldots, e_{n}(u)\,\}.
\end{aligned}
For each i\in\{\,0,\ldots,n\,\}, we define a \mathbf{T}-equivariant homeomorphism
\Psi_{i}\colon \pi^{-1}(V_{i})\to \pi^{-1}(V_0)
\quad\text{ by }\quad
e\mapsto \sigma_{i}(r(e)) \, e \, \sigma_{i}(r(e))^{-1}.
Note that every element e of \pi^{-1}(V_{i}) can be written as e=z\cdot e_{i}(u) for some unique z\in\mathbf{T} and u=r(e)\in r(V_{i})= U, since \pi(e)\in V_{i}\cap G(u)=\{\,g_{i}(u)\,\}. Since we assumed that s_0(u)=u and \phi_0 (z,u)=\iota(z,u) for all u and z\in\mathbf{T}, we further have that \sigma_0(u)=
\iota(1,u) \in \mathcal{E}^{(0)} for all u, so that we can write \Psi_{i} as follows:
\begin{aligned}
\Psi_{i}(z\cdot e_{i}(u))
&=
z\cdot
\sigma_{i}(u) e_{i}(u) \sigma_{i}(u)^{-1}
&&\text{by definition of $\Psi_{i}$}
\\&=
z\cdot
\psi (1,g_{0}(u))
&&\text{by definition of
$e_{i}(\cdot)$
}
\\
&=
z\cdot
\sigma_{0}(u) e_{0}(u)
\sigma_{0}(u)^{-1}
&&
\text{by definition of
$e_{0}(\cdot)$
}
\\
&=
z\cdot e_0(u)
&&\text{
since $\sigma_0(u)=\iota(1,u)$.
}
\end{aligned}
Now assume that \sigma\in \mathcal{E}_{S}(u), so it is composable with e=z\cdot e_i(u)\in\pi^{-1}(V_{i}). Then it follows from Equations (?) and (?) that
\sigma^{-1} e \sigma
\in
\operatorname{Ad}_{S}(e)
=
\{\,z\cdot e_{0}(u),\ldots, z\cdot e_{n}(u)\,\}.
On the other hand, by Equation (?), we have s_{i}(u)\pi(\sigma)\in S_{j}(u) for a unique j, so that (?), (?) implies
\begin{aligned}
\pi(\sigma^{-1} e\sigma)
&=
\pi(\sigma)^{-1} g_{i}(u) \pi(\sigma)
&&\text{since $\pi(e)=\pi(e_i(u))$}
\\
&=
\pi(\sigma)^{-1} (s_{i}(u)^{-1} g_{0}(u) s_{i}(u)) \pi(\sigma)
&&
\text{by~\eqref{eq:s_i conjugate}}
\\
&=
g_{j}(u)
&&
\text{since $s_{i}(u) \pi(\sigma)\in S_{j}(u)$ and by~\eqref{eq:s_i conjugate}}
&&
\\
&=
\pi(e_{j}(u))&&
\text{by~\eqref{diags:psi and phi_i}}.
\end{aligned}
In total, we have that
\sigma^{-1} e\sigma \in \{\,z\cdot e_{0}(u),\ldots, z\cdot e_{n}(u)\,\} \cap \mathbf{T}\cdot e_{j}(u),
so \sigma^{-1} e \sigma = z\cdot e_{j}(u). Hence by (?), \Psi_{j}(\sigma^{-1} e \sigma) = z\cdot e_{0}(u) = \Psi_{i}(e). (Lemma 3.21)