Here, we establish the proof of Theorem ?.
As the cdf G_{\alpha} of the PL-APT belongs to the attraction domain of Gumbel, it is known that to estimate the extremal index it is equivalent to use in the calculus the G_{\alpha}^{-1}(1-u) or \log G_{\alpha}^{-1}(1-u).
Let
\begin{aligned}
R_{j,n} &=f(j)\left( \log X_{n-j+1,n}-\log X_{n-j,n}\right) ^{s} \\
&=f(j)\left( G_{\alpha}^{-1}\left( 1-U_{j,n}\right) - G_{\alpha}^{-1}\left(
1-U_{j+1,n}\right) \right) ^{s} \\
&=f(j)\left( (1/\theta) \log \left( \frac{U_{j+1,n}}{U_{j,n}}\right) +C_{2}A_{n}+O_{
\mathbb{P}}\left( B_{n}\right) \right) ^{s} \\
&=f(j)\left( (1/\theta)j^{-1}E_{j,n}+C_{2}A_{n}+O_{\mathbb{P}}\left( B_{n}\right)
\right) ^{s}
\end{aligned}
with
A_{n}=\max \left\{ U_{j,n}^{2};U_{j+1,n}^{2}\right\} \leq -2\log \left(
\frac{U_{j+1,n}}{U_{j,n}}\right)
and
O_{\mathbb{P}}\left( B_{n}\right) =O_{\mathbb{P}}\left( \left( \log n\right)
^{-2}\right).
By the mean value theorem and j\in \left\{
1,...,k\left( n\right) \right\} , s>1, we get
\begin{aligned}
& & \\
& & R_{j,n}-(1/\theta)^{s}f(j)j^{-s}E_{j,n}^{s} \\
&\leq &sf(j)\left\vert C_{2}A_{n}+O_{
\mathbb{P}}\left( B_{n}\right) \right\vert \left( (1/\theta)j^{-1}E_{j,n}+\left\vert
C_{2}A_{n}\right\vert +\left\vert O_{\mathbb{P}}\left( B_{n}\right)
\right\vert \right) ^{s-1} \\
&\leq & (1/\theta)sf(j)j^{-1}E_{j,n}\left( (1/\theta)j^{-1}E_{j,n}+\left\vert
C_{2}A_{n}\right\vert +\left\vert O_{\mathbb{P}}\left( B_{n}\right)
\right\vert \right) ^{s-1} \\
&\leq &(1/\theta)sf(j)j^{-1}E_{j,n}\left(
D_{s}(1/\theta)^{s-1}j^{s-1}E_{j,n}^{s-1}+(1/\theta)D_{s}^{2}j^{-1}E_{j,n}+O_{\mathbb{P}
}\left( \frac{D_{s}^{2}}{\left( \log n\right) ^{-2}}\right) \right) .
\end{aligned}
Applying the sum , we get
\begin{aligned}
& & \\
& & \sum_{j=1}^{k\left( n\right) }\left(
R_{j,n}-(1/\theta)^{s}f(j)j^{-s}E_{j,n}^{s}\right) \\
&\leq &(1/\theta)s\sum_{j=1}^{k\left(
n\right) }f(j)j^{-1}E_{j,n}\left(
D_{s}(1/\theta)^{s-1}j^{s-1}E_{j,n}^{s-1}+(1/\theta)D_{s}^{2}j^{-1}E_{j,n}+O_{\mathbb{P}
}\left( \frac{D_{s}^{2}}{\left( \log n\right) ^{-2}}\right) \right) \\
\end{aligned}
and
\begin{aligned}
& & \\
& & \sum_{j=1}^{k\left( n\right) }R_{j,n}-(1/\theta)^{s}\sum_{j=1}^{k\left( n\right)
}f(j)j^{-s}E_{j,n}^{s} \\
&\leq & (1/\theta)s\sum_{j=1}^{k\left( n\right)
}f(j)j^{-1}E_{j,n}\left(
D_{s}(1/\theta)^{s-1}j^{s-1}E_{j,n}^{s-1}+(1/\theta)D_{s}^{2}j^{-1}E_{j,n}+O_{\mathbb{P}
}\left( \frac{D_{s}^{2}}{\left( \log n\right) ^{-2}}\right) \right) \\
\end{aligned}
\begin{aligned}
& & \left\vert T_{n}\left( f,s\right) -(1/\theta)^{s}S_{n}\left( f,s\right) \right\vert \\
&\leq &(1/\theta)sS_{n}\left( f,1\right) \left( D_{s}(1/\theta)^{s-1}S_{n}\left( f,s-1\right)
+(1/\theta)D_{s}^{2}S_{n}\left( f,1\right) +O_{\mathbb{P}}\left( \frac{D_{s}^{2}}{
\left( \log n\right) ^{-2}}\right) \right) \text{ (IL)}
\end{aligned}
where
S_{n}\left( f,s\right) =\sum_{j=1}^{k\left( n\right) }f(j)j^{-s}E_{j,n}^{s}
and
T_{n}\left( f,s\right) =\sum_{j=1}^{k\left( n\right) }f(j)\left( \log
X_{n-j+1,n}-\log X_{n-j,n}\right) ^{s}.
The random variable S_{n}\left( f,s\right) is a sequence of partial sum of random real values and independent random variables indexed by j\in \left\{ 1,...,k\left( n\right) \right\} with first and second moments
\mu _{1}=\Gamma \left( s+1\right) f\left( j\right) j^{-s}\text{ \ and \ }\mu
_{2}=\left( \text{\ }\Gamma \left( 2s+1\right) -\Gamma \left( s+1\right)
^{2}\right) f\left( j\right) j^{-s}.
Its asymptotic normality is given as follows by using the Levy-Feller-Linderberg (see [ips-mfpt-ang], Theorem 20),
\left( \frac{1}{s_{n}\left( f,s\right) }\sum_{j=1}^{k\left( n\right) }\left(
f(j)j^{-s}\left( E_{j,n}^{s}-\text{\ }\Gamma \left( s+1\right) \right)
\right) \leadsto \mathcal{N}\left( 0,1\right) \right)
\text{ \ and }
B_{n}\left( f,s\right) \longrightarrow 0\text{ as }n\longrightarrow +\infty,
where
B_{n}\left( f,s\right) =\frac{1}{C\left( s\right) }\left\{ \frac{\mathbb{V}
ar\left( f(j)j^{-s}\left( E_{j,n}^{s}-\text{\ }\Gamma \left( s+1\right)
\right) \right) }{\sum_{j=1}^{k\left( n\right) }\mathbb{V}ar\left(
f(j)j^{-s}\left( E_{j,n}^{s}-\text{\ }\Gamma \left( s+1\right) \right)
\right) }\right\} .
By combining the Lindeberg condition, the Cauchy-Schwarz inequality and the central limit theorem, we get for S_{n}\left(
f,s\right) the result below
\frac{S_{n}\left( f,s\right) -(1/\theta)^{s}a_{n}\left( f,s\right) }{s_{n}\left(
f,s\right) }\leadsto \mathcal{N}\left( 0,1\right) .
The continuation of the inequality (IL) implies
\begin{aligned}
\frac{T_{n}\left( f,s\right) -(1/\theta)^{s}S_{n}\left( f,s\right) }{s_{n}\left(
f,s\right) }-\frac{S_{n}\left( f,s\right) -(1/\theta)^{s}a_{n}\left( f,s\right) }{
s_{n}\left( f,s\right) } &\leq &O_{\mathbb{P}}\left( \frac{S_{n}\left(
f,1\right) }{S_{n}\left( f,s\right) \log n}\right) \\
\end{aligned}
\begin{aligned}
\frac{T_{n}\left( f,s\right) -S_{n}\left( f,s\right) }{s_{n}\left(
f,s\right) }-\frac{(1/\theta)^{s}\left( S_{n}\left( f,s\right) -a_{n}\left(
f,s\right) \right) }{s_{n}\left( f,s\right) } &\leq &O_{\mathbb{P}}\left(
\frac{S_{n}\left( f,1\right) }{S_{n}\left( f,s\right) \log n}\right). (IL1)
\end{aligned}
The right hand side of the inequality (IL1) above tends to zero in probability if and only if
\begin{aligned}
\left( \frac{S_{n}\left( f,1\right) }{S_{n}\left( f,s\right) \log n}\right) \longrightarrow 0 \ \ \ as \ \ \ n \longrightarrow +\infty .
\end{aligned}
Let
M_{n}\left( f,s\right) =\left( \frac{T_{n}\left( f,s\right) }{a_{n}\left(
f,s\right) }\right) ^{1/s}.
By combining the results above for the left hand side of the inequality (IL1) above, we arrive at
\frac{T_{n}\left( f,s\right) -(1/\theta)^{s}S_{n}\left( f,s\right) }{s_{n}\left(
f,s\right) }=\frac{a_{n}\left( f,s\right) }{s_{n}\left( f,s\right) }\left[
\frac{T_{n}\left( f,s\right) }{a_{n}\left( f,s\right) }-(1/\theta)^{s}\right].
So, we have
\frac{a_{n}\left( f,s\right) }{s_{n}\left( f,s\right) }\left[
\frac{T_{n}\left( f,s\right) }{a_{n}\left( f,s\right) }-(1/\theta)^{s}\right] = Z_{n}+O_{\mathbb{P}}(1).
If a_{n}\left( f,s\right) / s_{n}\left( f,s\right) \longrightarrow +\infty, then we have by using the delta method applied to g(t)=t^{1/s} ,
\frac{a_{n}\left( f,s\right) }{s_{n}\left( f,s\right) }\left[
\frac{T_{n}\left( f,s\right) }{a_{n}\left( f,s\right) }-(1/\theta)^{s}\right] \rightsquigarrow \mathcal{N} \left(0,(1/\theta)^{2s}C^{2}(s) \right).