Last part of the proof of Theorem 1. It remains to show
$$\begin{aligned} J_n \cdot I_{B_n} \rightarrow 0 \quad a.s. \end{aligned}$$
Application of the conditional McDiarmid inequality as in the proof of Theorem 1 yields
$$\begin{aligned} J_n \cdot I_{B_n} - {\mathbf E}\{ J_n \cdot I_{B_n} | X_1, \dots , X_n \} \rightarrow 0 \quad a.s. \end{aligned}$$
Hence, it suffices to show
$$\begin{aligned} {\mathbf E}\{ J_n | X_1, \dots , X_n \} \rightarrow 0 \quad a.s. \end{aligned}$$
(30)
By the inequality of Jensen, the independence of the data and \(|Y| \le L\)a.s., we get
$$\begin{aligned}&\left( {\mathbf E}\{ J_n | X_1, \dots , X_n \} \right) ^2\\&\quad \le {\mathbf E}\{ J_n^2 | X_1, \dots , X_n \}\\&\quad \le {\mathbf E}\left\{ \int \left| \sum _{i=1}^n W_{n,i}(x) \cdot (Y_i-m(X_i)) \right| ^2 \mu ({\hbox {d}}x) \bigg | X_1, \dots , X_n \right\} \\&\quad = {\mathbf E}\left\{ \left| \sum _{i=1}^n W_{n,i}(X) \cdot (Y_i-m(X_i)) \right| ^2 \bigg | X_1, \dots , X_n \right\} \\&\quad = {\mathbf E}\left\{ {\mathbf E}\left\{ \left| \sum _{i=1}^n W_{n,i}(X) \cdot (Y_i-m(X_i)) \right| ^2 \bigg | X, X_1, \dots , X_n \right\} \bigg | X_1, \dots , X_n \right\} \\&\quad = {\mathbf E}\left\{ \sum _{i=1}^n W_{n,i}(X)^2 \cdot {\mathbf E}\left\{ (Y_i-m(X_i))^2 \bigg | X, X_1, \dots , X_n \right\} \bigg | X_1, \dots , X_n \right\} \\&\quad \le 4 L^2 \cdot {\mathbf E}\left\{ \sum _{i=1}^n W_{n,i}(X)^2 \bigg | X_1, \dots , X_n \right\} \\&\quad = 4 L^2 \cdot \sum _{i=1}^n \int |W_{n,i}(x)|^2 \mu ({\hbox {d}}x). \end{aligned}$$
Thus, (30) follows from (29).
Proof of (29) in the context of Lemma 6. On the one hand, we have
$$\begin{aligned} \sum _{i=1}^n W_{n,i}(x)^2 = \frac{\sum _{i=1}^n K \left( \frac{x-X_i}{h_n} \right) ^2 }{ \left( \sum _{j=1}^n K \left( \frac{x-X_j}{h_n} \right) \right) ^2 } \le 1. \end{aligned}$$
On the other hand, it holds
$$\begin{aligned} \sum _{i=1}^n W_{n,i}(x)^2\le & {} c_2 \cdot \frac{\sum _{i=1}^n K \left( \frac{x-X_i}{h_n} \right) }{ \left( \sum _{j=1}^n K \left( \frac{x-X_j}{h_n} \right) \right) ^2 } \cdot I_{\left\{ \sum _{j=1}^n K \left( \frac{x-X_j}{h_n} \right) >0\right\} }\\\le & {} c_2 \cdot \frac{1}{ \sum _{j=1}^n K \left( \frac{x-X_j}{h_n} \right) }. \end{aligned}$$
Consequently,
$$\begin{aligned} \sum _{i=1}^n W_{n,i}(x)^2\le & {} \min \left\{ 1, c_2 \cdot \frac{1}{ \sum _{j=1}^n K \left( \frac{x-X_j}{h_n} \right) } \right\} \\\le & {} \min \left\{ 1, \frac{c_2}{c_1} \cdot \frac{1}{ \sum _{j=1}^n I_{S_{r_1}} \left( \frac{x-X_j}{h_n} \right) } \right\} \\\le & {} \max \left\{ 1, \frac{c_2}{c_1} \right\} \cdot \min \left\{ 1, \frac{1}{ \sum _{j=1}^n I_{S_{r_1}} \left( \frac{x-X_j}{h_n} \right) } \right\} \\\le & {} \max \left\{ 1, \frac{c_2}{c_1} \right\} \cdot \frac{2}{ 1+\sum _{j=1}^n I_{S_{r_1}} \left( \frac{x-X_j}{h_n} \right) }. \end{aligned}$$
Hence, it suffices to show
$$\begin{aligned} W_n:= \int \frac{1}{ 1+\sum _{j=1}^n I_{S_{r_1}} \left( \frac{x-X_j}{h_n} \right) } \mu ({\hbox {d}}x) \rightarrow 0 \quad a.s. \end{aligned}$$
(31)
For any bounded sphere S around 0, by Lemma 2a and by assumption (9), we get
$$\begin{aligned}&{\mathbf E}\left\{ \int _S \frac{1}{ 1+\sum _{j=1}^n I_{S_{r_1}} \left( \frac{x-X_j}{h_n} \right) } \mu ({\hbox {d}}x) \right\} \\&\quad = \int _S {\mathbf E}\left\{ \frac{1}{ 1+\sum _{j=1}^n I_{S_{r_1}} \left( \frac{x-X_j}{h_n} \right) } \right\} \mu ({\hbox {d}}x)\\&\quad \le \int _S \frac{1}{ n \cdot \mu (x+h_n \cdot S_{r_1}) } \mu ({\hbox {d}}x)\\&\quad \le \frac{const}{n \cdot h_n^d} \rightarrow 0 \quad (n \rightarrow \infty ), \end{aligned}$$
where the last inequality holds because of equation (5.1) in Györfi et al. (2002).
Thus, it suffices to show
$$\begin{aligned} W_n - {\mathbf E}\{W_n\} \rightarrow 0 \quad a.s. \end{aligned}$$
(32)
Analogously to the proof of (A4), with \(X_1^\prime \), \(X_1\), ..., \(X_n\) independent and identically distributed and
$$\begin{aligned} W_n^\prime := \int \frac{1}{ 1+ I_{S_{r_1}} \left( \frac{x-X_1^\prime }{h_n} \right) + \sum _{j=2}^n I_{S_{r_1}} \left( \frac{x-X_j}{h_n} \right) } \mu ({\hbox {d}}x), \end{aligned}$$
by Lemma 4.2 in Kohler et al. (2003), one has
$$\begin{aligned} {\mathbf E}\{|W_n-{\mathbf E}\{W_n\}|^4\} \le c_{11} \cdot n^2 \cdot {\mathbf E}\{ (W_n-W_n^\prime )^4\} \quad (n \in \mathbb {N}). \end{aligned}$$
Furthermore, by the second part of Lemma 5 one gets
$$\begin{aligned}&{\mathbf E}\{|W_n-W_n^\prime |^4\}\\&\quad \le 16 \cdot {\mathbf E}\left\{ \left( \int \frac{ I_{S_{r_1}} \left( \frac{x-X_1}{h_n} \right) }{ \left( 1 + \sum _{j=2}^n I_{S_{r_1}} \left( \frac{x-X_j}{h_n} \right) \right) ^2 } \mu ({\hbox {d}}x) \right) ^4 \right\} \\&\quad \le 16 \cdot {\mathbf E}\left\{ \left( \int \frac{ I_{S_{r_1}} \left( \frac{x-X_1}{h_n} \right) }{ 1 + \sum _{j=2}^n I_{S_{r_1}} \left( \frac{x-X_j}{h_n} \right) } \mu ({\hbox {d}}x) \right) ^4 \right\} \\&\quad \le \frac{const}{n^4}. \end{aligned}$$
From these relations, one obtains (32) by the Borel–Cantelli lemma and the Markov inequality.
Proof of (29) in the context of Lemma 9. Analogously to above it suffices to show
$$\begin{aligned} V_n := \int \frac{1}{ 1+\sum _{j=1}^n I_{A_{\mathcal{P}_n}(x)} \left( X_j \right) } \mu ({\hbox {d}}x) \rightarrow 0 \quad a.s. \end{aligned}$$
For any bounded sphere S around zero, by assumption (12) we get
$$\begin{aligned} \int _S \frac{1}{n \cdot \mu (A_{\mathcal{P}_n}(x))} \mu ({\hbox {d}}x) \rightarrow 0 \quad (n \rightarrow \infty ), \end{aligned}$$
from which by Lemma 2a we can conclude analogously to above
$$\begin{aligned} {\mathbf E}V_n \rightarrow 0 \quad (n \rightarrow \infty ). \end{aligned}$$
Hence, it suffices to show
$$\begin{aligned} V_n - {\mathbf E}\{V_n\} \rightarrow 0 \quad a.s., \end{aligned}$$
which follows analogously to above from the second part of Lemma 7.