The proof of “chi-square statistics follows chi-square distribution”

chi-square test(principle used in C4.5’s CVP Pruning),
also called chi-square statistics,
also called chi-square goodness-of fit

here is the contingency table:

The target is to prove:
$$\sum _{i=1}^{i=r}\sum _{j=1}^{j=s}\frac{[X_{ij}-N_{i\cdot }(\frac{N_{\cdot j}}{n}){]}^2}{N_{i\cdot }(\frac{N_j}{n})}\sim {\chi }^2[(r-1)(s-1)]①$$

Note:
the left side of above is “discrete”
the right side of above is “continuous”
----------------------------------------------
Let’s review the concepts of “Multi-dimensional Normal Distribution”,
according to[1]

$$X\sim N(\mu ,\sum )$$
$$\mu =[E[X_1],E[X_2],\cdot \cdot \cdot ,E[X_s]{]}^T$$
$$\sum =:[Cov[X_i,X_j];1\le i,j\le s]$$

-----------------------------------------------------------------------------------------------

${\sum }_{j=1}^{j=s}\frac{[X_{ij}-N_{i\cdot }(\frac{N_{\cdot j}}{n}){]}^2}{N_{i\cdot }(\frac{N_{\cdot j}}{n})}$

=$N_{i\cdot }{\sum }_{j=1}^{j=s}\frac{[\frac{X_{ij}}{N_{i\cdot }}-(\frac{N_{\cdot j}}{n}){]}^2}{(\frac{N_{\cdot j}}{n})}$

= N i ⋅ { [ ∑ j = 1 j = s − 1 [ X i j N i ⋅ − ( N ⋅ j n ) ] 2 N ⋅ j n ] + [ X i s N i ⋅ − ( N ⋅ s n ) ] 2 N ⋅ s n } N_{i·}\{[\sum_{j=1}^{j=s-1}\frac{[\frac{X_{ij}}{N_{i·}}-(\frac{N_{·j}}{n})]^2}{\frac{N_{·j}}{n}}]+ \frac{[\frac{X_{is}}{N_{i·}}-(\frac{N_{·s}}{n})]^2} {\frac{N_{·s}}{n}}\} Ni⋅​{[∑j=1j=s−1​nN⋅j​​[Ni⋅​Xij​​−(nN⋅j​​)]2​]+nN⋅s​​[Ni⋅​Xis​​−(nN⋅s​​)]2​}

= N i ⋅ { [ ∑ j = 1 j = s − 1 [ X i j N i ⋅ − ( N ⋅ j n ) ] 2 N ⋅ j N i ⋅ ] + [ ∑ j = 1 j = s − 1 ( X i j N i ⋅ − N ⋅ j n ) ] 2 N s N i ⋅ } N_{i·} \{[\sum_{j=1}^{j=s-1}\frac{[ \frac{X_{ij}}{N_{i·}}-(\frac{N_{·j}}{n})]^2 }{ \frac{N_{·j}}{N_{i·}} }]+ \frac{[\sum_{j=1}^{j=s-1}(\frac{X_{ij}}{N_{i·}}-\frac{N_{·j}}{n})]^2}{{\frac{Ns}{N_{i·}}}} \} Ni⋅​{[∑j=1j=s−1​Ni⋅​N⋅j​​[Ni⋅​Xij​​−(nN⋅j​​)]2​]+Ni⋅​Ns​[∑j=1j=s−1​(Ni⋅​Xij​​−nN⋅j​​)]2​}

Let’s set
$p^{\ast }=(\frac{N_{\cdot 1}}{n},...,\frac{N_{\cdot (s-1)}}{n}{)}^T$

${\overline{X}}^{\ast }=(\frac{X_{i1}}{N_{i\cdot }},\cdot \cdot \cdot ,\frac{X_{i(s-1)}}{N_{i\cdot }}{)}^T$

So,
$$N_{i\cdot }\sum _{j=1}^{j=s}\frac{[\frac{X_{ij}}{N_{i\cdot }}-(\frac{N_{\cdot j}}{n}){]}^2}{(\frac{N_{\cdot j}}{n})}$$

$=N_{i\cdot }({\overline{X}}^{\ast }-p^{\ast }{)}^T({\sum }^{\ast }{)}^{-1}({\overline{X}}^{\ast }-p^{\ast })$
where ${\sum }^{\ast }=$

$$\left[\begin{array}{cccc}{p}_1& 0& \cdot \cdot \cdot & 0\\ 0& p_2& \cdot \cdot \cdot & 0\\ ⋮& ⋮& \ddots & ⋮\\ 0& 0& \cdot \cdot \cdot & p_{s-1}\end{array}\right]-\left[\begin{array}{c}{p}_1\\ p_2\\ ⋮\\ p_{s-1}\end{array}\right]{\left[\begin{array}{c}{p}_1\\ p_2\\ ⋮\\ p_{s-1}\end{array}\right]}^T$$

According to Sherman-Morison Formula:
$({\sum }^{\ast }{)}^{-1}=$

$$\left[\begin{array}{cccc}\frac{1}{p_1}& 0& \cdot \cdot \cdot & 0\\ 0& \frac{1}{p_2}& \cdot \cdot \cdot & 0\\ ⋮& ⋮& \ddots & ⋮\\ 0& 0& \cdot \cdot \cdot & \frac{1}{p_{s-1}}\end{array}\right]-\frac{1}{p_s}\left[\begin{array}{cccc}1& 1& \cdot \cdot \cdot & 1\\ 1& 1& \cdot \cdot \cdot & 1\\ ⋮& ⋮& \ddots & ⋮\\ 1& 1& \cdot \cdot \cdot & 1\end{array}\right]$$

Let’s set $Y_i=\sqrt{N_{i\cdot }}\frac{{\overline{X}}^{\ast }-p^{\ast }}{\sqrt{{\sum }^{\ast }}}②$
according [3]:
------------------------the following are from wikipedia-------------------------------
$\left[\begin{array}{c}{X}_{1(1)}\\ ⋮\\ X_{1(k)}\end{array}\right]+\left[\begin{array}{c}{X}_{2(1)}\\ ⋮\\ X_{2(k)}\end{array}\right]+\cdots +\left[\begin{array}{c}{X}_{n(1)}\\ ⋮\\ X_{n(k)}\end{array}\right]=\left[\begin{array}{c}{\sum }_{i=1}^n\left[X_{i(1)}\right]\\ ⋮\\ {\sum }_{i=1}^n\left[X_{i(k)}\right]\end{array}\right]=\sum _{i=1}^n{\mathbf{X}}_i$

and the average is

$\frac{1}{n}\sum _{i=1}^n{\mathbf{X}}_i=\frac{1}{n}\left[\begin{array}{c}{\sum }_{i=1}^n{X}_{i(1)}\\ ⋮\\ {\sum }_{i=1}^n{X}_{i(k)}\end{array}\right]=\left[\begin{array}{c}{\bar{X}}_{i(1)}\\ ⋮\\ {\bar{X}}_{i(k)}\end{array}\right]={\bar{\mathbf{X}}}_{\mathbf{n}}\frac{1}{n}\sum _{i=1}^n{\mathbf{X}}_i=\frac{1}{n}\left[\begin{array}{c}{\sum }_{i=1}^n{X}_{i(1)}\\ ⋮\\ {\sum }_{i=1}^n{X}_{i(k)}\end{array}\right]=\left[\begin{array}{c}{\bar{X}}_{i(1)}\\ ⋮\\ {\bar{X}}_{i(k)}\end{array}\right]={\bar{\mathbf{X}}}_{\mathbf{n}}$
and therefore

$\frac{1}{\sqrt{n}}\sum _{i=1}^n\left[{\mathbf{X}}_i-E\left(X_i\right)\right]=\frac{1}{\sqrt{n}}\sum _{i=1}^n({\mathbf{X}}_i-\mathbf{\mu })=\sqrt{n}\left({\overline{\mathbf{X}}}_n-\mathbf{\mu }\right).\frac{1}{\sqrt{n}}\sum _{i=1}^n\left[{\mathbf{X}}_i-E\left(X_i\right)\right]=\frac{1}{\sqrt{n}}\sum _{i=1}^n({\mathbf{X}}_i-\mathbf{\mu })=\sqrt{n}\left({\overline{\mathbf{X}}}_n-\mathbf{\mu }\right).$
The multivariate central limit theorem states that
$$\sqrt{n}\left({\overline{\mathbf{X}}}_n-\mathbf{\mu }\right)\stackrel{D}{\to }{N}_k(0,\mathbf{\Sigma })$$

------------------------the above are from wikipedia-------------------------------

So,for②，we can get
$$Y_i\sim N_{s-1}(0,I_{s-1})③$$
where
$0=[0,0,\dots ,0{]}^T$
$I_{s-1}=E_{(s-1)\cdot (s-1)}$
then for ①

$$\begin{gathered} \sum _{i=1}^{i=r}\sum _{j=1}^{j=s}\frac{[X_{ij}-N_{i\cdot }(\frac{N_{\cdot j}}{n}){]}^2}{N_{i\cdot }(\frac{N_{\cdot j}}{n})} \\ ＝\sum _{i=1}^{i=r}{Y}_i^2 \end{gathered}$$

Because of ③，
$$\sum _{i=1}^{i=r}{Y}_i^2\sim {\chi }^2[(s-1)(r-1)]$$

The Chi-Square statistics was invented by Pearson[8].
Reference:
[1]https://en.wikipedia.org/wiki/Multivariate_normal_distribution
[2]《Seven different proofs for the Pearson independence test》
[3]https://en.wikipedia.org/wiki/Central_limit_theorem
[4]https://ocw.mit.edu/courses/mathematics/18-443-statistics-for-applications-fall-2003/lecture-notes/lec23.pdf
[5]https://arxiv.org/pdf/1808.09171.pdf
[6]https://www.math.utah.edu/~davar/ps-pdf-files/Chisquared.pdf
[7]http://personal.psu.edu/drh20/asymp/fall2006/lectures/ANGELchpt07.pdf
[8]https://download.csdn.net/download/appleyuchi/10834144