This is with TikZ, forgive me.
The problem is that what you want to draw is too large for only one column of a two-column document.
Here are two possible solutions: in one column, with different positions of the description, or in two columns.
\documentclass[twocolumn]{article}
\usepackage{mwe}% <--- for testing purpose only
\usepackage{cuted}
\usepackage{tikz}
\usetikzlibrary{matrix, fit}
\usetikzlibrary{arrows.meta}
\tikzset{mymatrixstyle/.style={matrix of math nodes,
ampersand replacement=\&,
left delimiter=(,
right delimiter=),
inner sep=0pt,
outer sep=0pt,
row sep=2pt, column sep=10pt},
myarrow/.style={-{Straight Barb[angle=60:3pt 3]}}}
\usepackage{mathtools}
\mathtoolsset{showonlyrefs}
\begin{document}
\blindtext % <--- for testing purpose only
In one column:
%\begin{eqnarray} see https://tex.stackexchange.com/a/197/101651
\begin{align}
\mathbf{N} &= \begin{pmatrix}
n_{11} & 0 & n_{13} & n_{14} \
n_{21} & n_{22} & n_{23} & n_{24} \
n_{31} & n_{32} & n_{33} & 0 \
n_{41} & n_{42} & n_{43} & n_{44}
\end{pmatrix}\
&= \begin{tikzpicture}[baseline]
\matrix[mymatrixstyle] (mymatr) {
\sqrt{\beta_x} & 0 & n_{13} & n_{14} &[-7pt]\
-\dfrac{\alpha_x}{\sqrt{\beta_x}} & n_{22} & n_{23} & n_{24} \
n_{31} & n_{32} & \sqrt{\beta_y} & 0 \
n_{41} & n_{42} & -\dfrac{\alpha_y}{\sqrt{\beta_y}} & n_{44}\};
\node[draw, fit=(mymatr-3-1)(mymatr-3-2)] (n3132) {};
\node[draw, fit=(mymatr-4-1)(mymatr-4-2)] (n4142) {};
\node[draw, fit=(mymatr-1-3)(mymatr-1-4)] (n1314) {};
\node[draw, fit=(mymatr-2-3)(mymatr-2-4)] (n2324) {};
\draw[myarrow, overlay] (n3132) -- ++(-1.8,0) -| +(0,-1.6) node[anchor=north] {$\zeta_y=n_{31}+in_{32}$};
\draw[myarrow] (n4142) -- +(0,-1.8) node[anchor=north] {$\tilde{\zeta}y=n{41}-in_{42}$};
\draw[myarrow, overlay] (n1314) -- ++(2,0) -| +(0,-3.6) node[anchor=20] {$\zeta_y=n_{13}-in_{14}$};
\draw[myarrow, overlay] (n2324) -- ++(1.4,0) -| +(0,-2.1) node[anchor=20] {$\tilde{\zeta}y=n{23}-in_{24}$};
\end{tikzpicture}
\end{align}
\blindtext[7] % <--- for testing purpose only
In two columns:
\begin{strip}
\begin{align}
\mathbf{N} &= \begin{pmatrix}
n_{11} & 0 & n_{13} & n_{14} \
n_{21} & n_{22} & n_{23} & n_{24} \
n_{31} & n_{32} & n_{33} & 0 \
n_{41} & n_{42} & n_{43} & n_{44}
\end{pmatrix}\
&= \begin{tikzpicture}[baseline]
\matrix[mymatrixstyle] (mymatr) {
\sqrt{\beta_x} & 0 & n_{13} & n_{14} &[-7pt]\
-\dfrac{\alpha_x}{\sqrt{\beta_x}} & n_{22} & n_{23} & n_{24} \
n_{31} & n_{32} & \sqrt{\beta_y} & 0 \
n_{41} & n_{42} & -\dfrac{\alpha_y}{\sqrt{\beta_y}} & n_{44}\};
\node[draw, fit=(mymatr-3-1)(mymatr-3-2)] (n3132) {};
\node[draw, fit=(mymatr-4-1)(mymatr-4-2)] (n4142) {};
\node[draw, fit=(mymatr-1-3)(mymatr-1-4)] (n1314) {};
\node[draw, fit=(mymatr-2-3)(mymatr-2-4)] (n2324) {};
\draw[myarrow, overlay] (n3132) -- +(-1.8,0) node[anchor=east] {$\zeta_y=n_{31}+in_{32}$};
\draw[myarrow, overlay] (n4142) -- +(-1.8,0) node[anchor=east] {$\tilde{\zeta}y=n{41}-in_{42}$};
\draw[myarrow, overlay] (n1314) -- +(1.8,0) node[anchor=west] {$\zeta_y=n_{13}-in_{14}$};
\draw[myarrow, overlay] (n2324) -- +(1.8,0) node[anchor=west] {$\tilde{\zeta}y=n{23}-in_{24}$};
\end{tikzpicture}
\end{align}
\end{strip}
\blindtext[7] % <--- for testing purpose only
\end{document}


Off-topic: do not use eqnarray, see here for the reasons: https://tex.stackexchange.com/a/197/101651
\tikzmarks into the matrix and overlay a tikzpicture. There are many example here. – John Kormylo Apr 01 '22 at 13:36\tikzmark(depending on how the matrices are written). FYI, It is always best to compose a fully compilable MWE that sets up the problem including the\documentclassand the appropriate packages so that those trying to help don't have to recreate it. In this case the MWE could produce the two matrcies -- the solution's depend on exactly how you are creating the matrices. – Peter Grill Apr 01 '22 at 17:22