By a proper ordering of vertices in G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E), the distance matrix D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E)) can be written as
\begin{aligned}
&D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E))\\
&= \scriptsize{\begin{bmatrix} J_{n_1}-I_{n_1}+A(G_{n_1}) & 2J_{n_1\times m_1}-Q(G_{n_1}) & J_{n_1\times n_2}& 2J_{n_1\times n_3} \\ 2J_{m_1\times n_1}-Q(G_{n_1})^T & 2(J_{m_1}-I_{m_1}) & 2J_{m_1\times n_2}& J_{m_1\times n_3}\\ J_{n_2\times n_1} & 2J_{n_2\times m_1} & 2(J_{n_2}-I_{n_2})-A(G_{n_2})& 3J_{n_2\times n_3}\\2J_{n_3\times n_1}& J_{n_3\times m_1}& 3J_{n_3\times n_2}& 2(J_{n_3}-I_{n_3})-A(G_{n_3}) \end{bmatrix}}.
\end{aligned}
Let k_i=\theta_{i1}\geq \theta_{i2}\geq\dots\geq \theta_{in_i} be the A-eigenvalues of G_{n_i}, where i = 1, 2, 3. Let U be a vector such that A(G_{n_1})U= \theta_{1j}U (for j= 2, 3,\ldots, n_1), then by Perron-Frobenius theory J_{1\times n_{1}}U= 0.
Now let, \phi_1= \begin{bmatrix} tU\\Q(G_{n_1})^TU\\0\\0\end{bmatrix} be an eigenvector of D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E)) for the eigenvalue \mu, where t is a real number. Then D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E))\phi_1= \mu \phi_1.
That is,
\scriptsize{\begin{aligned}
&\begin{bmatrix}J_{n_1}-I_{n_1}+A(G_{n_1}) & 2J_{n_1\times m_1}-Q(G_{n_1}) & J_{n_1\times n_2}& 2J_{n_1\times n_3} \\ 2J_{m_1\times n_1}-Q(G_{n_1})^T & 2(J_{m_1}-I_{m_1}) & 2J_{m_1\times n_2}& J_{m_1\times n_3}\\ J_{n_2\times n_1} & 2J_{n_2\times m_1} & 2(J_{n_2}-I_{n_2})-A(G_{n_2})& 3J_{n_2\times n_3}\\2J_{n_3\times n_1}& J_{n_3\times m_1}& 3J_{n_3\times n_2}& 2(J_{n_3}-I_{n_3})-A(G_{n_3}) \end{bmatrix}\begin{bmatrix} tU\\Q(G_{n_1})^TU\\0\\0\end{bmatrix}\\
&= \mu\begin{bmatrix} tU\\Q(G_{n_1})^TU\\0\\0\end{bmatrix}\\
\\
&\begin{bmatrix}(-t+\theta_{1j}t-\theta_{1j}-k_1)U\\(-t-2)Q(G_{n_1})^TU\\0\\0\end{bmatrix}= \begin{bmatrix} \mu tU\\\mu Q(G_{n_1})^TU\\0\\0\end{bmatrix}.
\end{aligned}}
Therefore,
\begin{aligned}
\quad -t+\theta_{1j}t-\theta_{1j}-k_1&= \mu t\\
\quad-t-2&= \mu
\end{aligned}
from this, we get
t^2+(\theta_{1j}+1)t-(\theta_{1j}+k_1)= 0
so that t has two values
\begin{aligned}
&t_1 = \frac{-(\theta_{1j}+1)+ \sqrt{(\theta_{1j}+1)^2+4(\theta_{1j}+k_1)}}{2},\\
& t_2 = \frac{-(\theta_{1j}+1)- \sqrt{(\theta_{1j}+1)^2+4(\theta_{1j}+k_1)}}{2}.
\end{aligned}
That is, -t_1-2 and -t_2-2 are eigenvalues of D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E)) corresponding to the eigenvalues \theta_{1j}\neq k_1 of A(G_{n_1}). Thus, we get 2(n_1-1) eigenvalues of D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E)).
By Lemma 2.1, we get Q(G_{n_1})V=0, where V is an eigenvector of L(G_{n_1}) corresponding to the eigenvalue -2 (repeated m_1-n_1 times).
If \phi_2= \begin{bmatrix}0\\V\\0\\0\end{bmatrix}, then we have
D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E))\phi_2= -2\phi_2.
Therefore, -2 is an eigenvalue of D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E)) repeated m_1-n_1 times.
Consider the eigenvalue \theta_{2j} (for j= 2,3,\ldots,n_2) of G_{n_2} with an eigenvector W such that J_{1\times n_2}W= 0.
Put \phi_3= \begin{bmatrix}0\\0\\W\\0 \end{bmatrix}, then we have
D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E))\phi_3= -(\theta_{2j} +2)\phi_3.
This shows that -(\theta_{2j} +2) are eigenvalues of D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E)) corresponding to the eigenvalues \theta_{2j}\neq k_2 of A(G_{n_2}).
Similarly, for each eigenvalue \theta_{3j} (for j= 2,3,\ldots,n_3) of G_{n_3}, we get an eigenvalue -(\theta_{3j} +2) of D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E)) with eigenvector \begin{bmatrix}0\\0\\0\\X\end{bmatrix}. Thus, we get 2n_1-2+m_1-n_1+n_2-1+n_3-1= n_1+m_1+n_2+n_3-4 eigenvalues of D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E)). All the other eigenvalues of D(G_{n_1}^C \triangleright (G_{n_2}^V\cup G_{n_3}^E)) are the eigenvalues of the equitable quotient matrix
M= \begin{bmatrix} n_1-1+k_1& 2m_1-k_1& n_2& 2n_3\\2n_1-2& 2m_1-2& 2n_2& n_3\\n_1& 2m_1& 2n_2-k_2-2& 3n_3\\2n_1& m_1& 3n_2& 2n_3-k_3-2\end{bmatrix}.