Appendix
The proof of Proposition 4
We have
$$\begin{aligned} X=A\otimes I+I\otimes B= \left( \begin{array}{cccc}B&{}a_1I&{}0&{}0\\ a_2I&{}B &{}0&{}0\\ 0&{}0&{}B&{}a_3I\\ 0&{}0&{}a_4I&{}B\end{array}\right) , \end{aligned}$$
and
$$\begin{aligned} XX^\dag= & {} \left( \begin{array}{cccc}B&{}a_1I&{}0&{}0\\ a_2I&{}B &{}0&{}0\\ 0&{}0&{}B&{}a_3I\\ 0&{}0&{}a_4I&{}B\end{array}\right) \left( \begin{array}{cccc}B^\dag &{}a_2^*I&{}0&{}0\\ a_1^*I&{}B^\dag &{}0&{}0\\ 0&{}0&{}B^\dag &{}a_4^*I\\ 0&{}0&{}a_3^*I&{}B^\dag \end{array}\right) \\= & {} \left( \begin{array}{cccc}BB^\dag +|a_1|^2I&{}a_2^*B+a_1B^\dag &{}0&{}0\\ a_2B^\dag +a_1^*B&{}BB^\dag +|a_2|^2I&{}0&{}0\\ 0&{}0&{}BB^\dag +|a_3|^2I &{} a_3B^\dag +a_4^*B\\ 0&{}0&{}a_4B^t+a_3^*B&{}|a_4|^2I+BB^\dag \end{array}\right) \\= & {} \left( \begin{array}{cc}C&{}0\\ 0&{}D\end{array}\right) =C\oplus D. \end{aligned}$$
where
$$\begin{aligned}&C=\left( \begin{array}{cc}BB^\dag +|a_1|^2&{}a_2^*B+a_1B^\dag \\ a_2B^\dag +a_1^*B&{}|a_2|^2I+BB^\dag \end{array}\right) \nonumber \\&\quad =\left( \begin{array}{cccccccc}|b_1|^2+|a_1|^2&{}0&{}0&{}0&{}0&{}a_2^*b_1&{}0&{}a_1b_4^*\\ 0&{}|b_2|^2+|a_1|^2&{}0&{}0&{}a_1b_1^*&{}0&{}a_2^*b_2&{}0\\ 0&{}0&{}|b_3|^2+|a_1|^2&{}0&{}0&{}a_1b_2^*&{}0&{}a_2^*b_3\\ 0&{}0&{}0&{}|b_4|^2+|a_1|^2&{}a_2^*b_4&{}0&{}a_1b_3^*&{}0\\ 0&{}a_1^*b_1&{}0&{}a_2b_4^*&{}|b_1|^2 +|a_2|^2&{}0&{}0&{}0\\ a_2b_1^*&{}0&{}a_1^*b_2&{}0&{}0&{}|b_2|^2+|a_2|^2&{}0&{}0\\ 0&{}a_2b_2^*&{}0&{}a_1^*b_3&{}0&{}0&{}|b_3|^2 +|a_2|^2&{}\\ a_1^*b_4&{}0&{}a_2b_3^*&{}0&{}0&{}0&{}0&{}|b_4|^2 + |a_2|^2\end{array}\right) \nonumber \\&\quad \sim \left( \begin{array}{cccccccc}|b_1|^2+|a_1|^2&{}0&{}a_2^*b_1&{}a_1b_4^*&{}0&{}0&{}0&{}0\\ 0&{}|b_3|^2+|a_1|^2&{}a_1b_2^*&{}a_2^*b_3&{}0&{}0&{}0&{}0\\ a_2b_1^*&{}a_1^*b_2&{}|b_2|^2+|a_2|^2&{}0&{}0&{}0&{}0&{}0\\ a_1^*b_4&{}a_2b_3^*&{}0&{}|b_4|^2+|a_2|^2&{}0&{}0&{}0&{}0\\ 0&{}0&{}0&{}0&{}|b_2|^2+|a_1|^2&{}0&{}a_1b_1^*&{}a_2^*b_2\\ 0&{}0&{}0&{}0&{}0&{}|b_4|^2+|a_1|^2&{}a_2^*b_4&{}a_1b_3^*\\ 0&{}0&{}0&{}0&{}a_1^*b_1&{}a_2b_4^*&{}|b_1|^2+|a_2|^2&{}0\\ 0&{}0&{}0&{}0&{}a_2b_2^*&{}a_1^*b_3&{}0&{}|b_3|^2+|a_2|^2\end{array}\right) \nonumber \\&\quad =C_1\oplus C_2, \end{aligned}$$
(14)
and
$$\begin{aligned}&D=\left( \begin{array}{cc}BB^\dag +|a_3|^2&{}a_4^*B+a_3B^\dag \\ a_4B^\dag +a_3^*B&{}|a_4|^2I+BB^\dag \end{array}\right) \nonumber \\&\quad \sim \left( \begin{array}{cccccccc}|b_1|^2+|a_3|^2&{}0&{}a_4^*b_1&{}a_3b_4^*&{}0&{}0&{}0&{}0\\ 0&{}|b_3|^2+|a_3|^2&{}a_3b_2^*&{}a_4^*b_3&{}0&{}0&{}0&{}0\\ a_4b_1^*&{}a_3^*b_2&{}|b_2|^2+|a_4|^2&{}0&{}0&{}0&{}0&{}0\\ a_3^*b_4&{}a_4b_3^*&{}0&{}|b_4|^2+|a_4|^2&{}0&{}0&{}0&{}0\\ 0&{}0&{}0&{}0&{}|b_2|^2+|a_3|^2&{}0&{}a_3b_1^*&{}a_4^*b_2\\ 0&{}0&{}0&{}0&{}0&{}|b_4|^2+|a_3|^2&{}a_4^*b_4&{}a_3b_3^*\\ 0&{}0&{}0&{}0&{}a_3^*b_1&{}a_4b_4^*&{}|b_1|^2+|a_4|^2&{}0\\ 0&{}0&{}0&{}0&{}a_4b_2^*&{}a_3^*b_3&{}0&{}|b_3|^2+|a_4|^2\end{array}\right) \nonumber \\&\quad =D_1\oplus D_2. \end{aligned}$$
(15)
Then, we have
$$\begin{aligned} XX^\dag =C_1\oplus C_2\oplus D_1 \oplus D_2, \end{aligned}$$
(16)
and
$$\begin{aligned} \mathrm{Tr}\,C_1= & {} \mathrm{Tr}\,C_2=\sum _{j=1}^4|b_j|^2+2(|a_1|^2+|a_2|^2), \nonumber \\ \mathrm{Tr}\,D_1= & {} \mathrm{Tr}\,D_2=\sum _{j=1}^4|b_j|^2+2(|a_3|^2+|a_4|^2). \end{aligned}$$
(17)
Let \(\lambda \) and \(\mu \) be two arbitrary eigenvalues of \(XX^\dag \). Using Lemma 3, proving Conjecture 1 is equivalent to proving \(\lambda +\mu \le \frac{1}{2}\). There are three cases for \(\lambda \) and \(\mu \).
Case 1. \(\lambda \) is the eigenvalue of some \(C_i\) and \(\mu \) is the eigenvalue of some \(D_j\) for any \(i,j =1,2\).
Without loss of generality, we assume that \(\lambda \) is the eigenvalue of \(C_1\) and \(\mu \) is the eigenvalue of \(D_1\). Equation (14) implies that \(\lambda \le \mathrm{Tr}\,C_1= \sum _{j=1}^4|b_j|^2+2(|a_1|^2+|a_2|^2)\). Equation (15) implies that \(\mu \le \mathrm{Tr}\,D_1= \sum _{j=1}^4|b_j|^2+2(|a_3|^2+|a_4|^2)\). Then, \( \lambda +\mu \le 2\sum _{j=1}^4(|b_j|^2+|a_j|^2)\), Eq. (2) shows that \(\lambda +\mu \le \frac{1}{2}\).
Case 2. \(\lambda \) and \(\mu \) are the eigenvalues of same \(C_i\) or \(D_j\) .
Assume \(\lambda \) and \(\mu \) are the eigenvalue of \(C_1\), Eq. (14) implies that \(\lambda +\mu \le \mathrm{Tr}\,C_1= \sum _{j=1}^4|b_j|^2+2(|a_1|^2+|a_2|^2)\le 2[\sum _{j=1}^4(|b_j|^2+|a_j|^2)]=\frac{1}{2}\) .
Case 3. \(\lambda \) and \(\mu \) are the eigenvalues of \(C_1\) and \(C_2\) or \(D_1\) and \(D_2\).
Without loss of generality, we can assume that \(\lambda \) is the maximum eigenvalue of \(C_1\) and \(\mu \) is the maximum eigenvalue of \(C_2\). Let \(x=(x_1, x_2, x_3, x_4), y=(y_1, y_2, y_3, y_4)\) and \(\Vert x\Vert =\Vert y\Vert =1\), then
$$\begin{aligned} \begin{array}{ll} xC_1x^\dag &{}=(x_1, x_2, x_3, x_4)\left( \begin{array}{cccc}|b_1|^2+|a_1|^2&{}0&{}a_2^*b_1&{}a_1b_4^*\\ 0&{}|b_3|^2+|a_1|^2&{}a_1b_2^*&{}a_2^*b_3\\ a_2b_1^*&{}a_1^*b_2&{}|b_2|^2+|a_2|^2&{}0\\ a_1^*b_4&{}a_2b_3^*&{}0&{}|b_4|^2+|a_2|^2\end{array}\right) \left( \begin{array}{c}x_1^*\\ x_2^*\\ x_3^*\\ x_4^*\end{array}\right) \\ &{}=(|b_1|^2+|a_1|^2)|x_1|^2+(|b_3|^2+|a_1|^2)|x_2|^2\\ &{}\qquad +(|b_2|^2+|a_2|^2)|x_3|^2+(|b_4|^2+|a_2|^2)|x_4|^2\\ &{}\quad \quad +a_2b_1^*x_1^*x_3+a_1^*b_4x_1^*x_4+a_1^*b_2x_2^*x_3+a_2^*b_3x_2^*x_4\\ &{}\quad \quad +a_2^*b_1x_1x_3^*+a_1b_4^*x_1x_4^*+a_1b_2^*x_2x_3^*+a_2b_3^*x_2^*x_4\\ &{}=|b_1x_1+a_2x_3|^2+|a_1x_1+b_4x_4|^2+|a_1x_2+b_2x_3|^2+|b_3x_4+a_2x_2|^2\\ &{}\le (|b_1|^2+|a_2|^2)(|x_1|^2+|x_3|^2)+(|a_1|^2+|b_2|^2)(|x_2|^2+|x_3|^2)\\ &{}\quad +(|a_1|^2+|b_4|^2)(|x_1|^2+|x_4|^2)+(|b_3|^2+|a_2|^2)(|x_2|^2+|x_4|^2)\\ &{}=(|b_1|^2+|a_2|^2+|a_1|^2+|b_4|^2)|x_1|^2+(|a_1|^2+|b_2|^2+|a_2|^2+|b_3|^2)|x_2|^2\\ &{}\quad +(|b_1|^2+|a_2|^2+|a_1|^2+|b_2|^2)|x_3|^2+(|a_1|^2+|b_4|^2+|a_2|^2+|b_3|^2)|x_4|^2\\ &{}=k_1|x_1|^2+k_2|x_2|^2+k_3|x_3|^2+k_4|x_4|^2, \end{array}\nonumber \\ \end{aligned}$$
(18)
where \(k_1=|b_1|^2+|a_2|^2+|a_1|^2+|b_4|^2, k_2=|a_1|^2+|b_2|^2+|a_2|^2+|b_3|^2, k_3=|b_1|^2+|a_2|^2+|a_1|^2+|b_2|^2, k_4=|a_1|^2+|b_4|^2+|a_2|^2+|b_3|^2, \) we have known that \(|x_1|^2+|x_2|^2+|x_3|^2+|x_4|^2=1\), then
$$\begin{aligned} xC_1x^\dag \le k_1x_1^2+k_2x_2^2+k_3x_3^2+k_4x_4^2\le \max \limits _{i=1,\cdots , 4}\{k_1, k_2, k_3, k_4\}. \end{aligned}$$
(19)
Equation (2) shows that \(k_i\le \sum _{i=1}^4 (|a_i|^2+|b_i|^2) = \frac{1}{4}\) for \( i=1, 2, 3, 4\), this implies that the maximal eigenvalue \(\lambda \) of matrix \(C_1\) satisfies
$$\begin{aligned} \lambda \le \frac{1}{4}. \end{aligned}$$
(20)
Similarly, we have
$$\begin{aligned} \begin{array}{ll} yC_2y^\dag &{}=(y_1, y_2, y_3, y_4)\left( \begin{array}{cccc}b_2^2+a_1^2&{}0&{}a_1b_1&{}a_2b_2\\ 0&{}b_4^2+a_1^2&{}a_2b_4&{}a_1b_3\\ a_1b_1&{}a_2b_4&{}b_1^2+a_2^2&{}0\\ a_2b_2&{}a_1b_3&{}0&{}b_3^2+a_2^2\end{array}\right) \left( \begin{array}{c}y_1^*\\ y_2^*\\ y_3^*\\ y_4^*\end{array}\right) \\ &{}\le (|b_2|^2+|a_2|^2+|a_1|^2+|b_1|^2)|y_1|^2+(|a_2|^2+|b_4|^2+|a_1|^2+|b_3|^2)|y_2|^2\\ &{}\quad +(|a_1|^2+|b_1|^2+|a_2|^2+|b_4|^2)|y_3|^2+(|b_2|^2+|a_2|^2+|a_1|^2+|b_3|^2)|y_4|^2\\ &{}=l_1|y_1|^2+l_2|y_2|^2+l_3|y_3|^2+l_4|y_4|^2.\end{array}\nonumber \\ \end{aligned}$$
(21)
Since \(|y_1|^2+|y_2|^2+|y_3|^2+|y_4|^2=1, \) we have
$$\begin{aligned} yC_2y^\dag \le \max \limits _{i=1,\cdots , 4}\{l_1, l_2, l_3, l_4\}. \end{aligned}$$
(22)
From Eq. (2), we can get that \(l_i\le \sum _{i=1}^4 (|a_i|^2+|b_i|^2) = \frac{1}{4}\) for \( i=1, 2, 3, 4\), this implies that the maximal eigenvalue \(\mu \) of matrix \(C_2\) satisfying
$$\begin{aligned} \mu \le \frac{1}{4}. \end{aligned}$$
(23)
Equations (20) and (23) imply that \(\lambda +\mu \le \frac{1}{2}.\)
This completes the proof.
The proof of Proposition 5
$$\begin{aligned} X=A\otimes I+I\otimes B= & {} \left( \begin{array}{cccc}0&{}a_1I &{}0&{}0\\ 0&{}0&{}a_2I &{}0\\ 0&{}0&{}0&{}a_3I\\ a_4I&{}0 &{} 0&{}0\end{array}\right) +\left( \begin{array}{cccc}B&{}0&{}0&{}0\\ 0&{}B&{}0&{}0\\ 0&{}0&{}B&{}0\\ 0&{}0&{}0&{}B\end{array}\right) \\= & {} \left( \begin{array}{cccc}B&{}a_1I&{}0&{}0\\ 0&{}B &{} a_2I&{}0\\ 0&{}0&{}B&{}a_3I\\ a_4I &{}0&{}0&{}B\end{array}\right) , \\ XX^\dag= & {} \left( \begin{array}{cccc}B&{}a_1I&{}0&{}0\\ 0&{}B &{} a_2I&{}0\\ 0&{}0&{}B&{}a_3I\\ a_4I &{}0&{}0&{}B\end{array}\right) \left( \begin{array}{cccc}B^\dag &{}0&{}0&{}a_4^*I\\ a_1^*I&{}B^\dag &{}0&{}0\\ 0&{}a_2^*I &{}B^\dag &{}0\\ 0&{}0&{}a_3^*I&{}B^\dag \end{array}\right) \\= & {} \left( \begin{array}{cccc}BB^\dag +|a_1|^2I&{}a_1B^\dag &{}0&{}a_4^*B\\ a_1^*B&{}BB^\dag +|a_2|^2I&{}a_2B^\dag &{}0\\ 0&{}a_2^*B&{}BB^\dag +|a_3|^2I &{} a_3B^\dag \\ a_4B^\dag &{}0&{}a_3^*B&{}|a_4|^2I+BB^\dag \end{array}\right) \\= & {} \left( \begin{array}{ll}Y_1&{}Y_2\\ Y_2^\dag &{}Y_3\end{array}\right) , \end{aligned}$$
where
$$\begin{aligned}&Y_1=\left( \begin{array}{cc}BB^\dag +|a_1|^2I&{}a_1B^\dag \\ a_1^*B&{}|a_2|^2I+BB^\dag \end{array}\right) \\&\quad =\left( \begin{array}{cccccccc}|b_1|^2+|a_1|^2&{}0&{}0&{}0&{}0&{}0&{}0&{}a_1b_4^*\\ 0&{}|b_2|^2+|a_1|^2&{}0&{}0&{}a_1b_1^*&{}0&{}0&{}0\\ 0&{}0&{}|b_3|^2+|a_1|^2&{}0&{}0&{}a_1b_2^*&{}0&{}0\\ 0&{}0&{}0&{}|b_4|^2+|a_1|^2&{}0&{}0&{}a_1b_3^*&{}0\\ 0&{}a_1^*b_1&{}0&{}0&{}|b_1|^2 +|a_2|^2&{}0&{}0&{}0\\ 0&{}0&{}a_1^*b_2&{}0&{}0&{}|b_2|^2+|a_2|^2&{}0&{}0\\ 0&{}0&{}0&{}a_1^*b_3&{}0&{}0&{}|b_3|^2 +|a_2|^2&{}0\\ a_1^*b_4&{}0&{}0&{}0&{}0&{}0&{}0&{}|b_4|^2 + |a_2|^2\end{array}\right) \\&\quad =P \left( \begin{array}{cccccccc}|b_1|^2+|a_1|^2&{}a_1b_4^*&{}0&{}0&{}0&{}0&{}0&{}0\\ a_1^*b_4&{}|b_4|^2+|a_2|^2&{}0&{}0&{}0&{}0&{}0&{}0\\ 0&{}0&{}|b_2|^2+|a_1|^2&{}a_1b_1^*&{}0&{}0&{}0&{}0\\ 0&{}0&{}a_1^*b_1&{}|b_1|^2+|a_2|^2&{}0&{}0&{}0&{}0\\ 0&{}0&{}0&{}0&{}|b_3|^2+|a_1|^2&{}a_1b_2^*&{}0&{}0\\ 0&{}0&{}0&{}0&{}a_1^*b_2&{}|b_2|^2+|a_2|^2&{}0&{}0\\ 0&{}0&{}0&{}0&{}0&{}0&{}|b_4|^2+|a_1|^2&{}a_1b_3^*\\ 0&{}0&{}0&{}0&{}0&{}0&{}a_1^*b_3&{}|b_3|^2+|a_2|^2\end{array}\right) P^\dag , \end{aligned}$$
where P is a permutation. We have
$$\begin{aligned}&Y_2=\left( \begin{array}{cc}0&{}a_4^*B\\ a_2B^\dag &{}0\end{array}\right) =\left( \begin{array}{cccccccc}0&{}0&{}0&{}0&{}0&{}a_4^*b_1&{}0&{}0\\ 0&{}0&{}0&{}0&{}0&{}0&{}a_4^*b_2&{}0\\ 0&{}0&{}0&{}0&{}0&{}0&{}0&{}a_4^*b_3\\ 0&{}0&{}0&{}0&{}a_4^*b_4&{}0&{}0&{}0\\ 0&{}0&{}0&{}a_2b_4^*&{}0&{}0&{}0&{}0\\ a_2b_1^*&{}0&{}0&{}0&{}0&{}0&{}0&{}0\\ 0&{}a_2b_2^*&{}0&{}0&{}0&{}0&{}0&{}0\\ 0&{}0&{}a_2b_3^*&{}0&{}0&{}0&{}0&{}0\end{array}\right) \\&\quad =P\left( \begin{array}{cccccccc}0&{}0&{}0&{}0&{}0&{}a_4^*b_1&{}0&{}0\\ 0&{}0&{}0&{}0&{}a_2b_3^*&{}0&{}0&{}0\\ 0&{}0&{}0&{}0&{}0&{}0&{}0&{}a_4^*b_2\\ 0&{}0&{}0&{}0&{}0&{}0&{}a_2b_4^*&{}0\\ 0&{}a_4^*b_3&{}0&{}0&{}0&{}0&{}0&{}0\\ a_2b_1^*&{}0&{}0&{}0&{}0&{}0&{}0&{}0\\ 0&{}0&{}0&{}a_4^*b_4&{}0&{}0&{}0&{}0\\ 0&{}0&{}a_2b_2^*&{}0&{}0&{}0&{}0&{}0\end{array}\right) P^t, \\&Y_3=\left( \begin{array}{cc}BB^\dag +|a_3|^2I&{}a_3B^\dag \\ a_3^*B&{}|a_4|^2I+BB^\dag \end{array}\right) \\&\quad =P \left( \begin{array}{cccccccc}|b_1|^2+|a_3|^2&{}a_3b_4^*&{}0&{}0&{}0&{}0&{}0&{}0\\ a_3^*b_4&{}|b_4|^2+|a_4|^2&{}0&{}0&{}0&{}0&{}0&{}0\\ 0&{}0&{}|b_2|^2+|a_3|^2&{}a_3b_1^*&{}0&{}0&{}0&{}0\\ 0&{}0&{}a_3^*b_1&{}|b_1|^2+|a_4|^2&{}0&{}0&{}0&{}0\\ 0&{}0&{}0&{}0&{}|b_3|^2+|a_3|^2&{}a_3b_2^*&{}0&{}0\\ 0&{}0&{}0&{}0&{}a_3^*b_2&{}|b_2|^2+|a_4|^2&{}0&{}0\\ 0&{}0&{}0&{}0&{}0&{}0&{}|b_4|^2+|a_3|^2&{}a_3b_3^*\\ 0&{}0&{}0&{}0&{}0&{}0&{}a_3^*b_3&{}|b_3|^2+|a_4|^2\end{array}\right) P^t. \end{aligned}$$
Here we can see that
$$\begin{aligned}&XX^\dag \sim \left( \begin{array}{cccccccc}N_1&{}0&{}0&{}0&{}0&{}0&{}N_2&{}0\\ 0&{}N_3&{}0&{}0&{}0&{}0&{}0&{}N_4\\ 0&{}0&{}N_5&{}0&{}N_6&{}0&{}0&{}0\\ 0&{}0&{}0&{}N_7&{}0&{}N_8&{}0&{}0\\ 0&{}0&{}N_6^\dag &{}0&{}N_9&{}0&{}0&{}0\\ 0&{}0&{}0&{}N_8^\dag &{}0&{}N_{10}&{}0&{}0\\ N_2^\dag &{}0&{}0&{}0&{}0&{}0&{}N_{11}&{}0\\ 0&{}N_4^\dag &{}0&{}0&{}0&{}0&{}0&{}N_{12}\end{array}\right) \nonumber \\&\quad \sim \left( \begin{array}{cccccccc}N_1&{}N_2&{}0&{}0&{}0&{}0&{}0&{}0\\ N_2^\dag &{}N_{11}&{}0&{}0&{}0&{}0&{}0&{}0\\ 0&{}0&{}N_3&{}N_4&{}0&{}0&{}0&{}0\\ 0&{}0&{}N_4^\dag &{}N_{12}&{}0&{}0&{}0&{}0\\ 0&{}0&{}0&{}0&{}N_5&{}N_6&{}0&{}0\\ 0&{}0&{}0&{}0&{}N_6^\dag &{}N_9&{}0&{}0\\ 0&{}0&{}0&{}0&{}0&{}0&{}N_7&{}N_8\\ 0&{}0&{}0&{}0&{}0&{}0&{}N_8^\dag &{}N_{10} \end{array}\right) =W, \end{aligned}$$
(24)
where \(N_i, i=1, 2,\dots , 12\) are \(2\times 2\) matrices. Then
$$\begin{aligned} XX^\dag \sim W=W_1\oplus W_2\oplus W_3 \oplus W_4, \end{aligned}$$
(25)
where
$$\begin{aligned} W_1=\left( \begin{array}{cc}N_1&{}N_2\\ N_2^\dag &{}N_{11}\end{array}\right) =\left( \begin{array}{cccc}|b_1|^2+|a_1|^2&{}a_1b_4^*&{}0&{}a_4^*b_1\\ a_1^*b_4&{}|b_4|^2+|a_2|^2&{}a_2b_3^*&{}0\\ 0&{}a_2^*b_3&{}|b_3|^2+|a_3|^2&{}a_3b_2^*\\ a_4b_1^*&{}0&{}a_3^*b_2&{}|b_2|^2+|a_4|^2\end{array}\right) , \end{aligned}$$
(26)
$$\begin{aligned} W_2=\left( \begin{array}{cc}N_3&{}N_4\\ N_4^t&{}N_{12}\end{array}\right) =\left( \begin{array}{cccc}|b_2|^2+|a_1|^2&{}a_1b_1^*&{}0&{}a_4^*b_2\\ a_1^*b_1&{}|b_1|^2+|a_2|^2&{}a_2b_4^*&{}0\\ 0&{}a_2^*b_4&{}|b_4|^2+|a_3|^2&{}a_3^*b_3\\ a_4b_2^*&{}0&{}a_3b_3^*&{}|b_3|^2+|a_4|^2\end{array}\right) , \end{aligned}$$
(27)
$$\begin{aligned} W_3=\left( \begin{array}{cc}N_5&{}N_6\\ N_6^t&{}N_{9}\end{array}\right) =\left( \begin{array}{cccc}|b_3|^2+|a_1|^2&{}a_1b_2^*&{}0&{}a_4^*b_3\\ a_1^*b_2&{}|b_2|^2+|a_2|^2&{}a_2b_1^*&{}0\\ 0&{}a_2^*b_1&{}|b_1|^2+|a_3|^2&{}a_3b_4^*\\ a_4b_3^*&{}0&{}a_3^*b_4&{}|b_4|^2+|a_4|^2\end{array}\right) , \end{aligned}$$
(28)
$$\begin{aligned} W_4=\left( \begin{array}{cc}N_7&{}N_8\\ N_8^t&{}N_{10}\end{array}\right) =\left( \begin{array}{cccc}|b_4|^2+|a_1|^2&{}a_1b_3^*&{}0&{}a_4^*b_4\\ a_1^*b_3&{}|b_3|^2+|a_2|^2&{}a_2b_2^*&{}0\\ 0&{}a_2^*b_2&{}|b_2|^2+|a_3|^2&{}a_3b_1^*\\ a_4b_4^*&{}0&{}a_3^*b_1&{}|b_1|^2+|a_4|^2\end{array}\right) ,\end{aligned}$$
(29)
and
$$\begin{aligned} \mathrm{Tr}\,W_i=\sum _{j=1}^4(|b_j|^2+|a_j|^2)=\frac{1}{4}, i=1,2,3,4. \end{aligned}$$
(30)
Let \(\lambda \) and \(\mu \) be two arbitrary eigenvalues of \(XX^\dag \). Then, proving Conjecture 1 is equivalent to proving \(\lambda +\mu \le \frac{1}{2}\). There are two cases for \(\lambda \) and \(\mu \).
Case 1. \(\lambda \) and \(\mu \) are the eigenvalues of different \(W_i\).
Without loss of generality, we can assume that \(\lambda \) is the maximum eigenvalue of \(W_i\), and \(\mu \) is the maximum eigenvalue of \(W_j\), \(1\le i, j\le 4, i\ne j\). \(\lambda +\mu \le \mathrm{Tr}\,W_i+\mathrm{Tr}\,W_j= 2[\sum _{j=1}^4(|b_j|^2+|a_j|^2)]=\frac{1}{2}\) .
Case 2. \(\lambda \) and \(\mu \) are the eigenvalues of same \(W_i, i=1, 2, 3, 4\). Equation (30) implies that \(\lambda +\mu \le \mathrm{Tr}\,W_i= \sum _{j=1}^4(|b_j|^2+|a_j|^2)=\frac{1}{4}\).
This completes the proof.
The proof of Proposition 6
$$\begin{aligned} X= & {} A\otimes I+I\otimes B=\left( \begin{array}{cccc}a_1I &{}0&{}0 &{}0\\ 0&{}-a_1I&{}0&{}0\\ 0&{}0&{}0&{}a_2I\\ 0&{}0 &{} a_3I&{}0\end{array}\right) +\left( \begin{array}{cccc}B&{}0&{}0&{}0\\ 0&{}B&{}0&{}0\\ 0&{}0&{}B&{}0\\ 0&{}0&{}0&{}B\end{array}\right) \\= & {} \left( \begin{array}{cccc}a_1I+B&{}0&{}0&{}0\\ 0&{}-a_1I+B &{}0&{}0\\ 0&{}0&{}B&{}a_2I\\ 0&{}0&{}a_3I&{}B\end{array}\right) . \end{aligned}$$
So we have
$$\begin{aligned} \begin{array}{lll} XX^\dag &{}= &{} \left( \begin{array}{cccc}a_1I+B&{}0&{}0&{}0\\ 0&{}-a_1I+B &{}0&{}0\\ 0&{}0&{}B&{}a_2I\\ 0&{}0&{}a_3I&{}B\end{array}\right) \left( \begin{array}{cccc}a_1^*I+B^\dag &{}0&{}0&{}0\\ 0&{}-a_1^*I+B^\dag &{}0&{}0\\ 0&{}0&{}B^\dag &{}a_3^*I\\ 0&{}0&{}a_2^*I&{}B^\dag \end{array}\right) \\ &{}=&{}\left( \begin{array}{cccc}BB^\dag +|a_1|^2I+a_1B^\dag +a_1^*B&{}0&{}0&{}0\\ 0&{}BB^\dag -a_1^*B-a_1B^\dag +|a_1|^2I&{}0&{}0\\ 0&{}0&{}BB^\dag +|a_2|^2I &{} a_2B^\dag +a_3^*B\\ 0&{}0&{}a_3B^\dag +a_2^*B&{}|a_3|^2I+BB^\dag \end{array}\right) \\ &{}=&{}\left( \begin{array}{cc}C&{}0\\ 0&{}D\end{array}\right) =C\oplus D,\end{array} \end{aligned}$$
where
$$\begin{aligned}&C=\left( \begin{array}{cc}BB^\dag +|a_1|^2I+a_1B^\dag +a_1^*B&{}0\\ 0&{}BB^\dag -a_1^*B-a_1B^\dag +|a_1|^2I \end{array}\right) \nonumber \\&=C_1\oplus C_2\oplus C_3\oplus C_4, \nonumber \\&C_1= \left( \begin{array}{cc} |b_1+a_1|^2&{}0\\ 0&{}|b_1-a_1|^2 \end{array}\right) , \nonumber \\&C_2= \left( \begin{array}{cc} |b_2|^2+|a_1|^2&{}a_1^*b_2+a_1b_3^*\\ a_1^*b_3+a_1b_2^*&{}|b_3|^2+|a_1|^2\end{array}\right) , \nonumber \\&C_3= \left( \begin{array}{cc} |b_1-a_1|^2&{}0\\ 0&{}|b_1+a_1|^2\end{array}\right) , \nonumber \\&C_4=\left( \begin{array}{cc} |b_2|^2 +|a_1|^2&{}-a_1^*b_2-a_1b_3^*\\ -a_1b_2^*-a_1^*b_3&{}|b_3|^2 + |a_1|^2\end{array}\right) , \end{aligned}$$
(31)
and
$$\begin{aligned}&D=\left( \begin{array}{cc}BB^\dag +|a_2|^2I&{}a_3^*B+a_2B^\dag \\ a_3B^\dag +a_2^*B&{}|a_3|^2I+BB^\dag \end{array}\right) \nonumber \\&\quad =\left( \begin{array}{cccccccc}|b_1|^2+|a_2|^2&{}0&{}0&{}0&{}a_3^*b_1+a_2b_1^*&{}0&{}0&{}0\\ 0&{}|b_1|^2+|a_2|^2&{}0&{}0&{}0&{}-b_1^*a_2-b_1a_3^*&{}0&{}0\\ 0&{}0&{}|b_2|^2+|a_2|^2&{}0&{}0&{}0&{}0&{}a_3^*b_2+a_2b_3^*\\ 0&{}0&{}0&{}|b_3|^2+|a_2|^2&{}0&{}0&{}a_3^*b_3+a_2b_2^*&{}0\\ a_3b_1^*+a_2^*b_1&{}0&{}0&{}0&{}|b_1|^2 +|a_3|^2&{}0&{}0&{}0\\ 0&{}-b_1a_2^*-b_1^*a_3&{}0&{}0&{}0&{}|a_3|^2+|b_1|^2&{}0&{}0\\ 0&{}0&{}0&{}a_3b_3^*+a_2^*b_2&{}0&{}0&{}|b_2|^2+|a_3|^2&{}0\\ 0&{}0&{}a_3b_2^*+a_2^*b_3&{}0&{}0&{}0&{}0&{}|b_3|^2 + |a_3|^2\end{array}\right) \nonumber \\&\quad \sim \left( \begin{array}{cccccccc}|b_1|^2+|a_2|^2&{}a_3^*b_1+a_2b_1^*&{}0&{}0&{}0&{}0&{}0&{}0\\ a_3b_1^*+a_2^*b_1&{}|b_1|^2+|a_3|^2&{}0&{}0&{}0&{}0&{}0&{}0\\ 0&{}0&{}|b_1|^2+|a_2|^2&{}-b_1^*a_2-b_1a_3^*&{}0&{}0&{}0&{}0\\ 0&{}0&{}-b_1a_2^*-b_1^*a_3&{}|b_1|^2+|a_3|^2&{}0&{}0&{}0&{}0\\ 0&{}0&{}0&{}0&{}|b_2|^2+|a_2|^2&{}a_3b_2^*+a_2^*b_3&{}0&{}0\\ 0&{}0&{}0&{}0&{}a_3^*b_2+a_2b_3^*&{}|b_3|^2+|a_3|^2&{}0&{}0\\ 0&{}0&{}0&{}0&{}0&{}0&{}|b_3|^2+|a_2|^2&{}a_3^*b_3+a_2b_2^*\\ 0&{}0&{}0&{}0&{}0&{}0&{}a_3b_3^*+a_2^*b_2&{}|b_2|^2+|a_3|^2\end{array}\right) \nonumber \\&\quad =D_1\oplus D_2\oplus D_3\oplus D_4. \end{aligned}$$
(32)
One can formulate the characteristic polynomial of C as follows.
$$\begin{aligned} \det (\lambda I-C)=f_1^2(\lambda )\cdot f_2^2(\lambda ), \end{aligned}$$
(33)
where
$$\begin{aligned} \begin{array}{lll} f_1(\lambda )&{}=&{}(|b_1|^2+|a_2|^2+2Re a_1^*b_1-\lambda )(|b_1|^2+|a_2|^2-2Re a_1^*b_1-\lambda ),\\ f_2(\lambda )&{}=&{}(|b_2|^2+|a_1|^2-\lambda )(|a_1|^2+|b_3|^2-\lambda )-|a_1^*b_2+a_1b_3^*|^2,\\ \end{array} \end{aligned}$$
(34)
and Rez denotes the real part of the complex number z.
We first claim the larger root of \(f_i(\lambda )=0, \forall i=1,2,3,4\), is not greater than \(\frac{1}{4}\).Take \(f_1(\lambda )\) as an example. It is clear that the sum of the two roots of \(f_1(\lambda )=0\) is \(|a_1|^2+|a_2|^2+2|b_1|^2\). Since \(XX^t=C\oplus D\) is a semipositive definite matrix. All the eigenvalues of C and D are non-negative. This implies that all the roots of \(f_i(\lambda )=0, i=1, 2\) are nonnegative. So we have the larger root of \(f_1(\lambda )=0\) is not greater than the sum of two roots, i.e., \(|a_1|^2+|a_2|^2+2|b_1|^2\). Recall the \(\sum _{i=1}^3(|a_i|^2+|b_i|^2)+|a_1|^2+|b_1|^2=\frac{1}{4}.\) Therefore, we conclude that the larger root of \(f_1(\lambda )=0\) is not larger than \(\frac{1}{4}.\) One can draw the same conclusion of \(f_2(\lambda )\). Then, our claim holds.
This implies the largest eigenvalue of C is not greater than \(\frac{1}{4}.\)
The characteristic polynomial of D can be expressed as follows:
$$\begin{aligned} \det (\lambda I-D)=g_1 (\mu )\cdot g_2(\mu )\cdot g_3(\mu )\cdot g_3(\mu ), \end{aligned}$$
(35)
where
$$\begin{aligned} \begin{array}{lll} g_1(\mu )&{}=&{}(|b_1|^2+|a_2|^2-\mu )(|a_3|^2+|b_1|^2-\mu )-|a_3^*b_1+a_2b_1^*|^2,\\ g_2(\mu )&{}=&{}(|b_1|^2+|a_2|^2-\mu )(|a_3|^2+|b_1|^2-\mu )-|b_1^*a_2+b_1a_3^*|^2,\\ g_3(\mu )&{}=&{}(|b_2|^2+|a_2|^2-\mu )(|a_3|^2+|b_3|^2-\mu )-|a_3b_2^*+a_2^*b_3|^2,\\ g_4(\mu )&{}=&{}(|b_3|^2+|a_2|^2-\mu )(|a_3|^2+|b_2|^2-\mu )-|a_3^*b_3+a_2b_2^*|^2. \end{array} \end{aligned}$$
(36)
By the same way, we conclude that the largest eigenvalue of D is not greater than \(\frac{1}{4}.\) Since \(XX^t=C\oplus D\), the sum of the largest eigenvalues of \(XX^t\) is at most \(\frac{1}{2}\). This complete the proof.
The proof of Proposition 7
$$\begin{aligned} \begin{array}{lll}X&{}=&{}A\otimes I+I\otimes B=\left( \begin{array}{cccc}a_1I &{}0&{}0 &{}0\\ 0&{}-a_1I&{}0&{}0\\ 0&{}0&{}0&{}a_2I\\ 0&{}0 &{} a_3I&{}0\end{array}\right) +\left( \begin{array}{cccc}B&{}0&{}0&{}0\\ 0&{}B&{}0&{}0\\ 0&{}0&{}B&{}0\\ 0&{}0&{}0&{}B\end{array}\right) \\ &{}=&{}\left( \begin{array}{cccc}a_1I+B&{}0&{}0&{}0\\ 0&{}-a_1I+B &{}0&{}0\\ 0&{}0&{}B&{}a_2I\\ 0&{}0&{}a_3I&{}B\end{array}\right) . \end{array} \end{aligned}$$
Hence
$$\begin{aligned} \begin{array}{lll}XX^\dag &{}= &{} \left( \begin{array}{cccc}a_1I+B&{}0&{}0&{}0\\ 0&{}-a_1I+B &{}0&{}0\\ 0&{}0&{}B&{}a_2I\\ 0&{}0&{}a_3I&{}B\end{array}\right) \left( \begin{array}{cccc}a_1^*I+B^\dag &{}0&{}0&{}0\\ 0&{}-a_1^*I+B^\dag &{}0&{}0\\ 0&{}0&{}B^\dag &{}a_3^*I\\ 0&{}0&{}a_2^*I&{}B^\dag \end{array}\right) \\ &{}=&{}\left( \begin{array}{cccc}BB^\dag +|a_1|^2I+a_1B^\dag +a_1^*B&{}0&{}0&{}0\\ 0&{}BB^\dag -a_1^*B-a_1B^\dag +|a_1|^2I&{}0&{}0\\ 0&{}0&{}BB^\dag +|a_2|^2I &{} a_2B^\dag +a_3^*B\\ 0&{}0&{}a_3B^\dag +a_2^*B&{}|a_3|^2I+BB^\dag \end{array}\right) \\ &{}=&{}\left( \begin{array}{cc}C&{}0\\ 0&{}D\end{array}\right) =C\oplus D,\end{array} \end{aligned}$$
where
$$\begin{aligned}&C=\left( \begin{array}{cc}BB^\dag +|a_1|^2I+a_1^*B+a_1B^\dag )&{}0\\ 0&{}BB^\dag +|a_1|^2I-a_1^*B-a_1B^\dag \end{array}\right) \nonumber \\&\quad =\left( \begin{array}{cccccccc}|b_1|^2+|a_1|^2&{}a_1^*b_1&{}0&{}a_1b_4^*&{}0&{}0&{}0&{}0\\ a_1b_1^*&{}|b_2|^2+|a_1|^2&{}a_1^*b_2&{}0&{}0&{}0&{}0&{}0\\ 0&{}a_1b_2^*&{}|b_3|^2+|a_1|^2&{}a_1^*b_3&{}0&{}&{}0&{}0\\ a_1^*b_4&{}0&{}a_1b_3^*&{}|b_4|^2+|a_1|^2&{}0&{}0&{}0&{}0\\ 0&{}0&{}0&{}0&{}|b_1|^2 +|a_1|^2&{}-a_1^*b_1&{}0&{}-a_1b_4^*\\ 0&{} 0&{}0&{}0&{}-a_1b_1^*&{}|b_2|^2+|a_1|^2&{}-a_1^*b_2&{}0\\ 0&{}0&{}0&{}0&{}0&{}-a_1b_2^*&{}|b_3|^2 +|a_1|^2&{}-a_1^*b_3\\ 0&{}0&{}0&{}0&{}-a_1^*b_4&{}0&{}-a_1b_3^*&{}|b_4|^2 +|a_1|^2\end{array}\right) \nonumber \\&\quad =C_1\oplus C_2, \end{aligned}$$
(37)
and
$$\begin{aligned}&D=\left( \begin{array}{cc}BB^\dag +|a_2|^2I&{}a_3^*B+a_2B^\dag \\ a_3B^\dag +a_2^*B&{}|a_3|^2I+BB^\dag \end{array}\right) \nonumber \\&\quad =\left( \begin{array}{cccccccc}|b_1|^2+|a_2|^2&{}0&{}0&{}0&{}0&{}a_3^*b_1&{}0&{}a_2b_4^*\\ 0&{}|b_2|^2+|a_2|^2&{}0&{}0&{}a_2b_1^*&{}0&{}a_3^*b_2&{}0\\ 0&{}0&{}|b_3|^2+|a_2|^2&{}0&{}0&{}a_2b_2^*&{}0&{}a_3^*b_3\\ 0&{}0&{}0&{}|b_4|^2+|a_2|^2&{}a_3^*b_4&{}0&{}a_2b_3^*&{}0\\ 0&{}a_2^*b_1&{}0&{}a_3b_4^*&{}|b_1|^2 +|a_3|^2&{}0&{}0&{}0\\ a_3b_1^*&{}0&{}a_2^*b_2&{}0&{}0&{}|b_2|^2+|a_3|^2&{}0&{}0\\ 0&{}a_3b_2^*&{}0&{}a_2^*b_3&{}0&{}0&{}|b_3|^2 +|a_3|^2&{}\\ a_2^*b_4&{}0&{}a_3b_3^*&{}0&{}0&{}0&{}0&{}|b_4|^2 + |a_3|^2\end{array}\right) \nonumber \\&\quad \sim \left( \begin{array}{cccccccc}|b_1|^2+|a_2|^2&{}0&{}a_3^*b_1&{}a_2b_4^*&{}0&{}0&{}0&{}0\\ 0&{}|b_3|^2+|a_2|^2&{}a_2b_2^*&{}a_3^*b_3&{}0&{}0&{}0&{}0\\ a_3b_1^*&{}a_2^*b_2&{}|b_2|^2+|a_3|^2&{}0&{}0&{}0&{}0&{}0\\ a_2^*b_4&{}a_3b_3^*&{}0&{}|b_4|^2+|a_3|^2&{}0&{}0&{}0&{}0\\ 0&{}0&{}0&{}0&{}|b_2|^2+|a_2|^2&{}0&{}a_2b_1^*&{}a_3^*b_2\\ 0&{}0&{}0&{}0&{}0&{}|b_4|^2+|a_2|^2&{}a_3^*b_4&{}a_2b_3^*\\ 0&{}0&{}0&{}0&{}a_2^*b_1&{}a_3b_4^*&{}|b_1|^2+|a_3|^2&{}0\\ 0&{}0&{}0&{}0&{}a_3b_2^*&{}a_2^*b_3&{}0&{}|b_3|^2+|a_3|^2\end{array}\right) \nonumber \\&\quad =D_1\oplus D_2. \end{aligned}$$
(38)
Then, we have
$$\begin{aligned} XX^t\sim C_1\oplus C_2\oplus D_1 \oplus D_2, \end{aligned}$$
(39)
and
$$\begin{aligned} \mathrm{Tr}\,C_1=\mathrm{Tr}\,C_2=\sum _{j=1}^4|b_j|^2+4|a_1|^2, \mathrm{Tr}\,D_1=\mathrm{Tr}\,D_2=\sum _{j=1}^4|b_j|^2+2(|a_2|^2+|a_3|^2).\nonumber \\ \end{aligned}$$
(40)
Let \(\lambda \) and \(\mu \) be two arbitrary eigenvalues of \(XX^t\). Then, proving Conjecture 1 is equivalent to proving \(\lambda +\mu \le \frac{1}{2}\). There are three cases for \(\lambda \) and \(\mu \).
Case 1. \(\lambda \) is the eigenvalue of some \(C_i\) and \(\mu \) is the eigenvalue of some \(D_j\) for any \(i,j =1,2\).
Without loss of generality, we assume \(\lambda \) is the eigenvalue of \(C_1\) and \(\mu \) is the eigenvalue of \(D_1\). Equations (37) and (40) imply that \(\lambda \le \mathrm{Tr}\,C_1= \sum _{j=1}^4|b_j|^2+4|a_1|^2\). Equations (38) and (40) imply that \(\mu \le \mathrm{Tr}\,D_1= \sum _{j=1}^4|b_j|^2+2(|a_2^2+|a_3|^2)\). Then \( \lambda +\mu \le 2\sum _{j=1}^3(|b_j|^2+|a_j|^2)+2|a_1|^2+2|b_4|^2\), Eq. (7) shows that \(\lambda +\mu \le \frac{1}{2}\).
Case 2. \(\lambda \) and \(\mu \) are the eigenvalues of same \(C_i\) or \(D_j\) .
Assume \(\lambda \) and \(\mu \) are the eigenvalue of \(C_1\), Eq. (40) implies that \(\lambda +\mu \le \mathrm{Tr}\,C_1= \sum _{j=1}^4|b_j|^2+4|a_1|^2\le 2[\sum _{j=1}^3(|b_j|^2+|a_j|^2)+|a_1|^2+|b_4|^2]=\frac{1}{2}\) .
Case 3. \(\lambda \) and \(\mu \) are the eigenvalues of \(C_1\) and \(C_2\) or \(D_1\) and \(D_2\).
Without loss of generality, we assume that \(\lambda \) is the maximum eigenvalue of \(C_1\) and \(\mu \) is the maximum eigenvalue of \(C_2\).
Let \(x=(x_1, x_2, x_3, x_4), y=(y_1, y_2, y_3, y_4)\) and \(\Vert x\Vert =\Vert y\Vert =1\), then
$$\begin{aligned} \begin{array}{ll} xC_1x^t&{}=(x_1, x_2, x_3, x_4)\left( \begin{array}{cccc}|b_1|^2+|a_1|^2&{}a_1^*b_1&{}0&{}a_1b_4^*\\ a_1b_1^*&{}|b_2|^2+|a_1|^2&{}a_1^*b_2&{}0\\ 0&{}a_1b_2^*&{}|b_3|^2+|a_2|^2&{}a_1^*b_3\\ a_1^*b_4&{}0&{}a_1b_3^*&{}|b_4|^2+|a_1|^2\end{array}\right) \left( \begin{array}{c}x_1^*\\ x_2^*\\ x_3^*\\ x_4^*\end{array}\right) \\ &{}=|b_1x_1+a_1x_2|^2+|a_1x_1+b_4x_4|^2+|a_1x_3+b_2x_2|^2+|b_3x_3+a_1x_4|^2\\ &{}\le (|b_1|^2+|a_1|^2)(|x_1|^2+|x_2|^2)+(|a_1|^2+|b_4|^2)(|x_1|^2+|x_4|^2)+(|a_1|^2+|b_2|^2)(|x_2|^2+|x_3|^2)\\ &{}\quad +(|b_3|^2+|a_1|^2)(|x_3|^2+|x_4|^2)\\ &{}=(|b_1|^2+2|a_1|^2+|b_4|^2)|x_1|^2+(2|a_1|^2+|b_1|^2+|b_2|^2)|x_2|^2+(|b_2|^2+|b_3|^2+2|a_1|^2)|x_3|^2\\ &{}\quad +(2|a_1|^2+|b_3|^2+|b_4|^2)|x_4|^2\end{array}.\nonumber \\ \end{aligned}$$
(41)
Set \(k_1=|b_1|^2+2|a_1|^2+|b_4|^2, k_2=2|a_1|^2+|b_1|^2+|b_2|^2, k_3=|b_2|^2+|b_3|^2+2|a_1|^2, k_4=2|a_1|^2+|b_3|^2+|b_4|^2, \) we have known that \(|x_1|^2+|x_2|^2+|x_3|^2+|x_4|^2=1\), then
$$\begin{aligned} xC_1x^t\le k_1|x_1|^2+k_2|x_2|^2+k_3|x_3|^2+k_4|x_4|^2\le \max \limits _{i=1,\cdots , 4}\{k_1, k_2, k_3, k_4\}. \end{aligned}$$
(42)
Equation (2) shows that \(k_i\le \sum _{i=1}^3 (|a_i|^2+|b_i|^2) +|a_1|^2+|b_4|^2= \frac{1}{4}\) for \( i=1, 2, 3, 4\), this implies that the maximal eigenvalue \(\lambda \) of matrix \(C_1\) satisfying
$$\begin{aligned} \lambda \le \frac{1}{4}. \end{aligned}$$
(43)
By the same way, we can show that the maximal eigenvalue \(\mu \) of matrix \(C_2\) satisfying
$$\begin{aligned} \mu \le \frac{1}{4}. \end{aligned}$$
(44)
Equations (43) and (44) imply that \(\lambda +\mu \le \frac{1}{2}.\)
This completes the proof.
The proof of Proposition 8
$$\begin{aligned} \begin{array}{lll}X&{}=&{}A\otimes I+I\otimes B=\left( \begin{array}{cccc}a_1I &{}0&{}0 &{}0\\ 0&{}-a_1I&{}0&{}0\\ 0&{}0&{}0&{}a_2I\\ 0&{}0 &{} a_3I&{}0\end{array}\right) +\left( \begin{array}{cccc}B&{}0&{}0&{}0\\ 0&{}B&{}0&{}0\\ 0&{}0&{}B&{}0\\ 0&{}0&{}0&{}B\end{array}\right) \\ &{}=&{}\left( \begin{array}{cccc}a_1I+B&{}0&{}0&{}0\\ 0&{}-a_1I+B &{}0&{}0\\ 0&{}0&{}B&{}a_2I\\ 0&{}0&{}a_3I&{}B\end{array}\right) . \end{array} \end{aligned}$$
We have
$$\begin{aligned} \begin{array}{lll}XX^\dag &{}= &{} \left( \begin{array}{cccc}a_1I+B&{}0&{}0&{}0\\ 0&{}-a_1I+B &{}0&{}0\\ 0&{}0&{}B&{}a_2I\\ 0&{}0&{}a_3I&{}B\end{array}\right) \left( \begin{array}{cccc}a_1^*I+B^\dag &{}0&{}0&{}0\\ 0&{}-a_1^*I+B^\dag &{}0&{}0\\ 0&{}0&{}B^\dag &{}a_3^*I\\ 0&{}0&{}a_2^*I&{}B^\dag \end{array}\right) \\ &{}=&{}\left( \begin{array}{cccc}BB^\dag +|a_1|^2I+a_1B^\dag +a_1^*B&{}0&{}0&{}0\\ 0&{}BB^\dag -a_1^*B-a_1B^\dag +|a_1|^2I&{}0&{}0\\ 0&{}0&{}BB^\dag +|a_2|^2I &{} a_2B^\dag +a_3^*B\\ 0&{}0&{}a_3B^\dag +a_2^*B&{}|a_3|^2I+BB^\dag \end{array}\right) ,\\ &{}=&{}C\oplus D.\end{array} \end{aligned}$$
where
$$\begin{aligned}&C=C_1\oplus C_2, \\&C_1=\left( \begin{array}{cccc} |a_1|^2+|b_1|^2&{}a_1b_2^*+a_1^*b_1&{}0&{}0\\ a_1b_1^*+a_1^*b_2&{}|a_1|^2+|b_2|^2&{}0&{}0\\ 0&{}0&{}|a_1|^2+|b_3|^2&{}a_1b_4^*+a_1^*b_3\\ 0&{}0&{}a_1b_3^*+a_1^*b_4&{}|a_1|^2+|b_4|^2\end{array}\right) , \\&C_2=\left( \begin{array}{cccc} |a_1|^2+|b_1|^2&{}a_1b_2^*-a_1^*b_1&{}0&{}0\\ a_1b_1^*-a_1^*b_2&{}|a_1|^2+|b_2|^2&{}0&{}0\\ 0&{}0&{}|a_1|^2+|b_3|^2&{}a_1b_4^*-a_1^*b_3\\ 0&{}0&{}a_1b_3^*-a_1^*b_4&{}|a_1|^2+|b_4|^2\end{array}\right) , \\ \end{aligned}$$
$$\begin{aligned}&D=\left( \begin{array}{cc}BB^\dag +|a_2|^2I&{}a_3^*B+a_2B^\dag \\ a_3B^\dag +a_2^*B&{}|a_3|^2I+BB^\dag \end{array}\right) \nonumber \\&\quad =\left( \begin{array}{cccccccc}|b_1|^2+|a_2|^2&{}0&{}0&{}0&{}0&{}a_3^*b_1+a_2b_2^*&{}0&{}0\\ 0&{}|b_2|^2+|a_2|^2&{}0&{}0&{}a_2b_1^*+a_3^*b_2&{}0&{}0&{}0\\ 0&{}0&{}|b_3|^2+|a_2|^2&{}0&{}0&{}0&{}0&{}a_2b_4^*+a_3^*b_3\\ 0&{}0&{}0&{}|b_4|^2+|a_2|^2&{}0&{}0&{}a_3^*b_4 +a_2b_3^*&{}0\\ 0&{}a_2^*b_1+a_3b_2^*&{}0&{}0&{}|b_1|^2 +|a_3|^2&{}0&{}0&{}0\\ a_3b_1^*+a_2^*b_2&{}0&{}0&{}0&{}0&{}|b_2|^2+|a_3|^2&{}0&{}0\\ 0&{}0&{}0&{}a_3b_4^*+a_2^*b_3&{}0&{}0&{}|b_3|^2 +|a_3|^2&{}\\ 0&{}0&{}a_2^*b_4 +a_3b_3^*&{}0&{}0&{}0&{}0&{}|b_4|^2 + |a_3|^2\end{array}\right) \nonumber \\&\quad \sim \left( \begin{array}{cccccccc}|b_1|^2+|a_2|^2&{}a_3^*b_1 +a_2b_2^*&{}0&{}0&{}0&{}0&{}0&{}0\\ a_3^*b_1+a_2^*b_2&{}|b_2|^2+|a_3|^2&{}0&{}0&{}0&{}0&{}0&{}0\\ 0&{}0&{}|b_2|^2+|a_2|^2&{}a_2b_1^* +a_3^*b_2&{}0&{}0&{}0&{}0\\ 0&{}0&{}a_2^*b_1 +a_3b_2^*&{}|b_1|^2+|a_3|^2&{}0&{}0&{}0&{}0\\ 0&{}0&{}0&{}0&{}|b_3|^2+|a_2|^2&{}a_2b_4^*+a_3^*b_3&{}0&{}0\\ 0&{}0&{}0&{}0&{} a_3b_3^*+a_2^*b_4&{}|b_4|^2+|a_3|^2&{}0&{}0\\ 0&{}0&{}0&{}0&{}0&{}0&{}|b_4|^2+|a_2|^2&{}a_3^*b_4+a_2b_3^*\\ 0&{}0&{}0&{}0&{}0&{}0&{} a_3b_4^*+a_2^*b_3&{}|b_3|^2+|a_3|^2\end{array}\right) \nonumber \\&\quad =D_1\oplus D_2\oplus D_3\oplus D_4. \end{aligned}$$
(45)
One can formulate the characteristic polynomial of C as follows.
$$\begin{aligned} \det (\lambda I-C)=f_1(\lambda )\cdot f_2(\lambda )\cdot f_3(\lambda )\cdot f_4(\lambda ), \end{aligned}$$
(46)
where
$$\begin{aligned} \begin{array}{lll} f_1(\lambda )&{}=&{}(|b_1|^2+|a_1|^2-\lambda )(|a_1|^2+|b_2|^2-\lambda )-|a_1b_2^*+a_1^*b_1|^2,\\ f_2(\lambda )&{}=&{}(|b_3|^2+|a_1|^2-\lambda )(|a_1|^2+|b_4|^2-\lambda )-|a_1b_4^*+a_1^*b_3|^2,\\ f_3(\lambda )&{}=&{}(|b_1|^2+|a_1|^2-\lambda )(|a_1|^2+|b_2|^2-\lambda )-|a_1b_2^*-a_1^*b_1|^2,\\ f_4(\lambda )&{}=&{}(|b_3|^3+|a_1|^2-\lambda )(|a_1|^2+|b_4|^2-\lambda )-|a_1b_4^*+a_1^*b_3|^2. \end{array} \end{aligned}$$
(47)
We first claim the larger root of \(f_i(\lambda )=0, \forall i=1,2,3,4\), is not greater than \(\frac{1}{4}\). Take \(f_1(\lambda )\) as an example. It’s clear that the sum of the two roots of \(f_1(\lambda )=0\) is \(2|a_1|^2+|b_1|^2+|b_2|^2\). Since \(XX^t=C\oplus D\) is a semipositive definite matrix. All the eigenvalues of C and D are nonnegative. This implies that all the roots of \(f_i(\lambda )=0, i=1, 2, 3, 4\) are nonnegative. So we have the larger root of \(f_1(\lambda )=0\) is not greater than the sum of two roots, i.e., \(2|a_1|^2+|b_1|^2+|b_2|^2\). Recall the \(\sum _{i=1}^3(|a_i|^2+|b_i|^2)+|a_1|^2+|b_4|^2=\frac{1}{4}.\) Therefore, we conclude that the larger root of \(f_1(\lambda )=0\) is not larger than \(\frac{1}{4}.\) One can draw the same conclusion of \(f_i(\lambda ), i=2, 3, 4\). Then, our claim holds.
This implies the largest eigenvalue of C is not greater than \(\frac{1}{4}, \)i.e.,
$$\begin{aligned} \lambda \le \frac{1}{4}. \end{aligned}$$
(48)
The characteristic polynomial of D can be expressed as follows:
$$\begin{aligned} \det (\lambda I-D)=g_1 (\mu )\cdot g_2(\mu )\cdot g_3(\mu )\cdot g_3(\mu ), \end{aligned}$$
(49)
where
$$\begin{aligned} \begin{array}{lll} g_1(\mu )&{}=&{}(|b_1|^2+|a_2|^2-\mu )(|a_3|^2+|b_2|^2-\mu )-|a_3^*b_1+a_2b_2^*|^2,\\ g_2(\mu )&{}=&{}(|b_2|^2+|a_2|^2-\mu )(|a_3|^2+|b_1|^2-\mu )-|b_1^*a_2+b_2a_3^*|^2,\\ g_3(\mu )&{}=&{}(|b_3|^2+|a_2|^2-\mu )(|a_3|^2+|b_4|^2-\mu )-|a_3^*b_3+a_2b_4^*|^2,\\ g_4(\mu )&{}=&{}(|b_4|^2+|a_2|^2-\mu )(|a_3|^2+|b_3|^2-\mu )-|a_3^*b_4+a_2b_3^*|^2. \end{array}\end{aligned}$$
(50)
By the same way, we can show that the maximal eigenvalue \(\mu \) of matrix D satisfying
$$\begin{aligned} \mu \le \frac{1}{4}. \end{aligned}$$
(51)
Equations (48) and (51) imply that \(\lambda +\mu \le \frac{1}{2}.\) This completes the proof.