1
$\begingroup$

Hi I am wondering how to find a Hermitian matrix $X \in M_{4}(\mathbb{C})$ that satisfies the equation $$ X + \overline{X}\operatorname{tr}(X^2)-2X\operatorname{tr}(X^2)-2X^2\overline{X}-2X^3+2\overline{X} X^2 =0, $$ where $\overline{X}$ is complex conjugate of $X$ (since $X$ is Hermitian, $\overline{X}=X^\top$). Note that $$ Tr(X^2) = \sum_{i,j} X_{ij} X_{ji}. $$ Thanks!

  • 0
    Equation is subject to that constraint, and my matrix is defined above. The matrix is hermitian, this does not mean it has to be real.2017-02-18
  • 0
    @user1551 the equation is solved with the constraint that $X=X^*$ thus it is hermitian. $X^*=\overline{X^\top}$2017-02-18
  • 0
    @user1551 not familiar with your language sorry I am confused. However okay let's consider X to be real, can we find a solution? What I meant was; The matrix is hermitian which is the same as complex symmetric. A hermitian matrix is equal to the complex conjugate of the real transpose. In terms of not using 'self adjoint' , I am specifically talking about matrices and not operators.2017-02-19
  • 1
    I see. You've confused "complex symmetric" ($X=X^T$) with "Hermitian" ($X=X^\ast$). In linear algebra, a complex symmetric matriix is just an ordinary symmetric matrix. Since the term "symmetric matrix" usually refers to real symmetric matrix, when the matrix is complex, we often say "complex symmetric" instead of merely "symmetric" to stress that the matrix is complex.2017-02-19
  • 0
    @user1551 that makes alot of sense thanks for clearing that up!2017-02-19

1 Answers 1

1

As $X$ is Hermitian, $\overline{X}=X^T$. So, we may rewrite the equation as $\renewcommand{\tr}{\operatorname{tr}}$ $$X + \tr(X^2) (X^T-2X) - 2X^2X^T - 2X^3 + 2X^TX^2 = 0,\tag{1}$$ Taking the conjugate transpose of both sides, we get $$X + \tr(X^2) (X^T-2X) - 2X^TX^2 - 2X^3 + 2X^2X^T = 0.\tag{2}$$ So, the difference and sum of $(1)$ and $(2)$ give \begin{align} &X^2X^T = X^TX^2,\tag{3}\\ &X + \tr(X^2) (X^T-2X) - 2X^3 = 0.\tag{4} \end{align} Let $X=H+iK$, where $H,K\in M_4(\mathbb R)$ are respectively the real and imaginary parts of $X$. As $X$ is Hermitian, $H$ is real symmetric and $K$ is real skew-symmetric. Therefore $X^2 = (H^2-K^2) + i(HK+KH)$, $X^T = H-iK$ and $(3)$ reduces to \begin{align} &HK^2 = K^2H,\tag{5}\\ &KH^2 = H^2K.\tag{6} \end{align} In the special case that $H$ and $K$ commute, $(5)$ and $(6)$ are simultaneously satisfied. In this case, we may assume that $$ X=(a_1I+K_1)\oplus\cdots\oplus(a_mI+K_m),\tag{7} $$ where each $a_j$ is real and each $K_j$ is either zero or a nonzero $2\times2$ real skew symmetric matrix.

Let $t=\tr(X^2)$. Substitute $(7)$ into $(4)$, and compare the real and imaginary parts on both sides, we get, for each $j=1,\ldots,m$, \begin{align} a_j\left[(1-t - 2a_j^2)I + 6K_j^2\right] &= 0,\tag{8}\\ K_j\left[(1-3t - 6a_j^2)I + 2K_j^2\right] &= 0.\tag{9} \end{align} There are three possibilities:

  1. $a_j=0$. Then $(9)$ gives $K_j=0$ or $0\ne K_j^2 = \frac{3t-1}2I$. Since $K_j^2$ is a non-positive multiple of the identity matrix, if the latter case occurs, we must have $t<\frac13$.
  2. $a_j\ne0$ and $K_j\ne0$. Substitute $(8)$ into $(9)$, we get $(1-t - 2a_j^2)I - 3(1-3t - 6a_j^2)=0$. Hence $a_j = \pm\sqrt{\frac{1-4t}8}$ and in turn, $K_j^2=-\frac18I$. Note that since $a_j$ is real, if this case does occur, we must have $t<\frac14$.
  3. $a_j\ne0$ and $K_j=0$. Then $(8)$ gives $a_j=\pm\sqrt{(1-t)/2}$. Since $a_j$ has to be real, if this case occurs, $t<1$.

It follows that, up to real orthogonal similarity, $X$ is equal to \begin{align*} X&=0\oplus iK_\ell\\ &\oplus\left(\sqrt{\frac{1-4t}8}\,I_p+iK_p\right) \oplus\left(-\sqrt{\frac{1-4t}8}\,I_q+iK_q\right)\\ &\oplus\left(\sqrt{\frac{1-t}2}\,I_r\right) \oplus\left(-\sqrt{\frac{1-t}2}I_s\right),\tag{10} \end{align*} where $K_\ell,K_p,K_q$ are two real skew symmetric matrices such that $K_\ell^2=\frac{3t-1}2I_\ell,\ K_p^2=-\frac18I_p$ and $K_q^2=-\frac18I_q$ ($\ell,p,q$ are even numbers). Hence \begin{align*} t=\tr X^2 &=\frac{1-3t}2\ell + \left[\left(\frac{1-4t}8\right)+\frac18\right](p+q) + \frac{1-t}2(r+s)\\ &=\frac{1-3t}2\ell + \left(\frac{1-2t}4\right)(p+q) + \frac{1-t}2(r+s). \end{align*} That is, $$ t = \frac{2\ell + (p+q) + 2(r+s)}{4 + 6\ell + 2(p+q) + 2(r+s)}.\tag{11} $$ In other words, if the real part and imaginary part of $X$ commute ($HK=KH$), then up to real orthogonal similarity, all Hermitian solutions $X$ to $(1)$ are given by $(10)$ and $(11)$, subject to the requirements that \begin{cases} \ell,p,q\in2\mathbb N\cup\{0\},\\ r,s\in\mathbb N\cup\{0\},\\ \ell+p+q+r+s\le4,\\ t<\frac13&\text{ if } \ell>0,\\ t<\frac14&\text{ if } p+q>0,\\ t<1&\text{ if } r+s>0. \end{cases} Using a brute-force search, we see that all feasible values of $(\ell, p+q, r+s, t)$ are given by $(0,0,0,0), (0,0,1,\frac13), (0,0,2,\frac12), (0,0,3,\frac35), (0,0,4,\frac23), (2,0,0,\frac14), (4,0,0,\frac27)$. The first five cases mean that for some real orthogonal matrix $Q\in O(4,\mathbb R)$, $$ Q^TXQ=0\oplus\left(\sqrt{\frac{1-t}2}\,I_r\right) \oplus\left(-\sqrt{\frac{1-t}2}I_s\right);\ 0\le r+s\le 4,\ t=\frac{r+s}{r+s+2}. $$ The case $(2,0,0,\frac14)$ means that $$ Q^TXQ=\frac{i}{\sqrt{8}}\pmatrix{0&-1\\ 1&0}\oplus0 $$ and the case $(4,0,0,\frac27)$ means that $$ Q^TXQ=\frac{i}{\sqrt{14}}\left[\pmatrix{0&-1\\ 1&0}\oplus\pmatrix{0&-1\\ 1&0}\right]. $$ These give all self-adjoint solutions $X$ to $(1)$ subject to the additional condition that the real part and imaginary part of $X$ commute.

  • 0
    Thank you very much for this detailed answer! I have to go through it all, I may have some questions I will let you know.2017-02-19
  • 0
    Two questions; 1) what is the result for the simpler case of K=0? (When X is real). I am trying to put K=0 into your final answer I'm just not entirely sure how, this is a good way for me to check my result so I want to do this. 2) I'm just a little confused, what is the circular symbol with the 2 lines in it? It seems the first time you use it is in equation (7). Thanks a lot!2017-02-19
  • 1
    @Integrals The answer assumes that $HK=KH$, so the case $K=0$ is already included. In fact, it is the first of the three cases mentioned at the end of the answer, i.e. $Q^TXQ=\sqrt{\frac{1-t}{2}}(0\oplus I_r\oplus -I_s)$ for some appropriate $r,s,t$. The symbol "$\oplus$" is the standard notation for the direct sum of two matrices.2017-02-19
  • 0
    that makes sense, this is very helpful thanks. My one last question for you: is this easily extended to the case when $X$ is 3x3?; $X\in M_3(\mathbb{C})$?2017-02-19
  • 1
    @Integrals Yes. In fact, when $X$ is a $3\times3$ self-adjoint solution, we must have $HK=KH$ (but I'm not going to prove this here, for the lack of space), so this commutation is no longer an additional assumption. However, while the same argument applies, the conclusions are a bit different. E.g. since $X$ is 3x3, the last case (where $Q^TXQ$ is a direct sum of two 2x2 matrices) becomes invalid. Also, in the first case, the constraint $0\le r+s\le4$ becomes $0\le r+s\le 3$.2017-02-20
  • 0
    I don't think I have any more questions. This technique as to how to approach this kind of problem is very helpful for me. Thanks!2017-02-20