4
$\begingroup$

If $x\in\mathrm{M}_2(\mathbb{C})$, $y=\dfrac{x+x^{\dagger}}{2}$, and $z=\dfrac{z-z^{\dagger}}{2}$, then $x=y+z$. Also, $y$ and $z$ respectively are Hermitian and anti-Hermitian, i.e. $y^{\dagger}=y$ and $z^{\dagger}=-z$, where $^\dagger$ denotes the conjugate transpose. Now suppose $\det(x)=\det(y)=1$. Does this force $x^{\dagger}=x$? I cannot find a counterexample but I can't prove it either.

Some thoughts: Let $H$ and $A$ respectively denote the sets of $2\times2$ complex Hermitian and anti-Hermitian matrices, and consider the map $\phi:\mathrm{M}_2(\mathbb{C})\rightarrow H\times A$ which takes $x$ as above to $(y,z)$. Then $x\in H\iff\phi(x)=(x,0)$. Also, $\phi$ is a vector space isomorphism if we think of $\mathrm{M}_2(\mathbb{\mathbb{C}})$ as an 8 dimensional real vector space, so taking the first coordinate of its image gives a projection onto $H$ as a 4 dimensional real subspace. I want to say that this projection cannot fix the determinant of $x$ unless $x$ is already in its image, but it's difficult to say anything in particular about the determinant here because it maps into $\mathbb{C}$ and the vector spaces I'm thinking about are real.

$\endgroup$
  • 2
    $\begingroup$ Do you mean norm (as in the title) or determinant (as in your paragraph)? $\endgroup$ – Alex Zorn May 1 '15 at 18:33
  • $\begingroup$ I mean determinant, sorry I'll fix that. $\endgroup$ – j0equ1nn May 1 '15 at 18:53
  • 1
    $\begingroup$ It's not true in general in higher dimensions. Counterexample: $M=\pmatrix{1&3&4\\ -3&-1&5\\ -4&-5&-1}\oplus I$. Clearly $M$ is not Hermitian but both $M$ and its Hermitian part have determinant 1. Your question has an affirmative answer, however, if the Hermitian part of the matrix is positive definite. $\endgroup$ – user1551 May 2 '15 at 9:08
2
$\begingroup$

We have $x = y + z$. Any Hermitian matrix is orthogonally diagonalizable with real eigenvalues, so (after conjugating by a unitary matrix) we can assume $y$ is a diagonal matrix. (Note that conjugating by a unitary matrix will preserve the Hermitian/anti-Hermitian qualities of $y$ and $z$, as well as the determinants of each). By the determinant constraint we have:

$$y = \begin{bmatrix} \lambda & 0 \\ 0 & \lambda^{-1} \end{bmatrix}$$

Where $\lambda \in \mathbb{R}^{\times}$ is nonzero. We also have:

$$z = \begin{bmatrix} ia & w \\ -\overline{w} & ib \end{bmatrix}$$

Here $a,b \in \mathbb{R}$ and $w \in \mathbb{C}$. So $x$ has determinant $1$ if and only if:

$$(ia + \lambda)(ib + \lambda^{-1}) - (w)(-\overline{w}) = 1$$ $$-ab + i(a\lambda^{-1} + b\lambda) + 1 + w\overline{w} = 1$$ $$(w\overline{w} - ab) = 0 \qquad \text{and} \qquad a\lambda^{-1} + b\lambda = 0$$ The last line comes from separating the real and imaginary parts of the above equation. Now the second equation gives: $$a = -b\lambda^{2}$$ And so: $$w\overline{w} + b^{2}\lambda^{2} = 0$$ The last line is the sum of two nonnegative real numbers, which is zero if and only if both its terms are zero. So we get $w = 0$ and $b = 0$, it follows that $a = 0$ and so $z = 0$ and $x = y$ as conjectured.

This solves your question is 2 dimensions. I'm not sure about higher dimensions. This method will work in practice but it will quickly become burdensome, so if it is true in general then probably there is a nicer proof.

$\endgroup$
  • $\begingroup$ Excellent, I'd forgotten that Hermitian things diagonalize! I feel like it should hold in higher dimension but for my purposes this will do the trick. $\endgroup$ – j0equ1nn May 1 '15 at 18:56
  • $\begingroup$ Oh I would also mention that for the "up to conjugation by a unitary matrix" part to work, we need the fact that $z$ remains anti-Hermitian under such a conjugation. In general $A$ (and $H$) are not invariant under conjugation, but since it's done with something unitary it works. $\endgroup$ – j0equ1nn May 1 '15 at 19:30
  • $\begingroup$ That's a good point, I've edited in a note for potential future readers. $\endgroup$ – Alex Zorn May 1 '15 at 20:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.