Let $A$, $B$, and $C$ be $2\times 2$ complex matrices, with $A$ and $C$ rank $1$ Hermitian. Can we find a real number $a$ and a $2\times 2$ unitary $U$ such that $$A + BV + V^*B^* + V^*CV$$ is a scalar multiple of the identity, where $V = aU$?

The motivation is that this would answer this question in a large special case.


Yes, one can always do this. In fact, one can assume that $\det(U)=1$, as this follows by a homotopy argument, using the fact that $\pi_3(S^2)\simeq \mathbb{Z}$.

Here are the details: Assume that $\det(U)=1$ so that $V$ has a non-negative real determinant. Then one can write $V$ uniquely in the form $$ V(x) = \begin{pmatrix} x_0+i\,x_1 & -x_2 + i\,x_3\\ x_2-i\,x_3 & \phantom{-}x_0-i\,x_1 \end{pmatrix} $$ for some $x = (x_0,x_1,x_2,x_3)\in\mathbb{R}^4$.

Now consider the map $F$ from $\mathbb{R}^4$ into $\mathbb{R}^3$ (regarded as the traceless Hermitian $2$-by-$2$ matrices) defined by $$ F(x) = \left[A +BV(x) + V(x)^*B^* + V(x)^*CV(x)\right]_0\,, $$ where, for a Hermitian $2$-by-$2$ matrix, $H$, we write $H - \tfrac12\mathrm{tr}(H)\,I_2 = H_0$. For use below, define the norm on traceless Hermitian $2$-by-$2$-matrices $M$ by the rule $|M|^2 = \tfrac12 \mathrm{tr}(M^2)$.

Our task is to show that there exists an $x\in\mathbb{R}^4$ such that $F(x)=0$. Note that $F(x)$ is a quadratic polynomial in $x$ taking values in $\mathbb{R}^3$.

When $|x|$ very large, $F(x)/|x|^2$ is close to $G(x) = (V(x)/|x|)^*C_0(V(x)/|x|)$, a vector whose norm is $|C_0|$. Since $C$ has rank 1, $C_0$ is not zero. Thus, when $|x|>>0$, $\bigl|F(x)\bigr|/|x|^2\approx |C_0|>0$. Moreover, when restricted to the $3$-sphere of radius $R>0$ in $\mathbb{R}^4$, the homogeneous map $G: S^3(R)\to S^2(|C_0|)$ is simply the Hopf map $S^3\to S^2$, appropriately scaled, which is known not to be homotopically trivial. (In fact, it is a generator of $\pi_3(S^2)\simeq\mathbb{Z}$.)

If there were not a $y\in\mathbb{R}^4$ that satisfied $F(y)=0$, then, for $R>>0$ the map $H(x) = F(x)/|F(x)|$ would map the $4$-ball of radius $R$ to $S^2$ in such a way that the map on its boundary $3$-sphere would be homotopic to the Hopf map, which is impossible.

Thus, there is a $y\in\mathbb{R}^4$ such that $F(y)=0$, i.e., $$ A +BV(y) + V(y)^*B^* + V(y)^*CV(y) = \mu\,I_2 $$ for some number $\mu$, as desired.

  • $\begingroup$ Great answer. I used it to give a partial answer to my earlier question about simultaneous "orthonormalization" in $\mathbb{C}^4$. $\endgroup$ – Nik Weaver Dec 28 '17 at 15:08
  • $\begingroup$ I've just put a paper on the arXiv that uses this idea: arxiv.org/abs/1802.07394 $\endgroup$ – Nik Weaver Feb 22 '18 at 20:59

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.