When working on this MSE question, I was led to conjecture the following :

If $z_1,z_2,z_3,z_4,z_5$ are five complex numbers with modulus $1$, such that $z_1+z_2+z_3+z_4+z_5=0$, then

$$ \bigg|\sum_{cyc} z_kz_{k+1}\bigg|^2 \geq \bigg(\frac{7-3\sqrt{5}}{4}\bigg)\Bigg(\sum_{cyc} \big|z_k^2-z_{k-1}z_{k+1}\big|^2\Bigg) $$

Or in other words,

$$ \begin{array}{l} \bigg|z_1z_2+z_2z_3+z_3z_4+z_4z_5+z_5z_1\bigg|^2 \geq \big(\frac{7-3\sqrt{5}}{4}\big) \times \\ \Bigg( \big|z_1^2-z_5z_2\big|^2+\big|z_2^2-z_1z_3\big|^2+ \big|z_3^2-z_2z_4\big|^2+\big|z_4^2-z_3z_5\big|^2+ \big|z_5^2-z_4z_1\big|^2\Bigg) \end{array} $$

Does anyone have an idea on how to prove (or disprove) this ?

  • $\begingroup$ How precisely is your conjecture related to the MSE question? If I knew that, I would perhaps see my way ahead. $\endgroup$ – chizhek Aug 28 '14 at 10:22

What follows is not a definitive answer. I have only managed to convince myself that one should try to prove the inequality, not to disprove it, because it is (almost) certainly true. So far I have only a visual 'empirical evidence' for this belief of mine, which, however, is very compelling. $\newcommand{\conjug}{\overline} $$\newcommand{\RR}{\mathbb{R}} $$\newcommand{\defeq}{:=} $$\newcommand{\abs}[1]{{\left|#1\right|}} $$\newcommand{\union}{\cup} $

Let $S$ be the unit circle in the plane of complex numbers. Write the difference LHS $-$ RHS of the conjectured inequality as $f(z_1,z_2,z_3,z_4,z_5)$, with $z_1,z_2,z_3,z_4,z_5\in S$. The function $f$ is of course invariant under cyclic permutations of its arguments; it is also invariant under rotations of the complex plane $z_i\mapsto uz_i$ (where $u\in S$), and under the complex conjugation $z_i\mapsto\conjug{z}_i$, $1\leq i\leq 5$. With a suitable rotation we can make $x\defeq z_1+z_2$ a nonnegative real number, and if need be we apply complex conjugation to get $z_1=\exp(i\varphi)$ and $z_2=\exp(-i\varphi)$ for some $0\leq\varphi\leq\pi/2$, so that $x=2\cos\varphi$. Write $a\defeq z_1+z_2+z_3$. Now if we know $z_1$, $z_2$, and $z_3=\exp(i\theta)$, then we almost know also $z_4$ and $z_5$, because $z_4+z_5=-a$. If $0<\abs{a}<2$, then there are two possibilities: if $z_4=u$, $z_5=v$ is one, then $z_4=v$, $z_5=u$ is the other one. When $\abs{a}=2$, then $z_4=z_5=-a/2$. Finally, if $a=0$, then $z_4=z_5=u$, where $u\in S$ is arbitrary, so in this case we have a singularity with an infinity of possibilities and with infinitely many possible values of $f$; it occurs when $z_1=\exp(i\pi/3)$, $z_2=\exp(-i\pi/3)$, and $z_3=-1$. To eliminate the need to separately observe two 'branches' of $f$, we allow $\varphi$ in the range $-\pi/2\leq\varphi\leq\pi/2$, while we require, for $a\neq0$, that $a+z_4$ lies on the left side of the line (perhaps on the line itself) through the origin $0$ and the point $a$, with the line directed from $0$ to $a$; this time we use the complex conjugation to get the point $a+z_4$ to the left side of the line.

We have parameterized the representative $5$-tuples $(z_1,z_2,z_3,z_4,z_5)$ by $\varphi\in[-\pi/2,\pi/2]$ and by $\theta$. If $x=2\cos\varphi\leq 1$, then $\theta$ can range through the whole interval $[0,2\pi]$. If $x>1$, then $\theta$ is restricted to the interval $[\alpha,2\pi\!-\!\alpha]$, $\alpha\defeq\arccos\bigl((3-x^2)/2x\bigr)$; this restriction comes from the condition $\abs{z_1+z_2+z_3}\leq 2$. Here is the diagram of $f$ as a function of the parameters $\varphi$ and $\theta$:

$\qquad$diagram of f

Note the discontinuities at the points $(\varphi,\theta)=(-\pi/3,\pi)$ and $(\varphi,\theta)=(\pi/3,\pi)$. At each of these two singularities the function $f$ can have any value between $-11-7\sqrt{3}+6\sqrt{5}+3\sqrt{15}\doteq1.911$ and $5(19-3\sqrt{5})/8\doteq6.432$ -- every one of these values is actually reachable as the limit of $f(\varphi,\theta)$ as $(\varphi,\theta)$ approaches the singularity from appropriate direction.

This is the view of the diagram from a viewpoint at the level $0$ and some distance away from the diagram to its left:

$\qquad$diagram of f, side view

This view of the diagram sits on the line that marks the level $0$. The function $f$ attains the zero value at two points, the point $(-\pi/5,3\pi/5)$ and the point $(-2\pi/5,6\pi/5)$. Let's look at the vectors $z_1$, $z_2$, $z_3$, $z_4$, $z_5$ which, added in this order, yield the result $0$. The vectors corresponding to the first point are shown on the left, and the vectors corresponding to the second point on the right:

$\qquad\qquad$ zi's for the first point $\qquad\qquad\quad$ zi's for the second point

So this is my non-definitive answer. Perhaps it already contains a germ of a bona fide proof of the inequality.


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.