Consider a polynomial: $$P_n(z)=\left(z-\frac{1}{2}\right)^{2n}+\left(z+\frac{1}{2}\right)^{2n}+1$$ where $n$ is a positive integer.

Obviously the polynomial has no real roots. The following properties are valid by numerical evidence:

At least one of the statements: $$ \left|z_i-\frac{1}{2}\right|=1,\;\left|z_i+\frac{1}{2}\right|=1, \text{ or } \Re(z_i)=0 $$ is valid for any root $z_i$ of the polynomial $P_n(z)$.

Further, all roots of the polynomial are distinct except for $\pm\frac{i\sqrt{3}}{2}$, which have multiplicity $n\;\text{mod}\;3$, and thus are two-fold degenerate for $n=2\;\text{mod}\;3$.

I would appreciate any hint on a proof of these properties.

  • $\begingroup$ The 'distinct' part fails for $n=2$, since $(x-1/2)^4+(x+1/2)^4+1=\frac{(4x^2+3)^2}{8}$ has multiple roots. $\endgroup$ – user547557 Apr 2 '18 at 0:32
  • $\begingroup$ @yixing Thank you. I overlooked the special character of $\pm\frac{i\sqrt{3}}{2}$ root. Corrected. $\endgroup$ – user Apr 2 '18 at 2:56
  • 4
    $\begingroup$ Idea. Once you find the set $Z_L$ of zeros of $P_n$ on the left arc $\{ \frac{1}{2} - e^{i\theta} : -\frac{\pi}{3} < \theta < \frac{\pi}{3} \}$, then $$ Z_R = \{ -z : z \in Z_0 \} $$ is the set of zeros of $P_n$ on the right arc and $$ Z_I = \left\{ \frac{1}{\frac{1}{2}+z} - \frac{1}{2} : z \in Z_0\right\} $$ is the set of zeros of $P_n$ on the imaginary axis, possibly except for $\pm \frac{i\sqrt{3}}{2}$. (In particular, $Z_R, Z_L, Z_I$ are disjoint.) Now it seems not terribly hard to show that $|Z_R| \geq 2\lfloor n/3 \rfloor$, and this easily leads to the desired claim. $\endgroup$ – Sangchul Lee Apr 2 '18 at 5:43
  • 1
    $\begingroup$ I would very appreciate if the five people who closed my question as "off-topic" let me know the reason for such strange decision. I would like also to thank other five people who have reopened it. $\endgroup$ – user Apr 2 '18 at 20:31

Here is a more detailed explanation of my idea. We first introduce four sets

\begin{align*} \mathsf{S}_T &= \left\{ -\frac{i\sqrt{3}}{2}, \frac{i\sqrt{3}}{2} \right\} \\ \mathsf{S}_L &= \{ z \in \mathbb{C} : \text{$|z - \tfrac{1}{2}| = 1$ and $\operatorname{Re}(z) < 0$} \} \\ \mathsf{S}_R &= \{ z \in \mathbb{C} : \text{$|z + \tfrac{1}{2}| = 1$ and $\operatorname{Re}(z) > 0$} \} \\ \mathsf{S}_I &= \{ z \in \mathbb{C} : \text{$\operatorname{Re}(z) = 0$ and $|z| > 1$} \} \end{align*}

Likewise, define $ \mathsf{Z}_{*} $ as the set of zeros of $P_n$ lying on $\mathsf{S}_{*}$, counted with multiplicity, for each symbol $* \in \{ T, L, R, I \} $. (Here, $T=$ triple junction, $L=$ left arc, $R=$ right arc, $I=$ imaginary axis.)

We first make some observations on the sets $\mathsf{S}_{*}$ and $\mathsf{Z}_{*}$.

  1. $\mathsf{S}_{*}$ are disjoint for different symbols $* \in \{ T, L, R, I \} $. In particular, it follows that

    $$|\mathsf{Z}_L| + |\mathsf{Z}_R| + |\mathsf{Z}_I| + |\mathsf{Z}_T| \leq 2n. $$

  2. The followings are equivalent:

    $$ z \in \mathsf{S}_L \qquad \Leftrightarrow \qquad -z \in \mathsf{S}_R \qquad \Leftrightarrow \qquad \frac{1}{z+\frac{1}{2}}-\tfrac{1}{2} \in \mathsf{S}_I. $$

    Equivalence of first two is clear. For equivalence of the first one and the third one, notice that $z = \frac{1}{2} - e^{it}$ if and only if $\frac{1}{z+\frac{1}{2}}-\tfrac{1}{2} = \frac{i}{2}\cot\left(\frac{t}{2}\right)$.

  3. We have

    $$ \left( z+\tfrac{1}{2} \right)^{2n}P_n \left( \frac{1}{z+\frac{1}{2}}-\tfrac{1}{2} \right) = P_n(z) = P_n(-z). $$

    In particular, together with the previous part, we find that $|\mathsf{Z}_L| = |\mathsf{Z}_R| = |\mathsf{Z}_I|$.

  4. As OP has already observed, we have $|\mathsf{Z}_T| = 2(n \text{ mod } 3)$. This follows by investigating

    \begin{align*} P_n\left(\pm \frac{i\sqrt{3}}{2} \right) &= 1 + 2\cos\left(\frac{2n\pi}{3} \right) \\ P_n^{(1)}\left(\pm \frac{i\sqrt{3}}{2} \right) &= 4ni\sin\left(\frac{(2n-1)\pi}{3} \right) \\ P_n^{(2)}\left(\pm \frac{i\sqrt{3}}{2} \right) &= 4n(2n-1)\cos\left(\frac{(2n-2)\pi}{3} \right) \end{align*}

In particular, we find that $|\mathsf{Z}_L| \leq 2 \lfloor n/3 \rfloor$. To conclude, it suffices to show the following claim:

Claim. We have $|\mathsf{Z}_L| \geq 2 \lfloor n/3 \rfloor$.

Indeed, let us plug $z = \frac{1}{2} - e^{it}$ with $|t| < \frac{\pi}{3}$ to the equation $P_n(z) = 0$. Upon simplification, we obtain

$$ \cos(nt) = \frac{(-1)^{n-1}}{2} (2\sin(t/2))^{2n}. \tag{*} $$

Notice that the RHS of $\text{(*)}$ is $<\frac{1}{2}$ whenever $|t| < \frac{\pi}{3}$. Hence for each $1 \leq k \leq \lfloor n/3 \rfloor$, there exists at least one $t = t_k \in \left( \frac{(k-1)\pi}{n}, \frac{k\pi}{n} \right)$ such that the equation $\text{(*)}$ is satisfied by the intermediate value theorem. Since $t = -t_k$ also solves $\text{(*)}$, it follows that

$$ \mathsf{Z}_L \supseteq \{ \tfrac{1}{2} - e^{it} : \text{$t = \pm t_k$ for some $1 \leq k \leq \lfloor n/3\rfloor$} \} $$

and therefore $|\mathsf{Z}_L| \geq 2\lfloor n/3 \rfloor$ as desired.

| cite | improve this answer | |
  • $\begingroup$ Thank you very much for the splendid proof. I believe however there is a misprint in "4". $P_n^{(1)}$ should be equal to $4ni\sin\frac{(2n-1)\pi}{3}$. Could you suggest an intuitively understandable reason why the "arc" roots of the polynomial are so extremly close to $\pm\frac{1}{2}\mp\exp\left(\frac{2k+1}{2n}\pi i\right)$? $\endgroup$ – user Apr 2 '18 at 20:18
  • 1
    $\begingroup$ @user, You are right, I made some mistake when simplifying the derivatives. As for the additional question, those zeros come from $t_k$'s and since the RHS of $\text{(*)}$ decays exponentially to $0$, it loosely follows that $t_k$'s become close to the zero of $\cos(nt) = 0$ in the interval $\left(\frac{(k-1)\pi}{n}, \frac{k\pi}{n} \right)$. I guess this explains the phenomena. $\endgroup$ – Sangchul Lee Apr 2 '18 at 21:29
  • $\begingroup$ The explanation sounds quite reasonable. Thank you once more. $\endgroup$ – user Apr 3 '18 at 12:10

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.