Let $P(z)=\sum_{k=0}^na_kz^k$ be a polynomial of degree $n$ having all its zeros on the unit circle. Let $M=\max_{0\leq k\leq n}\lvert a_k\rvert$. The polynomial $P(z)=z^n+1$ has $\max_{\lvert z\rvert=1}\lvert P(z)\rvert= 2$. Intuitively, it appears that $\max_{\lvert z\rvert=1}\lvert p(z)\rvert\geq 2M$. Kindly share your opinion on the validity of this bound $2M$. Can it be further sharpened?
-
7$\begingroup$ What about $1+z+... +z^n$? $\endgroup$– Fedor PetrovCommented Oct 28, 2023 at 6:42
-
1$\begingroup$ See my comment to Conrad's post below. $\endgroup$– GH from MOCommented Oct 29, 2023 at 0:56
-
2$\begingroup$ @Conrad's answer referenced by @GHfromMO, and their comment on it. $\endgroup$– LSpiceCommented Oct 29, 2023 at 1:03
-
1$\begingroup$ The trivial lower bound is $\frac\pi 2M$. It is certainly not sharp and will improve to the desired bound $2M$ if we could show, for instance, that for every polynomial $P$ with the roots on the circle and maximal value $2$, we have $|\{|P|>t\}|\le|\{|z-1|>t\}|$ for all $t\in(0,2)$, which seems plausible but I have no proof at the moment :-). $\endgroup$– fedjaCommented Oct 30, 2023 at 19:14
-
2$\begingroup$ The "Close" votes are not justified. Please do not vote to close this question, as it is clearly on topic. $\endgroup$– GH from MOCommented Nov 1, 2023 at 19:00
2 Answers
The result follows with $M=|a_n|$ actually as below:
Note that if $P$ has all the roots $w_j$ on the unit circle it follows that $w_j=1/\bar w_j$ so there is $c \ne 0$ st $$P(z)=cz^n\overline {P(1/\bar z)}=c\sum_{k=0}^n \bar a_kz^{n-k} $$ hence $a_k=c \bar a_{n-k}$ and in particular since $a_n \ne 0$ we have $a_0 \ne 0, |c|=1$ so $|a_0|=|a_n|$
But $a_0+a_nz^n=\frac{1}{n}\sum_{\omega^n=1}P(wz)$, so $|a_0+a_nz^n| \le \max_{|\zeta|=1}|P(\zeta)|$ for all $|z|=1$ hence choosing $|z|=1$ st $|a_0+a_nz^n|=|a_0|+|a_n|=2|a_n|$ the result follows
Note also that we only used that $P$ is self inversive ($w_j$ root iff $1/\bar w_j$ root and no zero roots) and actually only that $|a_0|=|a_n|$ which is implied by self-inversivity.
-
4$\begingroup$ The statement (in slightly sharper form) was the 2nd problem in the 1991 Schweitzer Contest in Hungary. See Problem G.47 in Székely (ed.): Contests in Higher Mathematics - Miklós Schweitzer Competitions (1962-1991). $\endgroup$ Commented Oct 29, 2023 at 0:54
-
3$\begingroup$ So $|a_0|+|a_n|$ is a lower bound for all polynomials without any assumptions. $\endgroup$ Commented Oct 29, 2023 at 7:29
-
3$\begingroup$ This is well known Visser’s inequality. $\endgroup$ Commented Oct 29, 2023 at 10:27
-
2$\begingroup$ Visser's inequality has been generalized by several researchers (e.g. there was an immediate generalization by Visser and van der Corput), and Google finds many references. Visser's original publication is C. Visser: A simple proof of certain inequalities concerning polynomials, Nederl. Akad. Wetensch. Proc. 48, 276-281 = Indag. Math. 7 (1945), 81-86. $\endgroup$ Commented Oct 29, 2023 at 16:55
-
3$\begingroup$ @fedja the result is true but nontrivial with the equality cases determined (they are essentially coming from complex polynomials of the type $c(z^m+e^{ia})^2, a\in \mathbb R$) - not sure who proved this first but a (4 page) proof is presented in the comprehensive book Analytic Theory of Polynomials by Rahman and Schmeisser, Th $16.1.9$ pages 646-650; see global.oup.com/academic/product/… $\endgroup$– ConradCommented Nov 1, 2023 at 17:28
Answer became outdated when the question was edited. Originally $2M$ was stated as an upper not lower bound, due to a typo. Watch your signs!
Yup, as suggested by Fedor in the comments the premise is false. For example, $P(z)=z^4+z^3+z^2+z+1$ has all zeroes on the boundary of the unit disk and all coefficients unity, but $|P(1)|>2$.
If $z$ is inside or on the unti disk, then all polynomial functions (regardless of their zeroes) satisfy $|P(z)|\le\sum|a_n|$ from the Triangle Inequality.
-
3$\begingroup$ It was a typo, a reverse bound is asked about $\endgroup$ Commented Oct 28, 2023 at 11:26