🧮 Brain Teaser

The Monic Polynomial That Cannot Hide Below 1

Let p(z)=zn+an−1zn−1+⋯+a1z+a0,ak∈C,p(z) = z^n + a_{n-1}z^{n-1} + \cdots + a_1 z + a_0, \qquad a_k \in \mathbb{C}, be a monic polynomial of degree n≥1n \ge 1.

(a) Prove that max⁡∣z∣=1∣p(z)∣  ≥  1.\max_{|z|=1} |p(z)| \;\ge\; 1.

(b) Show that equality holds if and only if p(z)=znp(z) = z^n.

Warm-up thought: You may choose the lower coefficients a0,…,an−1a_0,\dots,a_{n-1} freely to make pp as small as possible on the unit circle. One might expect a clever choice to push the maximum below 11. Show that no choice works.

Contrast (no proof required): On the real interval [−1,1][-1,1], the smallest possible value of max⁡x∈[−1,1]∣p(x)∣\max_{x\in[-1,1]}|p(x)| over monic pp is 21−n2^{1-n}. This value is attained by a rescaled Chebyshev polynomial. Why is the unit circle so much less forgiving than the interval?

maximum modulusreversed polynomialmean value propertyParsevalChebyshev contrast

Answer: The Monic Polynomial That Cannot Hide Below 1

Key Idea / Intuition

Reverse the polynomial. The map z↦znp(1/z)z \mapsto z^n p(1/z) turns the leading coefficient 11 into a constant term 11, and on the unit circle it has the same modulus as pp. A holomorphic function's value at the center is the average of its values on the circle. So if the value at the center is 11, the function cannot be smaller than 11 everywhere on the circle.

On an interval there is no such averaging constraint, because the interval does not enclose a point where a mean value property forces anything. That is why Chebyshev polynomials can make the maximum exponentially small on [−1,1][-1,1].


Formal Proof / Solution

Step 1: The reversed polynomial. Define q(z)=zn p(1/z)=1+an−1z+an−2z2+⋯+a0zn.q(z) = z^n\, p(1/z) = 1 + a_{n-1} z + a_{n-2} z^2 + \cdots + a_0 z^n. This is a polynomial, so it is entire. Its constant term is q(0)=1q(0) = 1.

Step 2: Same size on the circle. For ∣z∣=1|z| = 1 we have 1/z=zˉ1/z = \bar z, which also lies on the unit circle. Hence ∣q(z)∣=∣z∣n ∣p(zˉ)∣=∣p(zˉ)∣.|q(z)| = |z|^n\,|p(\bar z)| = |p(\bar z)|. As zz runs over the unit circle, so does zˉ\bar z. Therefore max⁡∣z∣=1∣q(z)∣=max⁡∣z∣=1∣p(z)∣=:M.\max_{|z|=1}|q(z)| = \max_{|z|=1}|p(z)| =: M.

Step 3: Mean value property, for part (a). By Cauchy's integral formula, q(0)q(0) is the average of qq on the unit circle: 1=q(0)=12π∫02πq(eiθ) dθ.1 = q(0) = \frac{1}{2\pi}\int_0^{2\pi} q(e^{i\theta})\,d\theta. Taking absolute values gives 1≤12π∫02π∣q(eiθ)∣ dθ≤M.1 \le \frac{1}{2\pi}\int_0^{2\pi} |q(e^{i\theta})|\,d\theta \le M. This proves (a).

Step 4: Equality case, part (b). Suppose M=1M = 1. Then ∣q∣≤1|q| \le 1 on ∣z∣=1|z|=1. By the maximum modulus principle, ∣q∣≤1|q| \le 1 on the closed unit disk. But ∣q(0)∣=1|q(0)| = 1, so ∣q∣|q| attains its maximum at an interior point. Hence qq is constant, and q≡q(0)=1q \equiv q(0) = 1.

This means an−1=⋯=a0=0a_{n-1} = \cdots = a_0 = 0, so p(z)=znp(z) = z^n.

Conversely, p(z)=znp(z) = z^n has ∣p∣≡1|p| \equiv 1 on the circle, so equality holds.

Alternative proof via Parseval. Since {eikθ}\{e^{ik\theta}\} is orthonormal on the circle, 12π∫02π∣p(eiθ)∣2 dθ=1+∑k=0n−1∣ak∣2  ≥  1.\frac{1}{2\pi}\int_0^{2\pi}|p(e^{i\theta})|^2\,d\theta = 1 + \sum_{k=0}^{n-1}|a_k|^2 \;\ge\; 1. The maximum of ∣p∣2|p|^2 is at least its average. So M2≥1+∑∣ak∣2M^2 \ge 1 + \sum|a_k|^2, with equality only if every ak=0a_k = 0. This gives (a) and (b) at once, and even the sharper bound M≥1+∑∣ak∣2.M \ge \sqrt{1 + \textstyle\sum |a_k|^2}.

About the contrast. On [−1,1][-1,1], the polynomial 21−nTn(x)2^{1-n}T_n(x) (rescaled Chebyshev) is monic and has sup norm 21−n2^{1-n}. The interval is not the boundary of a region containing a point where the reversal trick applies.

In potential-theoretic language, the minimal sup norm of a monic polynomial on a compact set KK behaves like cap⁡(K)n\operatorname{cap}(K)^n. The capacity of the unit circle is 11, while the capacity of [−1,1][-1,1] is 1/21/2.

Type: Complex AnalysisEdit on GitHub ↗