🧮 Brain Teaser
Complex Analysis
Gauss–Lucas Theorem
2026-06-11
✏︎

The Zeros of a Derivative via the Gauss–Lucas Theorem

Let p(z)=(z1)(z2)(z3)(z4)p(z) = (z-1)(z-2)(z-3)(z-4) be a degree-4 polynomial with all real roots at 1,2,3,41, 2, 3, 4.

(a) The Gauss–Lucas theorem states: the zeros of p(z)p'(z) all lie in the convex hull of the zeros of p(z)p(z).

Without computing p(z)p'(z) explicitly, use this theorem to locate the zeros of p(z)p'(z).

(b) Now here is the surprise: prove the Gauss–Lucas theorem itself. That is, show that if p(z)p(z) is any polynomial with zeros z1,,znCz_1, \ldots, z_n \in \mathbb{C}, then every zero of p(z)p'(z) lies in the convex hull of {z1,,zn}\{z_1, \ldots, z_n\}.

(Hint for (b)): Write p(z)p(z)=k=1n1zzk\dfrac{p'(z)}{p(z)} = \sum_{k=1}^n \dfrac{1}{z - z_k} and think about what happens if zz is outside the convex hull.

logarithmic derivativeconvex hullroots of derivativesseparation theorem

Answer: Gauss–Lucas Theorem

Key Idea / Intuition

The logarithmic derivative of pp is a sum of simple fractions 1zzk\frac{1}{z - z_k}. If zz lies outside the convex hull of the roots, all the vectors zzkz - z_k point into a common open half-plane — meaning their reciprocals also all point into a common half-plane — so their sum cannot be zero. Hence p(z)0p'(z) \neq 0 outside the convex hull.


Formal Proof / Solution

Part (a): Locating the zeros of pp'

By the Gauss–Lucas theorem, all zeros of p(z)p'(z) lie in the convex hull of {1,2,3,4}\{1, 2, 3, 4\}. Since all roots are real, the convex hull is simply the interval [1,4][1, 4] on the real line.

Moreover, since pp has only real roots, pp' is also a real polynomial (degree 3), so its zeros are either real or come in conjugate pairs. Since they must lie in [1,4]R[1, 4] \subset \mathbb{R}, all three zeros of pp' are real and lie in the interval [1,4][1, 4].

By Rolle's theorem (which is the real-line version), there is exactly one zero of pp' in each of (1,2)(1,2), (2,3)(2,3), (3,4)(3,4) — Gauss–Lucas tells us this is the whole story even in C\mathbb{C}.


Part (b): Proof of the Gauss–Lucas Theorem

Setup. Write p(z)=ck=1n(zzk)p(z) = c\prod_{k=1}^n (z - z_k). Taking the logarithmic derivative:

p(z)p(z)=k=1n1zzk,for z{z1,,zn}.\frac{p'(z)}{p(z)} = \sum_{k=1}^n \frac{1}{z - z_k}, \quad \text{for } z \notin \{z_1, \ldots, z_n\}.

Goal. Show that if p(w)=0p'(w) = 0, then ww lies in the convex hull K=conv{z1,,zn}K = \operatorname{conv}\{z_1,\ldots,z_n\}.

Equivalently, we show: if wKw \notin K, then p(w)0p'(w) \neq 0.

Key geometric step. If wKw \notin K, by the separating hyperplane theorem (in R2C\mathbb{R}^2 \cong \mathbb{C}), there exists a real linear functional — i.e., a direction ξC\xi \in \mathbb{C} with ξ=1|\xi|=1 — such that:

Re(ξ(wzk))>0for all k=1,,n.\operatorname{Re}(\xi \cdot (w - z_k)) > 0 \quad \text{for all } k = 1, \ldots, n.

In other words, all the differences wzkw - z_k lie strictly on one side of a line through the origin.

Consequence for the sum. For each kk:

Re ⁣(ξwzk)=Re(ξ(wzk))wzk2=Re(ξ(wzk))-relatedwzk2.\operatorname{Re}\!\left(\frac{\xi}{w - z_k}\right) = \frac{\operatorname{Re}(\xi\,\overline{(w-z_k)})}{|w-z_k|^2} = \frac{\operatorname{Re}(\xi(w-z_k))^*\text{-related}}{|w-z_k|^2}.

More carefully: since Re(ξ(wzk))>0\operatorname{Re}(\xi(w-z_k)) > 0, taking the reciprocal preserves the real part's sign:

Re ⁣(ξˉwzk)=Re ⁣(ξ(wzk)1wzk2wzk2)wzk2.\operatorname{Re}\!\left(\frac{\bar{\xi}}{w-z_k}\right) = \frac{\operatorname{Re}\!\left(\overline{\xi(w-z_k)}\cdot\frac{1}{|w-z_k|^2}\cdot|w-z_k|^2\right)}{|w-z_k|^2}.

Let me write it cleanly. Write wzk=rkeiθkw - z_k = r_k e^{i\theta_k}. The assumption says Re(ξrkeiθk)>0\operatorname{Re}(\xi \cdot r_k e^{i\theta_k}) > 0 for all kk, i.e., Re(ξeiθk)>0\operatorname{Re}(\xi e^{i\theta_k}) > 0. Then:

Re ⁣(ξˉ1wzk)=Re ⁣(ξˉrkeiθk)=1rkRe(ξˉeiθk)=1rkRe(ξeiθk)=1rkRe(ξeiθk)>0.\operatorname{Re}\!\left(\bar\xi \cdot \frac{1}{w-z_k}\right) = \operatorname{Re}\!\left(\frac{\bar\xi}{r_k e^{i\theta_k}}\right) = \frac{1}{r_k}\operatorname{Re}(\bar\xi e^{-i\theta_k}) = \frac{1}{r_k}\operatorname{Re}\overline{(\xi e^{i\theta_k})} = \frac{1}{r_k}\operatorname{Re}(\xi e^{i\theta_k}) > 0.

Summing over kk:

Re ⁣(ξˉk=1n1wzk)=k=1nRe ⁣(ξˉwzk)>0.\operatorname{Re}\!\left(\bar\xi \sum_{k=1}^n \frac{1}{w-z_k}\right) = \sum_{k=1}^n \operatorname{Re}\!\left(\frac{\bar\xi}{w-z_k}\right) > 0.

Conclusion. Since the sum k1wzk0\sum_k \frac{1}{w-z_k} \neq 0, we get p(w)p(w)0\frac{p'(w)}{p(w)} \neq 0, so p(w)0p'(w) \neq 0.

Therefore, every zero of p(z)p'(z) must lie inside (or on the boundary of) the convex hull of {z1,,zn}\{z_1,\ldots,z_n\}. \blacksquare


Why this is beautiful: The proof uses nothing but the separating hyperplane theorem and a one-line observation about complex reciprocals — yet it gives a nontrivial geometric constraint on the roots of derivatives.

Source: Mathematical folklore / Stein–Shakarchi Complex Analysis, standard graduate complex analysis

Type: Complex AnalysisSource: Mathematical folklore / Stein–Shakarchi Complex Analysis, standard graduate complex analysisEdit on GitHub ↗