🧮 Brain Teaser

The Holomorphic Function That Cannot Be Analytically Continued

Let ff be holomorphic on the open unit disk D={∣z∣<1}\mathbb{D} = \{|z| < 1\}, defined by

f(z)=∑n=0∞zn!.f(z) = \sum_{n=0}^{\infty} z^{n!}.

Show that ff cannot be analytically continued beyond D\mathbb{D}; that is, D\mathbb{D} is the natural domain of ff (every point of the unit circle ∣z∣=1|z|=1 is a singularity).

Hint: Consider what happens at roots of unity.

lacunary seriesnatural boundaryanalytic continuationroots of unitypower series

Answer: Natural Boundary via Factorial Lacunary Series

Key Idea / Intuition

The series ∑zn!\sum z^{n!} has radius of convergence exactly 11, so ff is holomorphic inside D\mathbb{D}. The key observation is that roots of unity are dense on the unit circle, and at every root of unity the partial sums of ff blow up — meaning ff cannot be extended even locally past any arc of the circle. The factorials in the exponents are perfectly tuned to make this blow-up happen at a dense set, which then propagates to the whole boundary.


Formal Proof / Solution

Step 1: Radius of convergence.

By the ratio / root test, lim sup⁡n∣an∣1/n=1\limsup_{n} |a_n|^{1/n} = 1 (since the coefficients are 00 or 11 and 11's appear), so the series converges for ∣z∣<1|z|<1 and diverges for ∣z∣>1|z|>1. Thus ff is holomorphic on D\mathbb{D} and the radius of convergence is exactly 11.


Step 2: Blow-up at roots of unity.

Fix a primitive pp-th root of unity ζ=e2πik/p\zeta = e^{2\pi i k/p} for integers p≥1p \geq 1, gcd⁡(k,p)=1\gcd(k,p)=1.

For n≥pn \geq p, we have p∣n!p \mid n!, so

ζn!=e2πik⋅n!/p=1.\zeta^{n!} = e^{2\pi i k \cdot n!/p} = 1.

Therefore, for 0<r<10 < r < 1,

f(rζ)=∑n=0∞(rζ)n!=∑n=0p−1rn!ζn!⏟finite, bounded+∑n=p∞rn!.f(r\zeta) = \sum_{n=0}^{\infty} (r\zeta)^{n!} = \underbrace{\sum_{n=0}^{p-1} r^{n!}\zeta^{n!}}_{\text{finite, bounded}} + \sum_{n=p}^{\infty} r^{n!}.

The tail is a positive real series:

∑n=p∞rn!→r→1−∑n=p∞1=+∞.\sum_{n=p}^{\infty} r^{n!} \xrightarrow{r \to 1^-} \sum_{n=p}^{\infty} 1 = +\infty.

So ∣f(rζ)∣→+∞|f(r\zeta)| \to +\infty as r→1−r \to 1^-.


Step 3: Every boundary point is singular.

The set of all roots of unity {e2πik/p:p≥1, 0≤k<p}\{e^{2\pi i k/p} : p \geq 1,\, 0 \leq k < p\} is dense in the unit circle ∣z∣=1|z|=1.

Suppose for contradiction that there exists an arc UU on the unit circle and a holomorphic function gg on D∪U\mathbb{D} \cup U extending ff. Since gg is continuous on the compact set D‾∩(U∪D)\overline{\mathbb{D}} \cap (U \cup \mathbb{D}), it is bounded near every point of UU. But roots of unity are dense in UU, and we showed ∣f(rζ)∣→∞|f(r\zeta)| \to \infty as r→1−r \to 1^- for every root of unity ζ\zeta. This contradicts the boundedness of gg near any point in UU.

Therefore, no analytic continuation of ff exists across any arc of the unit circle.


Conclusion.

Every point of ∣z∣=1|z|=1 is a singularity of ff, so D\mathbb{D} is the natural domain of ff. The unit circle is a natural boundary. ■\blacksquare


Remark (why factorials?)

The key property used is that n!n! is eventually divisible by every positive integer pp. Any lacunary series ∑zan\sum z^{a_n} where ana_n grows fast enough (Hadamard's gap condition: an+1/an≥λ>1a_{n+1}/a_n \geq \lambda > 1 for all nn) also has the unit circle as a natural boundary, but the factorial case gives the cleanest density argument via roots of unity.

Source: Complex Analysis (Stein–Shakarchi), Chapter 2; classical folklore

Type: Complex AnalysisSource: Complex Analysis (Stein–Shakarchi), Chapter 2; classical folkloreEdit on GitHub ↗