Notes on Complex Analysis
Export PDFGitHub

§4.4.5 Hadamard Factorization Theorem

4.4.5 Hadamard Factorization Theorem

Theorem 4.4.5.1.

Let 𝑓(𝑧)=e𝜑(𝑧)𝑃(𝑧) be the Weierstrass canonical factorization of 𝑓, where 𝑓 is entire with finite order 𝜌=𝜌(𝑓) and 𝑓(0)=1. Then 𝜑 is a polynomial of degree ≤𝜌.
Proof.

By logarithmic differentiation and by taking 𝑝>𝜌−1 subsequent derivatives, we have

𝑓′(𝑧)𝑓(𝑧)=𝜑′(𝑧)+𝑃′(𝑧)𝑃(𝑧)⟹d𝑝d𝑧𝑝(𝑓′(𝑧)𝑓(𝑧))=𝜑(𝑝+1)(𝑧)+d𝑝d𝑧𝑝(𝑃′(𝑧)𝑃(𝑧)).

By applying Proposition 4.4.4.17Proposition 4.4.4.17Proposition 4.4.4.17 and Lemma 4.4.4.18Lemma 4.4.4.18Lemma 4.4.4.18, we have

−∑𝑘=1∞𝑝!(𝑎𝑘−𝑧)𝑝+1=𝜑(𝑝+1)(𝑧)−∑𝑘=1∞𝑝!(𝑎𝑘−𝑧)𝑝+1⟹𝜑(𝑝+1)≡0.
Hence, 𝜑 is a polynomial of degree ≤𝑝. Choosing 𝑝=1+⌊𝜌−1⌋>𝜌−1 so that 𝑝≤𝜌, the assertion follows. ⁠□

Corollary 4.4.5.1.1.

Let 𝑓(𝑧)=𝑧𝑚e𝜑(𝑧)𝑃(𝑧) be the Weierstrass canonical factorization of 𝑓, where 𝑓 is entire with finite order 𝜌=𝜌(𝑓). Then 𝜑 is a polynomial of degree ≤𝜌.
Proof.

Let 𝑓(𝑧)=𝑧𝑚𝑔(𝑧), where 𝑓 and 𝑔 are entire, 𝑔(0)≠0, and 𝑓 has finite order 𝜌(𝑓) and 𝑀(𝑟,𝑓)=𝑟𝑚𝑀(𝑟,𝑔) for 𝑟>0. For 𝜀>0, ∃𝑟′>0 such that 𝑟>𝑟′ implies

𝑀(𝑟,𝑓)=𝑟𝑚𝑀(𝑟,𝑔)≤e𝑟𝜌(𝑓)+𝜀⟹𝑀(𝑟,𝑔)≤e𝑟𝜌(𝑓)+𝜀.

Thus, 𝜌(𝑔)≤𝜌(𝑓) by letting 𝜀→0+. If 𝜌(𝑓)=0, this already gives 𝜌(𝑔)=𝜌(𝑓). Otherwise, for any 0<𝜀<𝜌(𝑓)2, ∀𝑟′>0, ∃𝑟>𝑟′ such that

𝑀(𝑟,𝑓)≥e𝑟𝜌(𝑓)−𝜀⟹𝑀(𝑟,𝑔)≥exp(𝑟𝜌(𝑓)−2𝜀(𝑟𝜀−𝑚log𝑟𝑟𝜌(𝑓)−2𝜀))≥e𝑟𝜌(𝑓)−2𝜀

because for sufficiently large 𝑟,

𝑟𝜀−𝑚log𝑟𝑟𝜌(𝑓)−2𝜀>1.

Hence, 𝜌(𝑔)≥𝜌(𝑓)−2𝜀. Letting 𝜀→0+ implies 𝜌(𝑓)=𝜌(𝑔). Let 𝑔(𝑧)=𝑐ℎ(𝑧) where 𝑐 is a constant, so that ℎ(0)=1. It is also trivial that 𝜌(𝑔)=𝜌(ℎ). Explicitly, we have ℎ(𝑧)=e𝜑−Log𝑐𝑃(𝑧).

By Theorem 4.4.5.1Theorem 4.4.5.1Theorem 4.4.5.1 on ℎ, 𝜑−Log𝑐 is a polynomial of degree ≤𝜌, and so is 𝜑. ⁠□

Proposition 4.4.5.2.

Let 𝜇∈ℤ≥0 and let {𝑎𝑛}𝑛∈ℕ⊂ℂ∗ be a sequence such that ∑𝑛=1∞|𝑎𝑛|−𝜇−1 converges. If

𝑃(𝑧)=∏𝑛=1∞𝐸𝜇(𝑧𝑎𝑛),

then for any 𝑧∈ℂ,

log|𝑃(𝑧)|≤(2𝜇+1)|𝑧|𝜇+1∑𝑛=1∞|𝑎𝑛|−𝜇−1.
Proof.

For |𝑤|<1, Taylor’s formula gives

log|𝐸𝜇(𝑤)|=ℜ𝔢(log(1−𝑤)+∑𝑗=1𝜇𝑤𝑗𝑗)=ℜ𝔢(−∑𝑗=𝜇+1∞𝑤𝑗𝑗)≤∑𝑗=𝜇+1∞|𝑤|𝑗𝑗≤|𝑤|𝜇+11−|𝑤|.

Multiplying by 1−|𝑤| gives

(1−|𝑤|)log|𝐸𝜇(𝑤)|≤|𝑤|𝜇+1for|𝑤|<1.

We now prove by induction that

log|𝐸𝜇(𝑤)|≤(2𝜇+1)|𝑤|𝜇+1

for any 𝑤∈ℂ. If 𝜇=0, then

log|𝐸0(𝑤)|=log|1−𝑤|≤log(1+|𝑤|)≤|𝑤|.

Suppose the assertion holds for 𝜇−1. Since

𝐸𝜇(𝑤)=𝐸𝜇−1(𝑤)exp(𝑤𝜇𝜇)⟹log|𝐸𝜇(𝑤)|=log|𝐸𝜇−1(𝑤)|+ℜ𝔢𝑤𝜇𝜇,

it follows that

log|𝐸𝜇(𝑤)|≤(2𝜇−1)|𝑤|𝜇+|𝑤𝜇𝜇|≤(2𝜇−1+1𝜇)|𝑤|𝜇.

Let 𝐴=2𝜇−1+1𝜇. If |𝑤|≥𝐴2𝜇+1, then

𝐴|𝑤|𝜇≤(2𝜇+1)|𝑤|𝜇+1.

If |𝑤|<𝐴2𝜇+1, then 𝐴≤2𝜇 implies |𝑤|<2𝜇2𝜇+1<1, giving 11−|𝑤|<2𝜇+1, so from the first estimate, we have

log|𝐸𝜇(𝑤)|≤|𝑤|𝜇+11−|𝑤|≤(2𝜇+1)|𝑤|𝜇+1.

Therefore,

log|𝑃(𝑧)|≤∑𝑛=1∞log|𝐸𝜇(𝑧𝑎𝑛)|≤(2𝜇+1)|𝑧|𝜇+1∑𝑛=1∞|𝑎𝑛|−𝜇−1.□

Then the results of Corollary 4.4.5.1.1Corollary 4.4.5.1.1Corollary 4.4.5.1.1, Theorem 4.4.4.12Theorem 4.4.4.12Theorem 4.4.4.12, and Proposition 4.4.5.2Proposition 4.4.5.2Proposition 4.4.5.2 may be consolidated into a single statement:

Theorem 4.4.5.3 (Hadamard Factorization Theorem).

Let 𝜇 be the genus of 𝑓 and let 𝜌 be the order of 𝑓, where 𝑓 is entire with finite order. Then 𝜇≤𝜌≤𝜇+1.
Proof.

Let 𝑝 be the rank of 𝑓 and let 𝑞 be the degree of the polynomial 𝜑 in its canonical factorization (Definition 4.4.4.14Definition 4.4.4.14Definition 4.4.4.14). By Theorem 4.4.4.12Theorem 4.4.4.12Theorem 4.4.4.12 and Corollary 4.4.5.1.1Corollary 4.4.5.1.1Corollary 4.4.5.1.1, 𝑝≤⌊𝜌⌋ and 𝑞≤𝜌. Therefore, 𝜇=max{𝑝,𝑞}≤𝜌.

By Proposition 4.4.5.2Proposition 4.4.5.2Proposition 4.4.5.2,

log|𝑃(𝑧)|≤(2𝑝+1)|𝑧|𝑝+1∑𝑛=1∞|𝑎𝑛|−𝑝−1.

Also, max|𝑧|=𝑟ℜ𝔢𝜑(𝑧)≤𝐶𝑟𝑞 for sufficiently large 𝑟 and some 𝐶>0. Hence,

log𝑀(𝑟,𝑓)≤𝑚log𝑟+𝐶𝑟𝑞+(2𝑝+1)𝑟𝑝+1∑𝑛=1∞|𝑎𝑛|−𝑝−1≤𝐶′𝑟𝜇+1

for sufficiently large 𝑟 and some 𝐶′>0. Therefore, 𝜌≤𝜇+1. ⁠□

Theorem 4.4.5.4.

The factorization

sin𝑧=𝑧∏𝑘=1∞(1−𝑧2π2𝑘2)
defines an entire function and uniformly converges on any compact disk 𝐷(0,𝑟)‾.
Proof.

The zeros of sin are simple at each of ℤ. Aside from the simple zero at 𝑧=0, let

𝑎𝑘={−π𝑘/2if𝑘∈2ℕ,π(𝑘+1)/2if𝑘∈ℕ∖2ℕ

enumerate the zeros of sin. By Example 4.4.4.9Example 4.4.4.9Example 4.4.4.9, and the Hadamard Factorization Theorem (Theorem 4.4.5.3Theorem 4.4.5.3Theorem 4.4.5.3), the order of sin is 1, the genus does not exceed 1, and

sin𝑧=𝑧e𝜑(𝑧)∏𝑘=1∞𝐸1(𝑧𝑎𝑘)=𝑧e𝜑(𝑧)∏𝑘=1∞(1−𝑧𝑎𝑘)exp(𝑧𝑎𝑘),

where 𝜑(𝑧)=𝑎𝑧+𝑏 is a polynomial (and where the product locally uniformly converges in ℂ). Since the partial products {𝑃𝑛}𝑛∈ℕ, where

𝑃𝑛=∏𝑘=1𝑛(1−𝑧𝑎𝑘)exp(𝑧𝑎𝑘),

have a single accumulation point, the subsequence {𝑃2𝑛}𝑛∈ℕ converges to the same point. Since

𝑃2𝑛=∏𝑘=12𝑛(1−𝑧𝑎𝑘)exp(𝑧𝑎𝑘)=∏𝑘=1𝑛[(1−𝑧π𝑘)exp(𝑧π𝑘)(1+𝑧π𝑘)exp(−𝑧π𝑘)],

we have

sin𝑧=𝑧e𝑎𝑧+𝑏∏𝑘=1∞(1−𝑧2π2𝑘2).

Then from sin𝑧=−sin(−𝑧) we have

𝑧e𝑎𝑧+𝑏∏𝑘=1∞(1−𝑧2π2𝑘2)≡𝑧e−𝑎𝑧+𝑏∏𝑘=1∞(1−𝑧2π2𝑘2)⟹e2𝑎𝑧≡1⟹𝑎=0.

Since lim𝜁→0sin𝜁𝜁=1, we have

lim𝑧→0e𝑏∏𝑘=1∞(1−𝑧2π2𝑘2)=1⟹𝑏=0.□