Notes on Complex Analysis
Export PDFGitHub

§3.3 Zeros of a Holomorphic Function

3.3 Zeros of a Holomorphic Function

For a region 𝑈⊆ℂ and a holomorphic function 𝑓:𝑈→ℂ, a point 𝑧0∈𝑈 is a zero of 𝑓 iff 𝑓(𝑧0)=0. Furthermore, if 𝑓 has the Taylor expansion at 𝑧0 of

𝑎𝑚(𝑧−𝑧0)𝑚+𝑎𝑚+1(𝑧−𝑧0)𝑚+1+⋯,𝑚∈ℕ,𝑎𝑚≠0,

then the zero at 𝑧0 has multiplicity 𝑚.

We will introduce a fundamental application of Liouville’s Theorem (Theorem 3.2.5Theorem 3.2.5Theorem 3.2.5) below.

Theorem 3.3.1 (Fundamental Theorem of Algebra).

Every non-constant polynomial 𝑝(𝑧) with complex coefficients has at least one complex zero.
Proof.For the sake of contradiction, suppose that 𝑝(𝑧) has no complex zeros. Then the function 𝑓(𝑧)=1𝑝(𝑧) is continuous and entire, because 𝑝(𝑧) has no zeros in ℂ. Moreover, as 𝑧→∞, 𝑝(𝑧)→∞, so 𝑓(𝑧)→0, and thus 𝑓(𝑧) is bounded. By Liouville’s Theorem (Theorem 3.2.5Theorem 3.2.5Theorem 3.2.5), every bounded entire function is constant. Thus, 𝑓(𝑧) is constant, and so 𝑝(𝑧) must also be constant. By contradiction, 𝑝(𝑧) has at least one complex zero. ⁠□

Theorem 3.3.2.

Let 𝑈⊆ℂ be open and connected, and 𝑓:𝑈→ℂ be holomorphic over 𝑈. Then if the set defined by

𝑆={𝑧∈𝑈:𝑓(𝑧)=0}
has an accumulation point in 𝑈, then 𝑓≡0 over 𝑈.
Proof.

Let {𝑧𝑛}𝑛∈ℕ be a subset of 𝑆 and assume it has an accumulation point 𝑧∞ in 𝑈. Since 𝑓 is holomorphic over 𝑈, ∃𝜀>0 such that 𝑓 is holomorphic over 𝐷(𝑧∞,𝜀)⊆𝑈. Then over this disk, 𝑓 has the Taylor expansion

𝑓(𝑧)=∑𝑛=0∞𝑎𝑛(𝑧−𝑧∞)𝑛.(3.3.1)

By Definition 1.1.3Definition 1.1.3Definition 1.1.3, ∃𝑁∈ℕ such that ∀𝑛>𝑁, 𝑧𝑛∈𝐷(𝑧∞,𝜀). Since 𝑧𝑛 is a zero of 𝑓, 𝑓(𝑧𝑛)=0. Then, by the continuity of 𝑓,

lim𝑛→∞𝑓(𝑧𝑛)=𝑓(lim𝑛→∞𝑧𝑛)=𝑓(𝑧∞)=0.

Using this result in comparison to (3.3.1)(3.3.1)(3.3.1), we get that 𝑎0=0.

The function

𝑓1(𝑧)=𝑓(𝑧)𝑧−𝑧∞

has a Taylor expansion over 𝐷(𝑧∞,𝜀) of

𝑓1(𝑧)=∑𝑛=0∞𝑎𝑛+1(𝑧−𝑧∞)𝑛.

Let 𝑧=𝑧𝑛≠𝑧∞ for some 𝑛>𝑁. Then 𝑓1 vanishes, leaving

0=𝑎1+𝒪︀(𝑧𝑛−𝑧∞).

Letting 𝑛→∞, 𝑧𝑛→𝑧∞, and 𝑎1=0. Define

𝑓2(𝑧)=𝑓1(𝑧)𝑧−𝑧∞.

Then,

𝑓2(𝑧)=∑𝑛=0∞𝑎𝑛+2(𝑧−𝑧∞)𝑛.

Similarly, 𝑎2=0. Letting

𝑓𝑛(𝑧)=𝑓(𝑧)(𝑧−𝑧∞)𝑛,

the sequence {𝑎𝑛}𝑛∈ℤ≥0 vanishes, and 𝑓≡0 on 𝐷(𝑧∞,𝜀).

Let

𝑆̃={𝑧∈𝑈:∀𝑛∈ℤ≥0,𝑓(𝑛)(𝑧)=0}.

For all 𝑧∈𝐷(𝑧∞,𝜀), since 𝑓(𝑧) locally vanishes (and has vanishing derivatives as a consequence),

𝐷(𝑧∞,𝜀)⊆𝑆̃.

Furthermore, for all 𝑧′∈𝑆̃, ∃𝜀′>0 such that 𝑓(𝑧) has a convergent Taylor series with vanishing coefficients on 𝐷(𝑧′,𝜀′)⊆𝑈. Then 𝑓≡0 on 𝐷(𝑧′,𝜀′). Then for all 𝑧∈𝐷(𝑧′,𝜀′), since 𝑓 is constant at 𝑧, it also has vanishing derivatives. It follows that

𝐷(𝑧′,𝜀′)⊆𝑆̃.

Since every point in 𝑆̃ has an open neighborhood also in 𝑆̃, 𝑆̃ is open.

It is evident that for all 𝑘∈ℤ≥0, 𝑓(𝑘) is continuous in 𝑈 by the holomorphy of 𝑓. Let

𝑆𝑘={𝑧∈𝑈:𝑓(𝑘)(𝑧)=0}.

For any sequence {𝑧̃𝑛}∈𝑆𝑘 converging to some 𝑧̃∞∈𝑈, by the continuity of 𝑓,

lim𝑛→∞𝑓(𝑘)(𝑧̃𝑛)=𝑓(𝑘)(lim𝑛→∞𝑧̃𝑛)=𝑓(𝑘)(𝑧̃∞)=0,

and therefore 𝑧̃∞∈𝑆𝑘. Thus, 𝑆𝑘 contains all of its accumulation points in 𝑈 and is therefore closed in 𝑈 (if 𝑧̃∞∉𝑈, then it is no longer relevant; we are concerned about it being closed within 𝑈). Since

𝑆̃=⋂𝑘∈ℤ≥0𝑆𝑘

and each of 𝑆𝑘 is closed in 𝑈, 𝑆̃ is the intersection of closed sets and consequently closed.

Since 𝑆̃ is nonempty and clopen in the connected set 𝑈, 𝑆̃=𝑈 (by Theorem 3.2.1.19Theorem 3.2.1.19Theorem 3.2.1.19). It follows that 𝑓≡0 on 𝑈. ⁠□

Remark.

This is a trivial property of holomorphic functions that allows for the uniqueness of analytic continuations. It is oftentimes stated in the form below:

Theorem 3.3.3 (Identity Theorem).

Let 𝑈⊆ℂ be open and connected, and define 𝑓(𝑧) and 𝑔(𝑧) to be two holomorphic functions on 𝑈. For a set 𝑆⊆𝑈 with an accumulation point in 𝑈, if 𝑓≡𝑔 on 𝑆, then 𝑓≡𝑔 on 𝑈.
Proof.Let ℎ=𝑓−𝑔 be holomorphic over 𝑈. Since 𝑆 has an accumulation point in 𝑈, and ℎ≡0 over 𝑆, then by Theorem 3.3.2Theorem 3.3.2Theorem 3.3.2, ℎ≡0 over 𝑈. ⁠□

Theorem 3.3.4 (Holomorphic Argument Principle).

Let 𝑈⊆ℂ be a region and 𝑓:𝑈→ℂ be holomorphic. Let 𝛾⊂𝑈 be a simple, closed, positively oriented curve that is null-homotopic in 𝑈. If 𝑓 has no zeros on 𝛾, then 𝑓 has finitely many zeros in the region bounded by 𝛾, and this number, counting multiplicities, is given by

𝑘=12πi∮Γ𝑓′(𝑧)𝑓(𝑧)d𝑧.

Let Γ be the image of 𝛾 under the map 𝑤=𝑓(𝑧). Then

𝑘=12πΔΓarg(𝑤),

where ΔΓarg(𝑤) denotes the total change in argument of 𝑤 as it traverses Γ.

Proof.

Let 𝑧1,…,𝑧𝑛 be the distinct zeros of 𝑓 enclosed by 𝛾 with the respective multiplicities 𝑘1,…,𝑘𝑛. Choose disjoint disks 𝐷(𝑧𝑗,𝜀𝑗) centered at each 𝑧𝑗 with radii 𝜀𝑗>0, each contained in the interior of 𝛾 and avoiding 𝛾. The function

𝑓′(𝑧)𝑓(𝑧)

is holomorphic on the domain

int(𝛾)∖⋃𝑗=1𝑛𝐷(𝑧𝑗,𝜀𝑗)‾,

where int(𝛾) denotes the interior relative to 𝛾. The oriented boundary of this domain is

𝛾+∪⋃𝑗=1𝑛𝜕𝐷(𝑧𝑗,𝜀𝑗)−.

By Cauchy–Goursat (Theorem 3.1.9Theorem 3.1.9Theorem 3.1.9),

∫𝛾+∪⋃𝑗=1𝑛𝜕𝐷(𝑧𝑗,𝜀𝑗)−𝑓′(𝑧)𝑓(𝑧)d𝑧=0,

which rearranges to

∮𝛾+𝑓′(𝑧)𝑓(𝑧)d𝑧=∑𝑗=1𝑛∮𝜕𝐷(𝑧𝑗,𝜀𝑗)+𝑓′(𝑧)𝑓(𝑧)d𝑧.

Near each 𝑧𝑗, express

𝑓(𝑧)=(𝑧−𝑧𝑗)𝑘𝑗ℎ𝑗(𝑧)

where ℎ𝑗 is holomorphic and non-vanishing on 𝐷(𝑧𝑗,𝜀𝑗). Differentiation yields

𝑓′(𝑧)=𝑘𝑗(𝑧−𝑧𝑗)𝑘𝑗−1ℎ𝑗(𝑧)+(𝑧−𝑧𝑗)𝑘𝑗ℎ𝑗′(𝑧),

and thus

𝑓′(𝑧)𝑓(𝑧)=𝑘𝑗𝑧−𝑧𝑗+ℎ𝑗′(𝑧)ℎ𝑗(𝑧).

Since ℎ𝑗 is holomorphic and non-vanishing on 𝐷(𝑧𝑗,𝜀𝑗), the function ℎ𝑗′/ℎ𝑗 is holomorphic there. By the Cauchy–Goursat Theorem,

∮𝜕𝐷(𝑧𝑗,𝜀𝑗)ℎ𝑗′(𝑧)ℎ𝑗(𝑧)d𝑧=0.

The Cauchy–Goursat Formula (Theorem 3.1.10Theorem 3.1.10Theorem 3.1.10) gives

∮𝜕𝐷(𝑧𝑗,𝜀𝑗)𝑘𝑗𝑧−𝑧𝑗d𝑧=2πi𝑘𝑗.

Combining results,

∮Γ𝑓′(𝑧)𝑓(𝑧)d𝑧=∑𝑗=1𝑛2πi𝑘𝑗=2πi𝑘.

Finally, parameterize Γ by 𝑤=𝑓(𝑧). Then d𝑤=𝑓′(𝑧)d𝑧, and

𝑘=12πi∮Γd𝑤𝑤=12πiΔΓlog(𝑤)=12πΔΓarg(𝑤),

which proves the result. ⁠□

Thus, one defines the winding index (Ind) to quantify how many times a closed curve winds counterclockwise around a given point in the complex plane. Formally, if 𝛾=𝛾([0,1]) is a counterclockwise-oriented closed curve and 𝑧 is a point satisfying 𝑧∉𝛾, then

IndΓ(𝑧)=12πi∮𝛾d𝜁𝜁−𝑧=12πi∫01𝛾′(𝑡)d𝑡𝛾(𝑡)−𝑧.

Theorem 3.3.5.

Let {𝑓𝑛(𝑧)} be a sequence of holomorphic functions on the open set 𝑈⊆ℂ that uniformly converges to 𝑓(𝑧) on every compact subset of 𝑈. If ∀𝑛∈ℕ, 𝑓𝑛(𝑧) has no zeros in 𝑈, then 𝑓 is either identically 0 or has no zeros in 𝑈.
Proof.

By the holomorphy of 𝑓𝑛(𝑧), for any simple closed rectifiable curve 𝛾⊂𝑈 (whose interior is a subset of 𝑈), by the Cauchy–Goursat Theorem (Theorem 3.1.9Theorem 3.1.9Theorem 3.1.9),

∮Γ𝑓𝑛(𝜁)d𝜁=0.

Since 𝛾 is a subset of any compact subset of 𝑈, {𝑓𝑛(𝜁)} uniformly converges on 𝛾, and by Theorem 2.3.7Theorem 2.3.7Theorem 2.3.7,

lim𝑛→∞∮Γ𝑓𝑛(𝜁)d𝜁=∮Γlim𝑛→∞𝑓𝑛(𝜁)d𝜁=∮Γ𝑓(𝜁)d𝜁=0.(3.3.2)

Then by Morera’s Theorem (Theorem 3.2.6Theorem 3.2.6Theorem 3.2.6), 𝑓(𝑧) is holomorphic, and 𝑓′(𝑧) is holomorphic. We aim to show that 𝑓𝑛′(𝑧)⇉𝑓′(𝑧).

Let 𝐾⋐𝑈 be arbitrary and compact and 𝑉⊃𝐾 be open and relatively compact in 𝑈. Since {𝑓𝑛′(𝑧)} is holomorphic, by Corollary 3.2.7.1Corollary 3.2.7.1Corollary 3.2.7.1, there exists a finite constant 𝑐>0 such that

lim𝑛→∞sup𝑧∈𝐾|𝑓𝑛′(𝑧)−𝑓′(𝑧)|≤𝑐lim𝑛→∞sup𝑧∈𝑉|𝑓𝑛(𝑧)−𝑓(𝑧)|.

By the definition of uniform convergence, the right-hand side approaches 0, and {𝑓𝑛′(𝑧)} is then uniformly convergent to 𝑓′(𝑧) by the same reasoning.

Through the proof of Theorem 3.3.2Theorem 3.3.2Theorem 3.3.2, if 𝑓≢0 over 𝑈, then the zeros of 𝑓 do not have an accumulation point in 𝑈 and are therefore discrete. In this case, let 𝛾⊂𝑈 be a curve that does not pass through the zeros of 𝑓. Since each function in the sequence 𝑓𝑛 does not contain zeros in 𝑈, by the Argument Principle (Theorem 3.3.4Theorem 3.3.4Theorem 3.3.4),

lim𝑛→∞∮Γ𝑓𝑛′(𝑧)𝑓𝑛(𝑧)d𝑧=0.(3.3.3)

Since 𝑓 and 𝑓′ are holomorphic over 𝛾, by Theorem 1.2.22Theorem 1.2.22Theorem 1.2.22, there exists a finite value 𝑀>0 such that ∀𝑧∈𝛾, max{|𝑓(𝑧)|,|𝑓′(𝑧)|}<𝑀.

Since 𝛾 does not pass through the zeros of 𝑓, ∃𝜆>0 such that ∀𝑧∈𝛾, |𝑓(𝑧)|>𝜆. By the uniform convergence of {𝑓𝑛(𝑧)}, ∃𝑁∈ℕ such that

|𝑓𝑛(𝑧)−𝑓(𝑧)|<𝜆2,∀𝑛>𝑁,∀𝑧∈𝛾.

Then |𝑓𝑛(𝑧)|>𝜆2 on 𝛾. Hence, 1𝑓𝑛(𝑧) and its limit are uniformly bounded;

|1𝑓(𝑧)|<1𝜆,|1𝑓𝑛(𝑧)|<2𝜆,∀𝑧∈𝛾,∀𝑛>𝑁.
|𝑓′𝑓−𝑓𝑛′𝑓𝑛|=|𝑓′𝑓𝑛−𝑓𝑛′𝑓𝑓𝑛𝑓|<2|𝑓′𝑓𝑛−𝑓′𝑓|+|𝑓′𝑓−𝑓𝑛′𝑓|𝜆2<2𝑀𝜆2⋅(|𝑓𝑛−𝑓|+|𝑓′−𝑓𝑛′|).
sup𝑧∈𝛾|𝑓′(𝑧)𝑓(𝑧)−𝑓𝑛′(𝑧)𝑓𝑛(𝑧)|≤2𝑀𝜆2(sup𝑧∈𝛾|𝑓𝑛(𝑧)−𝑓(𝑧)|+sup𝑧∈𝛾|𝑓′(𝑧)−𝑓𝑛′(𝑧)|)lim𝑛→∞sup𝑧∈𝛾|𝑓′(𝑧)𝑓(𝑧)−𝑓𝑛′(𝑧)𝑓𝑛(𝑧)|≤2𝑀𝜆2(lim𝑛→∞sup𝑧∈𝛾|𝑓𝑛(𝑧)−𝑓(𝑧)|+|𝑓′(𝑧)−𝑓𝑛′(𝑧)|)=0.

Therefore, 𝑓′(𝑧)𝑓(𝑧) is uniformly convergent on 𝛾. By Theorem 2.3.7Theorem 2.3.7Theorem 2.3.7, we can pass the limit through the integral in (3.3.3)(3.3.3)(3.3.3). Then,

lim𝑛→∞∮𝛾𝑓𝑛′(𝑧)𝑓𝑛(𝑧)d𝑧=∮𝛾𝑓′(𝑧)𝑓(𝑧)d𝑧=0.

By the Argument Principle (Theorem 3.3.4Theorem 3.3.4Theorem 3.3.4), 𝑓(𝑧) has no zeros in the interior of 𝛾. Since 𝛾 was arbitrarily chosen, either 𝑓(𝑧)≡0 on 𝑈 or has no zeros in 𝑈. ⁠□

Theorem 3.3.6 (Rouché).

Let 𝑈⊆ℂ be open and 𝑓,𝑔 be two holomorphic functions over 𝑈. Let 𝛾⊂𝑈 be a simple, closed, rectifiable curve, and for all 𝑧∈𝛾

|𝑓(𝑧)−𝑔(𝑧)|<|𝑓(𝑧)|.(3.3.4)

Then 𝑓 and 𝑔 have the same number of zeros enclosed by 𝛾 and do not vanish on 𝛾.

Proof.

It is obvious that 𝑔(𝑧) has no zeros on 𝛾. Otherwise, ∃𝑧0∈𝛾 such that 𝑔(𝑧0)=0, implying that |𝑓(𝑧0)|<|𝑓(𝑧0)| which is impossible. Similarly, 𝑓(𝑧) has no zeros on 𝛾, since |𝑔(𝑧)|<0 is an impossibility.

Let 𝑘𝑓 and 𝑘𝑔 denote the number of zeros of 𝑓 and 𝑔 enclosed by 𝛾, respectively. By the Argument Principle (Theorem 3.3.4Theorem 3.3.4Theorem 3.3.4),

𝑘𝑔−𝑘𝑓=∮Γ𝑔′(𝑧)𝑔(𝑧)d𝑧−∮𝛾𝑓′(𝑧)𝑓(𝑧)d𝑧=∮𝛾𝑔′(𝑧)𝑓(𝑧)−𝑓′(𝑧)𝑔(𝑧)𝑔(𝑧)𝑓(𝑧)d𝑧=∮𝛾(𝑔(𝑧)𝑓(𝑧))′𝑔(𝑧)𝑓(𝑧)d𝑧.

Let 𝑤=ℎ(𝑧)=𝑔(𝑧)𝑓(𝑧) with Γ=ℎ(𝛾). Then,

𝑘𝑔−𝑘𝑓=∮Γd𝑤𝑤.

From (3.3.4)(3.3.4)(3.3.4), by dividing both sides by 𝑓(𝑧), we obtain |𝑤−1|<1. Then Γ lies in the open disk 𝐷(1,1), which will never intersect or enclose 0. Then by Lemma 3.1.8Lemma 3.1.8Lemma 3.1.8,

𝑘𝑔−𝑘𝑓=∮Γd𝑤𝑤=0,

as desired. ⁠□

By the Fundamental Theorem of Algebra (Theorem 3.3.1Theorem 3.3.1Theorem 3.3.1), any polynomial in the form 𝑝(𝑧)=∑𝑘=0𝑛𝑎𝑘𝑧𝑘 (𝑛∈ℕ, 𝑎𝑛≠0, 𝑎𝑘∈ℂ where 𝑘=1,…𝑛) has at least one complex zero. Consider the function 𝑞(𝑧)=𝑎𝑛𝑧𝑛, with a zero at 𝑧=0 with multiplicity 𝑛. By Rouché’s Theorem (Theorem 3.3.6Theorem 3.3.6Theorem 3.3.6), since ∃𝑅∈ℝ such that |𝑞(𝑧)−𝑝(𝑧)|=|∑𝑘=0𝑛−1𝑎𝑘𝑧𝑘|<|𝑎𝑛𝑧𝑛| over |𝑧|=𝑅, 𝑝 and 𝑞 have the same number of zeros, counting multiplicity.

Theorem 3.3.7.

Let 𝑈⊆ℂ be open and connected, and 𝑓(𝑧) be holomorphic and non-constant on 𝑈.

If 𝑧0∈𝑈 and 𝑤0=𝑓(𝑧0), and the multiplicity of the zero at 𝑧0 of 𝑓−𝑤0 is 𝑚, then for all 𝜌>0 such that 𝑓−𝑤0 is non-vanishing on 𝐷(𝑧0,𝜌)‾∖{𝑧0}, ∃𝛿>0 such that ∀𝜉∈𝐷(𝑤0,𝛿), 𝑓−𝜉 has 𝑚 zeros in 𝐷(𝑧0,𝜌), counting multiplicity.

Proof.

The zero at 𝑧0 is isolated by Theorem 3.3.2Theorem 3.3.2Theorem 3.3.2. Furthermore, |𝑓−𝑤0| is continuous on 𝜕𝐷(𝑧0,𝜌) and attains a positive infimum 𝛿. In other words, on this set, |𝑓−𝑤0|≥𝛿. Hence, ∀𝜉∈𝐷(𝑤0,𝛿), we have |𝜉−𝑤0|<𝛿≤|𝑓(𝑧)−𝑤0| for any 𝑧∈𝜕𝐷(𝑧0,𝜌).

By Rouché’s Theorem, since |(𝑓(𝑧)−𝑤0)−(𝑓(𝑧)−𝜉)|<|𝑓(𝑧)−𝑤0|, it follows that 𝑓−𝜉 and 𝑓−𝑤0 have the same number of zeros in 𝐷(𝑧0,𝜌). ⁠□

We also have the following generalization of Theorem 3.3.5Theorem 3.3.5Theorem 3.3.5, which is a heuristic restatement of Theorem 3.3.7Theorem 3.3.7Theorem 3.3.7:

Theorem 3.3.8 (Hurwitz).

Let 𝑈⊆ℂ be an open and connected set, and suppose {𝑓𝑛(𝑧)}𝑛∈ℕ is a holomorphic function sequence that uniformly converges to a non-constant function 𝑓(𝑧) on all compact sets of 𝑈.

If 𝑧0∈𝑈 and 𝑤0=𝑓(𝑧0), and the multiplicity of the zero at 𝑧0 of 𝑓−𝑤0 is 𝑚, then for all 𝜌>0 such that 𝑓−𝑤0 is non-vanishing on 𝐷(𝑧0,𝜌)‾∖{𝑧0}, ∃𝑁∈ℕ such that ∀𝑛>𝑁, 𝑓𝑛−𝑤0 has 𝑚 zeros in 𝐷(𝑧0,𝜌), counting multiplicity.

Proof.

The zero at 𝑧0 is isolated by Theorem 3.3.2Theorem 3.3.2Theorem 3.3.2. Furthermore, |𝑓−𝑤0| is continuous on 𝜕𝐷(𝑧0,𝜌) and attains a positive infimum 𝛿. In other words, on this set, |𝑓−𝑤0|≥𝛿. By uniform convergence, ∃𝑁∈ℕ such that ∀𝑛>𝑁, we have |𝑓(𝑧)−𝑓𝑛(𝑧)|<𝛿≤|𝑓(𝑧)−𝑤0| for any 𝑧∈𝜕𝐷(𝑧0,𝜌).

By Rouché’s Theorem (Theorem 3.3.6Theorem 3.3.6Theorem 3.3.6), since

|(𝑓(𝑧)−𝑤0)−(𝑓𝑛(𝑧)−𝑤0)|<|𝑓(𝑧)−𝑤0|,

it follows that 𝑓𝑛−𝑤0 and 𝑓−𝑤0 have the same number of zeros in 𝐷(𝑧0,𝜌). ⁠□