Notes on Complex Analysis
Export PDFGitHub

§2.3 Complex Power Series

2.3 Complex Power Series

Power series in real analysis can be generalized into complex series. Particularly, concepts such as uniform convergence are analogous to those in real analysis:

Definition 2.3.1 (Uniform Convergence).

For a set 𝑈⊆ℂ, a function sequence {𝑓𝑛(𝑧)} uniformly converges to a function 𝑓(𝑧) on 𝑈 iff ∀𝜀>0, ∃𝑁∈ℕ such that ∀𝑛>𝑁, ∀𝑧∈𝑈, |𝑓𝑛(𝑧)−𝑓(𝑧)|<𝜀.

Remark.

The definition above is equivalent to the following definition.

For a set 𝑈⊆ℂ, a function sequence {𝑓𝑛(𝑧)} uniformly converges to 𝑓(𝑧) iff

lim𝑛→∞sup𝑧∈𝑈|𝑓𝑛(𝑧)−𝑓(𝑧)|=0.

Informally, we will use the notation 𝑓𝑛(𝑧)⇉𝑓(𝑧) to represent uniform convergence.

Theorem 2.3.2 (Cauchy Criterion).

For a set 𝑈⊆ℂ, a function sequence {𝑓𝑛(𝑧)} uniformly converges to a function 𝑓(𝑧) iff ∀𝜀>0, ∃𝑁∈ℕ such that ∀𝑛,𝑚>𝑁, ∀𝑧∈𝑈, |𝑓𝑛(𝑧)−𝑓𝑚(𝑧)|<𝜀.
Proof.

If 𝑓𝑛(𝑧) uniformly converges to 𝑓(𝑧), then for ∀𝜀>0, ∃𝑁∈ℕ such that ∀𝑛,𝑚>𝑁 and ∀𝑧∈𝑈,

|𝑓𝑚(𝑧)−𝑓(𝑧)|<𝜀2,|𝑓𝑛(𝑧)−𝑓(𝑧)|<𝜀2.

Then,

|𝑓𝑚(𝑧)−𝑓𝑛(𝑧)|≤|𝑓𝑛(𝑧)−𝑓(𝑧)|+|𝑓𝑚(𝑧)−𝑓(𝑧)|<𝜀.

For the converse, refer to the analogous proof in Theorem 1.2.17Theorem 1.2.17Theorem 1.2.17. ⁠□

Function series are defined to be a sequence formed by the partial sums of function sequences. There are many ways to verify the uniform convergence of a function series. Perhaps the most widely known is the Weierstrass 𝑀-Test.

Theorem 2.3.3 (Weierstrass 𝑀-Test).

Let 𝑈⊆ℂ be a region and {𝑓𝑛} be a function sequence on 𝑈.

If ∃{𝑀𝑛}⊂ℝ≥0 such that ∀𝑛∈ℕ, ∀𝑧∈𝑈, |𝑓𝑛(𝑧)|≤𝑀𝑛 and the series ∑𝑛=1∞𝑀𝑛 converges, then the series ∑𝑛=1∞𝑓𝑛(𝑧) converges uniformly and absolutely on 𝑈.

Proof.

By the convergence of ∑𝑛=1∞𝑀𝑛, ∀𝜀>0, ∃𝑁∈ℕ such that ∀𝑚≥𝑛>𝑁,

|𝑀𝑚+𝑀𝑚−1+⋯+𝑀𝑛+1|<𝜀.

Since 𝑀𝑗 bounds 𝑓𝑗(𝑧), it follows that

|𝑓𝑚(𝑧)+𝑓𝑚−1(𝑧)+⋯+𝑓𝑛+1(𝑧)|≤|𝑀𝑚+𝑀𝑚−1+⋯+𝑀𝑛+1|<𝜀,

and the result follows from Theorem 2.3.2Theorem 2.3.2Theorem 2.3.2. ⁠□

The concept of uniform convergence is generalized to improper integrals with parameters, and the same theorems from series have a corresponding counterpart.

In both complex and real analysis, the concept of power series, a unique type of function series, is of trivial importance. Similar to real power series, complex series have the form

∑𝑛=0∞𝑎𝑛𝑧𝑛,

where {𝑎𝑛} are constants.

Let 𝐷(𝑎,𝑟)=𝐵1(𝑎,𝑟)={𝑧∈ℂ:|𝑧−𝑎|<𝑟} denote the open disk centered at 𝑎 with radius 𝑟. For simplicity, from now on we will have 𝔻 denote the unit open disk, or 𝐷(0,1). We will now observe the convergence of power series.

Theorem 2.3.4 (Abel’s Theorem).

For a power series 𝑓(𝑧)=∑𝑛=0∞𝑎𝑛𝑧𝑛, there exists a constant 𝑅∈ℝ≥0∪{∞}, known as the radius of convergence such that:

  1. 𝑓 absolutely converges on 𝐷(0,𝑅), and ∀0≤𝜌<𝑅, uniformly converges on 𝐷(0,𝜌)‾.
  2. 𝑓(𝑧) diverges when |𝑧|>𝑅.
  3. 𝑓 is holomorphic over 𝐷(0,𝑅) and 𝑓′(𝑧) can be obtained by termwise differentiation, or 𝑓′(𝑧)=∑𝑛=1∞𝑛𝑎𝑛𝑧𝑛−1, which also has a convergence radius of 𝑅.

The disk |𝑧|<𝑅 is known as the disk of convergence, a direct generalization of the interval of convergence for real series. There are many ways to determine the radius of convergence:

Theorem 2.3.5 (Cauchy–Hadamard).

The radius of convergence of the power series in the form ∑𝑛=0∞𝑎𝑛𝑧𝑛 can be determined by

𝑅=1lim sup𝑛→∞|𝑎𝑛|𝑛.(2.3.1)

Of course, a convergence radius of 0 implies that the series is divergent everywhere except for possibly at 0, and a convergence radius of ∞ means that the series absolutely converges everywhere.

Proof of Theorem 2.3.4Theorem 2.3.4Theorem 2.3.4 and Theorem 2.3.5Theorem 2.3.5Theorem 2.3.5.

We will prove that the value in (2.3.1)(2.3.1)(2.3.1) satisfies the criteria in Theorem 2.3.4Theorem 2.3.4Theorem 2.3.4.

Assume |𝑧|<𝑅. Then ∀𝜌∈(|𝑧|,𝑅), and consequently, 1𝜌>1𝑅. By Definition 1.2.18Definition 1.2.18Definition 1.2.18 and (2.3.1)(2.3.1)(2.3.1), ∃𝑁∈ℕ such that ∀𝑛>𝑁, |𝑎𝑛|𝑛<1𝜌 and |𝑎𝑛|<1𝜌𝑛. It follows that |𝑎𝑛𝑧𝑛|<|𝑧|𝑛𝜌𝑛<1 for all 𝑛>𝑁. Then ∑𝑛=0∞|𝑎𝑛𝑧𝑛| converges.

Let 𝜌′∈(𝜌,𝑅). Similarly, ∃𝑁′∈ℕ such that ∀𝑛>𝑁′, |𝑎𝑛|𝑛<1𝜌′, and |𝑎𝑛|<1𝜌′𝑛. Then |𝑎𝑛𝑧𝑛|<|𝑎𝑛𝜌𝑛|<𝜌𝑛𝜌′𝑛. By the Weierstrass 𝑀-Test (Theorem 2.3.3Theorem 2.3.3Theorem 2.3.3), ∑𝑛=0∞|𝑎𝑛𝑧𝑛| is uniformly bounded for 𝑛>𝑁′ by the convergent series ∑𝑛=0∞𝑎𝑛𝜌𝑛, and thus uniformly converges on |𝑧|<𝜌. This proves Part 1Part 1Part 1.

Assume that |𝑧|>𝑅. For all 𝜌∈(𝑅,|𝑧|), 1𝜌<1𝑅. By Definition 1.2.18Definition 1.2.18Definition 1.2.18, ∀𝑁∈ℕ, ∃𝑛𝑁>𝑁 such that |𝑎𝑛𝑁|𝑛𝑁>1𝜌. It follows that |𝑎𝑛𝑁𝑧𝑛𝑁|>|𝑧𝑛𝑁|𝜌𝑛𝑁>1. Since ∀𝑁∈ℕ,

|∑𝑘=0𝑛𝑁𝑎𝑘𝑧𝑘−∑𝑘=0𝑛𝑁−1𝑎𝑘𝑧𝑘|>1,

by the Cauchy Criterion (Theorem 1.2.17Theorem 1.2.17Theorem 1.2.17), ∑𝑛=0∞𝑎𝑛𝑧𝑛 is divergent. Thus, Part 2Part 2Part 2 is satisfied.

To prove Part 3Part 3Part 3, first observe that ∑𝑛=1∞𝑛𝑎𝑛𝑧𝑛 and ∑𝑛=1∞𝑎𝑛𝑧𝑛 have the same convergence radius since lim sup𝑛→∞𝑛𝑛=1. For 𝑧∈𝐷(0,𝑅), let 𝑓(𝑧)=𝑆𝑛(𝑧)+𝑅𝑛(𝑧), where

𝑆𝑛(𝑧)=∑𝑘=0𝑛−1𝑎𝑘𝑧𝑘,𝑅𝑛(𝑧)=∑𝑘=𝑛∞𝑎𝑘𝑧𝑘.

Let

𝑓1(𝑧)=lim𝑛→∞𝑆𝑛′(𝑧)=∑𝑛=1∞𝑛𝑎𝑛𝑧𝑛−1.

Let 𝜌<𝑅 be positive and |𝑧0|<𝜌. Then we aim to show that

lim𝑧→𝑧0(𝑓(𝑧)−𝑓(𝑧0)𝑧−𝑧0−𝑓1(𝑧))=0.

By analyzing the difference,

𝑓(𝑧)−𝑓(𝑧0)𝑧−𝑧0−𝑓1(𝑧)=[𝑆𝑛(𝑧)−𝑆𝑛(𝑧0)𝑧−𝑧0−𝑆𝑛′(𝑧)]+𝑆𝑛′(𝑧)−𝑓1(𝑧)+𝑅𝑛(𝑧)−𝑅𝑛(𝑧0)𝑧−𝑧0.(2.3.2)

Since 𝑆𝑛′(𝑧)→𝑓1(𝑧) as 𝑛→∞, it follows that ∀𝜀>0, ∃𝑁∈ℕ such that ∀𝑛>𝑁, |𝑆𝑛′(𝑧)−𝑓1(𝑧)|<𝜀3. Since

𝑅𝑛(𝑧)−𝑅𝑛(𝑧0)𝑧−𝑧0=∑𝑘=𝑛∞𝑎𝑘(𝑧𝑘−1+𝑧𝑘−2𝑧0+⋯+𝑧0𝑘−1)

for 𝑧≠𝑧0, with |𝑧|<𝜌<𝑅,

|∑𝑘=𝑛∞𝑎𝑘(𝑧𝑘−1+⋯+𝑧0𝑘−1)|≤∑𝑘=𝑛∞|𝑎𝑘|(|𝑧𝑘−1|+⋯+|𝑧0𝑘−1|)<∑𝑘=𝑛∞|𝑎𝑘|𝑘𝜌𝑘−1.

Since ∑𝑘=1∞𝑘|𝑎𝑘|𝜌𝑘−1 is absolutely convergent, ∑𝑘=𝑛∞|𝑎𝑘|𝑘𝜌𝑘−1 is the remainder term of a convergent series. Then, ∃𝑁′∈ℕ such that ∀𝑛>𝑁′, ∑𝑘=𝑛∞|𝑎𝑘|𝑘𝜌𝑘−1<𝜀3.

Finally, for a fixed 𝑛>max(𝑁,𝑁′), ∃𝛿>0 such that ∀𝑧∈𝐷(𝑧0,𝛿)∖{𝑧0},

|𝑆𝑛(𝑧)−𝑆𝑛(𝑧0)𝑧−𝑧0−𝑆𝑛′(𝑧)|<𝜀3.

From (2.3.2)(2.3.2)(2.3.2), we get

|𝑓(𝑧)−𝑓(𝑧0)𝑧−𝑧0−𝑓1(𝑧)|<𝜀,

confirming Part 3Part 3Part 3. ⁠□

Obviously, a substitution of 𝑧=𝜁−𝑎 where 𝑎∈ℂ translates the disk of convergence to 𝐷(𝑎,𝑅). The subsequent results on uniform convergence hold for complex functions:

Theorem 2.3.6 (Uniform Limit).

Let {𝑓𝑛(𝑧)}𝑛 be a sequence of continuous functions on 𝑈⊆ℂ and uniformly convergent to 𝑓(𝑧). Then 𝑓(𝑧) is continuous on 𝑈.
Proof.

Fix 𝜀>0 and 𝑧0∈𝑈. By uniform convergence, ∃𝑛∈ℕ such that ∀𝑧∈𝑈, |𝑓𝑛(𝑧)−𝑓(𝑧)|<𝜀3 and |𝑓𝑛(𝑧0)−𝑓(𝑧0)|<𝜀3. By continuity, ∃𝛿>0 such that ∀𝑧∈𝐷(𝑧0,𝛿)∩𝑈⊆𝑈, |𝑓𝑛(𝑧)−𝑓𝑛(𝑧0)|<𝜀3.

By the triangle inequality,

|𝑓(𝑧)−𝑓(𝑧0)|≤|𝑓(𝑧)−𝑓𝑛(𝑧)|+|𝑓𝑛(𝑧)−𝑓𝑛(𝑧0)|+|𝑓𝑛(𝑧0)−𝑓(𝑧0)|<𝜀.

for all 𝑧∈𝐷(𝑧0,𝛿)∩𝑈. Then the continuity of 𝑓 is satisfied. ⁠□

Lastly, the sufficient criteria to pass a limit through an integral:

Theorem 2.3.7.

Let 𝛾 be a rectifiable curve on which the function sequence {𝑓𝑛}𝑛∈ℕ is continuous. If {𝑓𝑛(𝑧)} uniformly converges to 𝑓, then

lim𝑛→∞∫𝛾𝑓𝑛(𝑧)d𝑧=∫𝛾𝑓(𝑧)d𝑧.
Proof.

Since {𝑓𝑛(𝑧)} uniformly converges to 𝑓(𝑧) on 𝛾, ∀𝜀>0, there exists 𝑁∈ℕ such that for all 𝑛>𝑁,

|𝑓𝑛(𝑧)−𝑓(𝑧)|<𝜀length(𝛾),∀𝑧∈𝛾.

Since each 𝑓𝑛 is continuous and 𝛾 is rectifiable, each integral ∫𝛾𝑓𝑛(𝑧)d𝑧 is convergent and well-defined.

Then ∀𝑛>𝑁,

|∫𝛾𝑓𝑛(𝑧)d𝑧−∫𝛾𝑓(𝑧)d𝑧|=|∫𝛾[𝑓𝑛(𝑧)−𝑓(𝑧)]d𝑧|≤∫𝛾|𝑓𝑛(𝑧)−𝑓(𝑧)||d𝑧|<∫𝛾𝜀length(𝛾)|d𝑧|=𝜀.

Therefore,

lim𝑛→∞∫𝛾𝑓𝑛(𝑧)d𝑧=∫𝛾𝑓(𝑧)d𝑧.□

Remark.

For a uniformly convergent series ∑𝑛=1∞𝑓𝑛(𝑧), the commutation between the limit and the integral becomes a summation-integral switch:

∑𝑛=1∞∫𝛾𝑓𝑛(𝑧)d𝑧=∫𝛾∑𝑛=1∞𝑓𝑛(𝑧)d𝑧.